diff --git a/configs/config_1.14G_dp16_tp4_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp4_pp1_acc1_mbs32_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c9376fe6a6165027696444716ad25be1ce8f9ea9 --- /dev/null +++ b/configs/config_1.14G_dp16_tp4_pp1_acc1_mbs32_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_dp16_tp4_pp1_acc1_mbs32_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: 16 + 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: 32 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp8_pp1_acc8_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp16_tp8_pp1_acc8_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..949b1a4e46aae633f27fada84f7fce1b01237251 --- /dev/null +++ b/configs/config_1.14G_dp16_tp8_pp1_acc8_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_dp16_tp8_pp1_acc8_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: 16 + 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: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs128_seq2048_zero0_tpmodeRED_vocab32k_prof.yaml b/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs128_seq2048_zero0_tpmodeRED_vocab32k_prof.yaml new file mode 100644 index 0000000000000000000000000000000000000000..112c24cb312193223e8da77c6d936c29e9e07880 --- /dev/null +++ b/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs128_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_dp1_tp8_pp1_acc1_mbs128_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: 1 + 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: 128 + sequence_length: 2048 + train_steps: 10 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp1_pp4_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp1_pp4_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fdedc9e1182f6d4e5b6b2bdc5c81b0829f67ad54 --- /dev/null +++ b/configs/config_1.14G_dp2_tp1_pp4_acc1_mbs16_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: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp1_pp4_acc1_mbs16_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: 2 + 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: 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_1.14G_dp2_tp1_pp8_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp1_pp8_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5b5c3404e3ae0dab70b3aaefc93e70b0e55e54dd --- /dev/null +++ b/configs/config_1.14G_dp2_tp1_pp8_acc256_mbs1_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: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp1_pp8_acc256_mbs1_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: 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: 256 + 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_1.14G_dp2_tp64_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp64_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c73eb6fab4ed85e12b438b1b89cd981c00badbed --- /dev/null +++ b/configs/config_1.14G_dp2_tp64_pp1_acc16_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_dp2_tp64_pp1_acc16_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: 2 + 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: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp64_pp1_acc1_mbs64_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp64_pp1_acc1_mbs64_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b720abe3437d1f9fd425e73b991072849dfe358 --- /dev/null +++ b/configs/config_1.14G_dp2_tp64_pp1_acc1_mbs64_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_tp64_pp1_acc1_mbs64_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: 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: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp64_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp64_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9337e0a5f96164f650514722ba0183251a5b72fb --- /dev/null +++ b/configs/config_1.14G_dp2_tp64_pp1_acc32_mbs32_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_tp64_pp1_acc32_mbs32_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: 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: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp8_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp8_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fe008b9151d4d51b06c902caa8e0df7c5fdf053f --- /dev/null +++ b/configs/config_1.14G_dp2_tp8_pp1_acc64_mbs16_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_tp8_pp1_acc64_mbs16_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: 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: 64 + 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_1.14G_dp32_tp4_pp1_acc16_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp32_tp4_pp1_acc16_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ad400d3031c748894c746a405c392e35f1e9edbe --- /dev/null +++ b/configs/config_1.14G_dp32_tp4_pp1_acc16_mbs4_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_dp32_tp4_pp1_acc16_mbs4_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: 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: 16 + 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_dp4_tp2_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp2_pp1_acc16_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..23ebc6f71543ba58b64e0ce27c7571aafc4b9561 --- /dev/null +++ b/configs/config_1.14G_dp4_tp2_pp1_acc16_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_dp4_tp2_pp1_acc16_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: 4 + 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: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp64_pp1_acc4_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp64_pp1_acc4_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3fe79f7406810e100b00a9c43cfa6530f9ccdca2 --- /dev/null +++ b/configs/config_1.14G_dp4_tp64_pp1_acc4_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_tp64_pp1_acc4_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: 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: 4 + 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_dp4_tp8_pp1_acc32_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp8_pp1_acc32_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b4a07cc2fe9a08423e3e61a6fa85e05aa7d4eff9 --- /dev/null +++ b/configs/config_1.14G_dp4_tp8_pp1_acc32_mbs1_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_dp4_tp8_pp1_acc32_mbs1_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: 4 + 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: 32 + 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_dp64_tp8_pp1_acc1_mbs8_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp64_tp8_pp1_acc1_mbs8_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..accee464fb33cec6a4110fddfbc78687da02391b --- /dev/null +++ b/configs/config_1.14G_dp64_tp8_pp1_acc1_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_dp64_tp8_pp1_acc1_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: 64 + 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: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp1_pp2_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp1_pp2_acc2_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..772fffd392b62a0cfc58e715b72416e4ad286351 --- /dev/null +++ b/configs/config_1.14G_dp8_tp1_pp2_acc2_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_dp8_tp1_pp2_acc2_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: 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: 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_tp8_pp1_acc2_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp8_pp1_acc2_mbs32_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0ef04abee0b9d9e9bc6bdc83adbec7546b510a6b --- /dev/null +++ b/configs/config_1.14G_dp8_tp8_pp1_acc2_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_dp8_tp8_pp1_acc2_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: 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: 2 + 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_tp1_pp1_acc8_mbs4_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp1_pp1_acc8_mbs4_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d80f218dd0594e8cdcea5352ab3740a6f498b041 --- /dev/null +++ b/configs/config_1.34G_dp16_tp1_pp1_acc8_mbs4_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_acc8_mbs4_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: 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_pp8_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp1_pp8_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b3f58e58399447198bee5ee67475f9617535d688 --- /dev/null +++ b/configs/config_1.34G_dp16_tp1_pp8_acc8_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: 1.34G_dp16_tp1_pp8_acc8_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: 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: 16 + 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: 8 + 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_1.34G_dp16_tp4_pp1_acc32_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp4_pp1_acc32_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..893065dd2dc414945879f4b8f4734b428a04d05a --- /dev/null +++ b/configs/config_1.34G_dp16_tp4_pp1_acc32_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_dp16_tp4_pp1_acc32_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: 16 + 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: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp4_pp1_acc4_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp16_tp4_pp1_acc4_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7bbd679099cfa2df3769f64222a584e5cb676e3c --- /dev/null +++ b/configs/config_1.34G_dp16_tp4_pp1_acc4_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_dp16_tp4_pp1_acc4_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: 16 + 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: 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_dp16_tp8_pp1_acc4_mbs32_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp8_pp1_acc4_mbs32_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0fbeed431a5af880eef24aafd9820913ecd1bb97 --- /dev/null +++ b/configs/config_1.34G_dp16_tp8_pp1_acc4_mbs32_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_dp16_tp8_pp1_acc4_mbs32_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: 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: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp16_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp16_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e1bb667e0d42e1d1e6a2fb4e8c67d2a5bc2a64d2 --- /dev/null +++ b/configs/config_1.34G_dp1_tp16_pp2_acc4_mbs64_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_pp2_acc4_mbs64_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: 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: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp4_pp16_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp4_pp16_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c50d88fa1a30461fd7f07f3853c0fb81a53cf01b --- /dev/null +++ b/configs/config_1.34G_dp1_tp4_pp16_acc16_mbs16_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_tp4_pp16_acc16_mbs16_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: 16 + 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: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp1_pp8_acc32_mbs32_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp1_pp8_acc32_mbs32_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a4c1de5d5fed6814d089f85efdbaf96c28a0a2c5 --- /dev/null +++ b/configs/config_1.34G_dp2_tp1_pp8_acc32_mbs32_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: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp1_pp8_acc32_mbs32_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: 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: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp1_pp8_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp1_pp8_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..751c0e1752965af5955e121dcda00af23d651bc9 --- /dev/null +++ b/configs/config_1.34G_dp2_tp1_pp8_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: 1.34G_dp2_tp1_pp8_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: 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: 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: 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_1.34G_dp2_tp32_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp32_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..44c3c1b4152e8772ae2a75fae018ce6052c8618d --- /dev/null +++ b/configs/config_1.34G_dp2_tp32_pp1_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_dp2_tp32_pp1_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: 2 + 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: 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_dp2_tp4_pp1_acc8.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp4_pp1_acc8.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5e2d086412bbbf7e06c054c85d233257e74053f8 --- /dev/null +++ b/configs/config_1.34G_dp2_tp4_pp1_acc8.0_mbs16_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_dp2_tp4_pp1_acc8.0_mbs16_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: 2 + 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: 8.0 + 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_1.34G_dp2_tp64_pp1_acc32_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp64_pp1_acc32_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0cc1605a84df643832392da588915f8f987d793e --- /dev/null +++ b/configs/config_1.34G_dp2_tp64_pp1_acc32_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_dp2_tp64_pp1_acc32_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: 2 + 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: 32 + 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_dp32_tp2_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp2_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..743aae07617113b79defb159c856c5e8ee44ad9e --- /dev/null +++ b/configs/config_1.34G_dp32_tp2_pp1_acc1_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_dp32_tp2_pp1_acc1_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: 32 + 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: 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_1.34G_dp32_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp32_tp2_pp1_acc32_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bdb6a379f05019ee2a601050e2d84fc7e3cd0459 --- /dev/null +++ b/configs/config_1.34G_dp32_tp2_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_tp2_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: 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_1.34G_dp32_tp8_pp1_acc1_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp32_tp8_pp1_acc1_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5b41908adb613ae33dbcbcde9175f2c99bc9e9f2 --- /dev/null +++ b/configs/config_1.34G_dp32_tp8_pp1_acc1_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_dp32_tp8_pp1_acc1_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: 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: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp128_pp1_acc32_mbs1_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp128_pp1_acc32_mbs1_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8c8bcd185207ac3e0eb7c9852e07b2ddae529ff4 --- /dev/null +++ b/configs/config_1.34G_dp4_tp128_pp1_acc32_mbs1_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_tp128_pp1_acc32_mbs1_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: 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: 32 + 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_dp4_tp1_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp1_pp2_acc64_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..13be3ecceff3cd52833366f390d50649dcfd9c09 --- /dev/null +++ b/configs/config_1.34G_dp4_tp1_pp2_acc64_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: 1.34G_dp4_tp1_pp2_acc64_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: 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: 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: 64 + 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_1.34G_dp4_tp32_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bbb8ecc7bc40337ee2c08bdb2dadae3f992dc239 --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp4_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: 1.34G_dp4_tp32_pp4_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: 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: 4 + 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: 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_1.34G_dp4_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4df5b260ab8708fd3e422fe07db8390fd3108f4 --- /dev/null +++ b/configs/config_1.34G_dp4_tp4_pp1_acc4_mbs32_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_tp4_pp1_acc4_mbs32_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: 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: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp64_pp1_acc128_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp64_pp1_acc128_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..86700ff3e61928a31f68d4488e2dcfe35a2a6060 --- /dev/null +++ b/configs/config_1.34G_dp4_tp64_pp1_acc128_mbs1_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_acc128_mbs1_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: 128 + 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_1.34G_dp64_tp4_pp1_acc8_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp64_tp4_pp1_acc8_mbs4_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..67273b15b699216c8e0d4c2c49f6656b254fd86b --- /dev/null +++ b/configs/config_1.34G_dp64_tp4_pp1_acc8_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_dp64_tp4_pp1_acc8_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: 64 + 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: 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_tp16_pp1_acc16_mbs1_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp16_pp1_acc16_mbs1_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cd23d6288cb889e841392e68fd41acfa9fdb4eb2 --- /dev/null +++ b/configs/config_1.34G_dp8_tp16_pp1_acc16_mbs1_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_dp8_tp16_pp1_acc16_mbs1_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: 8 + 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: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp16_pp1_acc2_mbs8_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp16_pp1_acc2_mbs8_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6d48329197a80ac41983e4c42cd9d2c8f5837d0f --- /dev/null +++ b/configs/config_1.34G_dp8_tp16_pp1_acc2_mbs8_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_tp16_pp1_acc2_mbs8_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: 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: 8 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp16_pp1_acc32_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp16_pp1_acc32_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b7fba8d70754d46c4ae9d9318ba19fe368b47ee --- /dev/null +++ b/configs/config_1.34G_dp8_tp16_pp1_acc32_mbs8_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_dp8_tp16_pp1_acc32_mbs8_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: 8 + 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: 32 + 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_1.34G_dp8_tp32_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp32_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d54b762bcd3b822dee04469e92c4f2d07e005ffb --- /dev/null +++ b/configs/config_1.34G_dp8_tp32_pp1_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_tp32_pp1_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: 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: 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_acc64_mbs1_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp32_pp1_acc64_mbs1_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a0faa6084d07c63b1e4f396fc8d57fb7dc4f4854 --- /dev/null +++ b/configs/config_1.34G_dp8_tp32_pp1_acc64_mbs1_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_tp32_pp1_acc64_mbs1_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: 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: 64 + 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_dp8_tp64_pp1_acc1_mbs64_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp64_pp1_acc1_mbs64_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f8a259cf6c0c4c9fdf9c91b18c2a51a3bca82e0c --- /dev/null +++ b/configs/config_1.34G_dp8_tp64_pp1_acc1_mbs64_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_dp8_tp64_pp1_acc1_mbs64_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: 8 + 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: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.27G_dp4_tp16_pp8_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml b/configs/config_3.27G_dp4_tp16_pp8_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f0bc8b0b4be16aff2af7bf3eeda13ad055a9a536 --- /dev/null +++ b/configs/config_3.27G_dp4_tp16_pp8_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.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: 3.27G_dp4_tp16_pp8_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24 + 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: 2048 + num_attention_heads: 24 + num_hidden_layers: 28 + num_key_value_heads: 24 + 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: 4 + 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: 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_3.27G_dp8_tp1_pp1_acc1_mbs2_seq4096_zero0_l28_h3072_heads24.yaml b/configs/config_3.27G_dp8_tp1_pp1_acc1_mbs2_seq4096_zero0_l28_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a369782cd9ff09a98c52fad1a051ec74935b2d49 --- /dev/null +++ b/configs/config_3.27G_dp8_tp1_pp1_acc1_mbs2_seq4096_zero0_l28_h3072_heads24.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: 3.27G_dp8_tp1_pp1_acc1_mbs2_seq4096_zero0_l28_h3072_heads24 + 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: 24 + num_hidden_layers: 28 + num_key_value_heads: 24 + 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: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + 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: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp16_tp2_pp16_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp16_tp2_pp16_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..807e4cb1ec9061a11a02607e7b4ffcfddf8fb5c0 --- /dev/null +++ b/configs/config_3.57G_dp16_tp2_pp16_acc16_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_dp16_tp2_pp16_acc16_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: 16 + 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: 16 + 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_dp1_tp1_pp32_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp1_tp1_pp32_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d6bc1c3c8684658d02da4d004d74dc548011e853 --- /dev/null +++ b/configs/config_3.57G_dp1_tp1_pp32_acc1_mbs256_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_tp1_pp32_acc1_mbs256_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: 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: 256 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp16_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp16_pp2_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..80621831b0e4d54f650f881122c0feb494dd59cf --- /dev/null +++ b/configs/config_3.57G_dp2_tp16_pp2_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: 3.57G_dp2_tp16_pp2_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: 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: 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: 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_3.57G_dp4_tp4_pp32_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp4_tp4_pp32_acc32_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..df95649a9abf0eea6b6365eccc7c71c315fa9a32 --- /dev/null +++ b/configs/config_3.57G_dp4_tp4_pp32_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: 3.57G_dp4_tp4_pp32_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: 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: 4 + expert_parallel_size: 1 + pp: 32 + 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: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_37.8G_dp64_tp4_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml b/configs/config_37.8G_dp64_tp4_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2843cc37f0e4fdb4f0b3b003791b2eea6c7b8bf0 --- /dev/null +++ b/configs/config_37.8G_dp64_tp4_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.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: 37.8G_dp64_tp4_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64 + 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: 131072 + is_llama_config: true + max_position_embeddings: 2048 + 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: 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: 64 + 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: 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_467G_dp4_tp4_pp8_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab49k_gqa8.yaml b/configs/config_467G_dp4_tp4_pp8_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab49k_gqa8.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6ec585a94a8e0331699b2504b09b761d5c0c4295 --- /dev/null +++ b/configs/config_467G_dp4_tp4_pp8_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab49k_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: 467G_dp4_tp4_pp8_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab49k_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: 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: 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: 49152 +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: 8 + 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: 64 + 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_dp1_tp16_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp1_tp16_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..be6d0e81358d063f50a216be108a6fc8233f61f0 --- /dev/null +++ b/configs/config_469G_dp1_tp16_pp4_acc16_mbs16_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_tp16_pp4_acc16_mbs16_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: 4 + 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: 16 + 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_469G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp32_pp2_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff50afc217a7cb0964c68ac94018e580a3bf897a --- /dev/null +++ b/configs/config_469G_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: 469G_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: 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: 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_469G_dp2_tp8_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp8_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..851aec37fe8a0c3a2526f5ca82dc0be28adb7a6a --- /dev/null +++ b/configs/config_469G_dp2_tp8_pp8_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_tp8_pp8_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: 8 + 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: 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_8.86G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12c8dd9725926c767246f4b1fcdbfb5dbd6e4712 --- /dev/null +++ b/configs/config_8.86G_dp16_tp32_pp1_acc4_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_dp16_tp32_pp1_acc4_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: 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: 4 + 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_dp1_tp4_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp4_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1e8278dc98deb2e8a655d3c553d15558aecfa677 --- /dev/null +++ b/configs/config_8.86G_dp1_tp4_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_tp4_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: 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: 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_dp1_tp8_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_8.86G_dp1_tp8_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..620f10a1b669caaf274fa841a85d1d9d1c12d4ea --- /dev/null +++ b/configs/config_8.86G_dp1_tp8_pp1_acc1_mbs4_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: 8.86G_dp1_tp8_pp1_acc1_mbs4_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: 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: 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: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp2_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp2_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7e7cf4cbfb07ae6a0a99ec5dbc3b7d9f8210338e --- /dev/null +++ b/configs/config_8.86G_dp32_tp2_pp1_acc8_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_dp32_tp2_pp1_acc8_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: 32 + 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: 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_8.86G_dp32_tp2_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp2_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d410f773df1585e39e889cdddc4c0d5837ceee8f --- /dev/null +++ b/configs/config_8.86G_dp32_tp2_pp4_acc4_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_dp32_tp2_pp4_acc4_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: 32 + 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: 4 + 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_8.86G_dp4_tp2_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp2_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..115bcdeb6e6960bcf4dd48b9c1fa73072175cebc --- /dev/null +++ b/configs/config_8.86G_dp4_tp2_pp4_acc64_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: 8.86G_dp4_tp2_pp4_acc64_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: 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: 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: 64 + 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_dp8_tp16_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp8_tp16_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8cdffa8c220a508c9088b7c3bdd8f444ba53efdb --- /dev/null +++ b/configs/config_8.86G_dp8_tp16_pp2_acc8_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_dp8_tp16_pp2_acc8_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: 8 + 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: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp8_tp1_pp1_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp8_tp1_pp1_acc32_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4f05751f712dd903873f2718df2d84f94a350cb --- /dev/null +++ b/configs/config_8.86G_dp8_tp1_pp1_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: 8.86G_dp8_tp1_pp1_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: 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: 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: 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_8.86G_dp8_tp2_pp8_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp8_tp2_pp8_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cd074cb5febe955dcda9618fd078c869b6ea628d --- /dev/null +++ b/configs/config_8.86G_dp8_tp2_pp8_acc8_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_dp8_tp2_pp8_acc8_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: 8 + expert_parallel_size: 1 + pp: 8 + 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: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp8_tp4_pp4_acc4_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp8_tp4_pp4_acc4_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d50590abb1e3f2255410467445d85d45eb6c873e --- /dev/null +++ b/configs/config_8.86G_dp8_tp4_pp4_acc4_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: 8.86G_dp8_tp4_pp4_acc4_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: 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: 8 + 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: 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_dp16_tp2_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp2_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..76109adf8bd3379e6ec4739d3b61e2e1610133da --- /dev/null +++ b/configs/config_80G_dp16_tp2_pp2_acc2_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_dp16_tp2_pp2_acc2_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: 16 + 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: 2 + 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_dp1_tp16_pp1_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp16_pp1_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1796dc89cdb10b1a5714b95e96c1087c98f6c8c1 --- /dev/null +++ b/configs/config_80G_dp1_tp16_pp1_acc16_mbs16_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_pp1_acc16_mbs16_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: 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: 16 + 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_dp1_tp2_pp64_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp2_pp64_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ae07a82bbd8b7f542acb8a66b56cd22af46cca0c --- /dev/null +++ b/configs/config_80G_dp1_tp2_pp64_acc1_mbs256_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_tp2_pp64_acc1_mbs256_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: 64 + 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: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp32_pp2_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp32_pp2_acc256_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1afb352de8c09cf5a9fdadd088b24eafdcdf63a4 --- /dev/null +++ b/configs/config_80G_dp1_tp32_pp2_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_tp32_pp2_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: 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: 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_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp1_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..678bdbc6cbc12c7057991502edf643c08da8c63b --- /dev/null +++ b/configs/config_80G_dp2_tp1_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: 80G_dp2_tp1_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: 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: 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: 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_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp2_pp32_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b3e7f620d448efa31ee3d607f23a222e67815d81 --- /dev/null +++ b/configs/config_80G_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: 80G_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: 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: 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_80G_dp64_tp2_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp64_tp2_pp2_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..45faddb37e715032d4f0a06f8e863c2df2cf2409 --- /dev/null +++ b/configs/config_80G_dp64_tp2_pp2_acc1_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: 80G_dp64_tp2_pp2_acc1_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: 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: 64 + 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: 1 + 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_dp8_tp2_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp2_pp2_acc16_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3b536445f041078b475621297320a39d71473a8c --- /dev/null +++ b/configs/config_80G_dp8_tp2_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_tp2_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: 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: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp2_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp2_pp2_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e25d7ad153954b8f6db5ae0499f3111e3cf3455b --- /dev/null +++ b/configs/config_80G_dp8_tp2_pp2_acc8_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: 80G_dp8_tp2_pp2_acc8_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: 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: 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: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp128_tp1_pp1_acc4_mbs2_seq4096_zero0_l30_h3072_heads24.yaml b/configs/config_dp128_tp1_pp1_acc4_mbs2_seq4096_zero0_l30_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a76ec8f4a0e10e1fd7467c83e56ab35f45cc9c9 --- /dev/null +++ b/configs/config_dp128_tp1_pp1_acc4_mbs2_seq4096_zero0_l30_h3072_heads24.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.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp128_tp1_pp1_acc4_mbs2_seq4096_zero0_l30_h3072_heads24 + 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: 24 + num_hidden_layers: 30 + num_key_value_heads: 24 + 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: 49152 +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: 128 + 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: 4 + 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_dp64_tp1_pp1_acc1_mbs8_seq2048.yaml b/configs/config_dp64_tp1_pp1_acc1_mbs8_seq2048.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eed06fb5d797a5598d8e7c889868cdcae3eb2ece --- /dev/null +++ b/configs/config_dp64_tp1_pp1_acc1_mbs8_seq2048.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.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp64_tp1_pp1_acc1_mbs8_seq2048 + 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: 24 + 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: 49152 +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: 64 + 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: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100