diff --git a/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs1_seq131072_zero0_tpmodeRED_l15_h2048_heads16.yaml b/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs1_seq131072_zero0_tpmodeRED_l15_h2048_heads16.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1c348c5b856a48970cb0480ed21ea3ec36af9796 --- /dev/null +++ b/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs1_seq131072_zero0_tpmodeRED_l15_h2048_heads16.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.07G_dp1_tp8_pp1_acc1_mbs1_seq131072_zero0_tpmodeRED_l15_h2048_heads16 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 131072 + num_attention_heads: 16 + num_hidden_layers: 15 + num_key_value_heads: 16 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 131072 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_l15_h2048_heads16.yaml b/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_l15_h2048_heads16.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8fa6bf5bfb4e45ed3e9db730234b320de72edbd --- /dev/null +++ b/configs/config_1.07G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_l15_h2048_heads16.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_tp.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.07G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_l15_h2048_heads16 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 16 + num_hidden_layers: 15 + num_key_value_heads: 16 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp16_pp1_acc2_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp16_pp1_acc2_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3957c2a8bc4146de7e5d3002567902ec5b268583 --- /dev/null +++ b/configs/config_1.14G_dp16_tp16_pp1_acc2_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp16_pp1_acc2_mbs64_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp32_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp32_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cf5b21840ba83e213f187c6704d11461eea3d15b --- /dev/null +++ b/configs/config_1.14G_dp16_tp32_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp32_pp1_acc2_mbs4_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp8_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp8_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5c0f50290735ccf5cb25e3029979b6f334b7a455 --- /dev/null +++ b/configs/config_1.14G_dp16_tp8_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp8_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp1_tp1_pp8_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp1_tp1_pp8_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b364b8d961175b46ce104230644465ff9646c55b --- /dev/null +++ b/configs/config_1.14G_dp1_tp1_pp8_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp1_tp1_pp8_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc128_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc128_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cfd559e212807c27ed372c72b004fbe3b3cdd029 --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc128_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc128_mbs8_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc8_mbs8_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc8_mbs8_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..38d89ba7a676e48bee8459b45e4cb38e0d5907b0 --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc8_mbs8_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc8_mbs8_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc2_mbs128_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc2_mbs128_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5f0e0dc935c81604a67baff11e1b421b61089484 --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc2_mbs128_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc2_mbs128_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp16_pp1_acc4_mbs4_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp32_tp16_pp1_acc4_mbs4_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3dfac4cb9908517f8ecf9039debed13df252a0e5 --- /dev/null +++ b/configs/config_1.14G_dp32_tp16_pp1_acc4_mbs4_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp16_pp1_acc4_mbs4_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp16_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp32_tp16_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eb1e3ed3a5b4c4de02b699cba98544288045756d --- /dev/null +++ b/configs/config_1.14G_dp32_tp16_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp16_pp1_acc8_mbs2_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp2_pp1_acc16_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp32_tp2_pp1_acc16_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7fd0b6501d6194af1abbcbe9041c830456d03ca5 --- /dev/null +++ b/configs/config_1.14G_dp32_tp2_pp1_acc16_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp2_pp1_acc16_mbs1_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp4_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp32_tp4_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ebf59893f1d3eb503c9808db87ed8c67d04fa0e8 --- /dev/null +++ b/configs/config_1.14G_dp32_tp4_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp4_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp3_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k_prof.yaml b/configs/config_1.14G_dp3_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k_prof.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b1bbd417006f53e871004e9cd8d14ea0e2ff2bc1 --- /dev/null +++ b/configs/config_1.14G_dp3_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k_prof.yaml @@ -0,0 +1,92 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp3_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeRED_vocab32k_prof + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 3 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: + profiler_export_path: ./tb_logs +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 10 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp128_pp1_acc16_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp128_pp1_acc16_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..18a952958f70b0afc9c11d63823aeccd4276e37c --- /dev/null +++ b/configs/config_1.14G_dp4_tp128_pp1_acc16_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp128_pp1_acc16_mbs2_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 128 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp128_pp1_acc8_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp128_pp1_acc8_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..748f9215e84a4051762380200ebfaf08b8a1049b --- /dev/null +++ b/configs/config_1.14G_dp4_tp128_pp1_acc8_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp128_pp1_acc8_mbs16_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 128 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp16_pp1_acc1_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp16_pp1_acc1_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..400082d423c1d0c5f4ab35e2c376860af419f134 --- /dev/null +++ b/configs/config_1.14G_dp4_tp16_pp1_acc1_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp16_pp1_acc1_mbs32_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp2_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp2_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7ba5eae941147f11c2e5480133251bf5df61e6fc --- /dev/null +++ b/configs/config_1.14G_dp8_tp2_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp2_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f9df07ace3009e8c91af3bf9c9bf1086287d07bd --- /dev/null +++ b/configs/config_1.14G_dp8_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp4_pp1_acc2_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp4_pp1_acc2_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..639aed53644e2b4aae89183f534562d0b9989340 --- /dev/null +++ b/configs/config_1.14G_dp8_tp4_pp1_acc2_mbs2_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp4_pp1_acc2_mbs2_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp64_pp1_acc1_mbs64_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp64_pp1_acc1_mbs64_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d6940705bf6f95c2bc6bdba92b81b81b829522fd --- /dev/null +++ b/configs/config_1.14G_dp8_tp64_pp1_acc1_mbs64_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp64_pp1_acc1_mbs64_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..79d7db8c53ca2e4c41f29cd235a4f2c833ab9af3 --- /dev/null +++ b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp4_pp1_acc1_mbs4_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp16_pp1_acc8_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp16_tp16_pp1_acc8_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..79cf3f9ea254eb35cdfa633bbeb6031948aa48fb --- /dev/null +++ b/configs/config_1.34G_dp16_tp16_pp1_acc8_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp16_pp1_acc8_mbs4_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp1_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp1_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1c1d89f814b05b86f9c082d666fc2b66a579de9f --- /dev/null +++ b/configs/config_1.34G_dp16_tp1_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp1_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp2_pp1_acc4_mbs8_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp2_pp1_acc4_mbs8_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..620ca125d752d4951dfd13ea50ed4d68623a3522 --- /dev/null +++ b/configs/config_1.34G_dp16_tp2_pp1_acc4_mbs8_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp2_pp1_acc4_mbs8_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp16_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp16_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..22f848772c8bb0fb4b3a5d079bdefdaff121f095 --- /dev/null +++ b/configs/config_1.34G_dp1_tp16_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp16_pp8_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp32_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp32_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..04de0fb92335dbfe3335cdf15a9f20eaa927328e --- /dev/null +++ b/configs/config_1.34G_dp1_tp32_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp32_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp256_tp1_pp1_acc2_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp256_tp1_pp1_acc2_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7e8d1a40aa1ff6e6a849aa9dc2d030ef16030f8 --- /dev/null +++ b/configs/config_1.34G_dp256_tp1_pp1_acc2_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp256_tp1_pp1_acc2_mbs4_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 256 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp16_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp16_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6e08188a90e0c7e83fe1a4f282c0618a3edfa8e7 --- /dev/null +++ b/configs/config_1.34G_dp2_tp16_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp16_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp16_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp16_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f1fbd194a552044933559a0732147d5414cf4530 --- /dev/null +++ b/configs/config_1.34G_dp2_tp16_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp16_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..776011d289c9cbc6fb502bc34690e0b263443933 --- /dev/null +++ b/configs/config_1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp256_pp1_acc32_mbs2_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp32_pp2_acc2_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp32_pp2_acc2_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..67563f5a900107c6e584ee70ca3e2d5f736648b8 --- /dev/null +++ b/configs/config_1.34G_dp2_tp32_pp2_acc2_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp32_pp2_acc2_mbs64_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp8_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp8_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..468c9b97711f4b33ecbf920157bc115530d806b2 --- /dev/null +++ b/configs/config_1.34G_dp2_tp8_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp8_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp8_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp8_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cbb111c6467647122faaa4ece3cda1600648573b --- /dev/null +++ b/configs/config_1.34G_dp2_tp8_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp8_pp4_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..759176290a84cbebc84659c6edf2b870b20e3e35 --- /dev/null +++ b/configs/config_1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp4_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp4_pp2_acc2_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp4_pp2_acc2_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..17a4e8f6f50998626cf017a920564467db618036 --- /dev/null +++ b/configs/config_1.34G_dp32_tp4_pp2_acc2_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp4_pp2_acc2_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp8_pp1_acc1_mbs4_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp32_tp8_pp1_acc1_mbs4_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9dc298ff88d16706f6a3d1b99df911c05960d12e --- /dev/null +++ b/configs/config_1.34G_dp32_tp8_pp1_acc1_mbs4_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp8_pp1_acc1_mbs4_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp16_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp16_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..344e12de297c9928821318c3e53daa4de309ae6a --- /dev/null +++ b/configs/config_1.34G_dp4_tp16_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp16_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a4e3508bc4cb5d30eacf1f2b1145afa362ecb74e --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp32_pp1_acc32_mbs4_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7552e0ff7bd34e3d873d25fd0b941cb807a4cda --- /dev/null +++ b/configs/config_1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp4_pp1_acc32_mbs16_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7e39ed75efe70e25aa2de7c6f1b1142e966bbbac --- /dev/null +++ b/configs/config_1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp64_pp1_acc16_mbs2_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4ad1ff7f5c2e12b4b43ae65dd09a3236ea807015 --- /dev/null +++ b/configs/config_1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp64_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp64_tp1_pp2_acc8_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp64_tp1_pp2_acc8_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..03db0f263819439ed7cadddff75c87ebe2eb62fe --- /dev/null +++ b/configs/config_1.34G_dp64_tp1_pp2_acc8_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp64_tp1_pp2_acc8_mbs1_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp64_tp2_pp1_acc8_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp64_tp2_pp1_acc8_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f824191534bf28bae163eedbfc29efae156fdce7 --- /dev/null +++ b/configs/config_1.34G_dp64_tp2_pp1_acc8_mbs4_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp64_tp2_pp1_acc8_mbs4_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp1_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp1_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e88cdea2fb5f823b4cd3793f81dd2c4bbedd5078 --- /dev/null +++ b/configs/config_1.34G_dp8_tp1_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp1_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp1_pp2_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp1_pp2_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..faf4e4573f68cf0e6099613b968337693f2d36c4 --- /dev/null +++ b/configs/config_1.34G_dp8_tp1_pp2_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp1_pp2_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp32_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp32_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e70d7169640dd655550e6af383dc13813adb86ef --- /dev/null +++ b/configs/config_1.34G_dp8_tp32_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp32_pp1_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e8a703da03eba6f68d16467ef2e501f90003cb53 --- /dev/null +++ b/configs/config_1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp4_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp8_pp1_acc8_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp8_pp1_acc8_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ec1762fa41e6976c313e565bf56c03ac010aa9dc --- /dev/null +++ b/configs/config_1.34G_dp8_tp8_pp1_acc8_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp8_pp1_acc8_mbs8_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_187G_dp2_tp8_pp32_acc1_mbs1_seq2048_zero0_tpmodeRED_l126_h16384_heads128.yaml b/configs/config_187G_dp2_tp8_pp32_acc1_mbs1_seq2048_zero0_tpmodeRED_l126_h16384_heads128.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6d00e9196f17061b6be6da22363752f45ec7851d --- /dev/null +++ b/configs/config_187G_dp2_tp8_pp32_acc1_mbs1_seq2048_zero0_tpmodeRED_l126_h16384_heads128.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 187G_dp2_tp8_pp32_acc1_mbs1_seq2048_zero0_tpmodeRED_l126_h16384_heads128 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 32 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp16_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp1_tp16_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dacd2f32656b8a44335956bc1b7639b7841d968a --- /dev/null +++ b/configs/config_3.57G_dp1_tp16_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp16_pp2_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp2_pp32_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp1_tp2_pp32_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b2f69b59e10fa34d23bcef54f8e24d4a5ede46a1 --- /dev/null +++ b/configs/config_3.57G_dp1_tp2_pp32_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp2_pp32_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 32 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp4_pp4_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp4_pp4_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eacf5364e1d342af53c2a515f93eedc48cc87735 --- /dev/null +++ b/configs/config_3.57G_dp2_tp4_pp4_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp4_pp4_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp8_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp8_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..622874713081653df0396ace28a8d75effee20af --- /dev/null +++ b/configs/config_3.57G_dp2_tp8_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp8_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.8G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_3.8G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b2f5f79811eae3835801337a15f7c9875c6946c0 --- /dev/null +++ b/configs/config_3.8G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.8G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeRED_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 30 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 7 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 3 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp1_tp2_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp1_tp2_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..838b7c9437f6ebd3ea391ff0aab9a8bb0cbeea95 --- /dev/null +++ b/configs/config_469G_dp1_tp2_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp1_tp2_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8dbb207feae3c682ecd96ae4889ad4c59dc860e4 --- /dev/null +++ b/configs/config_469G_dp2_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 32 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp32_pp4_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp32_pp4_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..558a48c1c09d7790b264c51e71902d45b0ab0955 --- /dev/null +++ b/configs/config_469G_dp2_tp32_pp4_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp32_pp4_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp32_tp2_pp2_acc8_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp32_tp2_pp2_acc8_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..354cd5cea7dddf0a7739b51adb61ba8bca2f90a4 --- /dev/null +++ b/configs/config_469G_dp32_tp2_pp2_acc8_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp32_tp2_pp2_acc8_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp8_tp1_pp4_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp8_tp1_pp4_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..169b719d3c645bd0af926d35ffadecacac46f85f --- /dev/null +++ b/configs/config_469G_dp8_tp1_pp4_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp8_tp1_pp4_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp8_tp4_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp8_tp4_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4e62ea862b7b1e10e46e097e06d8b2e8bb3a1eb3 --- /dev/null +++ b/configs/config_469G_dp8_tp4_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp8_tp4_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp16_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp16_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8a58d742fb7d15d65c12fcce083194f1f75b4446 --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp16_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp16_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp2_pp16_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp2_pp16_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8dd9ab440a1298ed6839cc314f86307631427ee7 --- /dev/null +++ b/configs/config_8.86G_dp1_tp2_pp16_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp2_pp16_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp2_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp2_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0b5328bf73ae8d127d8e8cec42f939ae928313db --- /dev/null +++ b/configs/config_8.86G_dp1_tp2_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp2_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp1_pp8_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp1_pp8_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..73ef5fcbeb722b7fb9c34c6e745f33cb8b729470 --- /dev/null +++ b/configs/config_8.86G_dp2_tp1_pp8_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp1_pp8_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: debug + log_level_replica: debug +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1eea3f9ef272621dc08e77e9b89d0f36e2013333 --- /dev/null +++ b/configs/config_8.86G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp1_pp2_acc8_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp1_pp2_acc8_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..699281a9122862c08c0adbd6dfa1d0bff4226db5 --- /dev/null +++ b/configs/config_8.86G_dp4_tp1_pp2_acc8_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp1_pp2_acc8_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp8_pp16_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp8_pp16_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a924f1bb35113bd05fdba8734ba8184a1bd184d6 --- /dev/null +++ b/configs/config_8.86G_dp4_tp8_pp16_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp8_pp16_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp8_pp4_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp8_pp4_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0152211c698e737cf4047eabeef04e249e9480cd --- /dev/null +++ b/configs/config_8.86G_dp4_tp8_pp4_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp8_pp4_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a16a5397e730c75691b6421219733733a180d4aa --- /dev/null +++ b/configs/config_80G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp16_pp8_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp32_pp1_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp32_pp1_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..06c92bd36b69280ba5d96d0016f9dd2f270f2109 --- /dev/null +++ b/configs/config_80G_dp1_tp32_pp1_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp32_pp1_acc64_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f175c6ffde4ae7c98f13e426e4b81459d5013b93 --- /dev/null +++ b/configs/config_80G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp8_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k_gqa8.yaml b/configs/config_80G_dp2_tp8_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k_gqa8.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a7243f0bdc8d4b0a7e5710621ac96e0a1472c1ef --- /dev/null +++ b/configs/config_80G_dp2_tp8_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k_gqa8.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp8_pp4_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k_gqa8 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp1_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp1_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d4314392c987db3e9b6f1c2065b9f1ec6cda2196 --- /dev/null +++ b/configs/config_80G_dp8_tp1_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp8_tp1_pp2_acc4_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp4_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp4_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b3f1ba03fe5c8abb4fd676fc6d3e7f07baa5a96b --- /dev/null +++ b/configs/config_80G_dp8_tp4_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp8_tp4_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100