See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-0.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - c64ff0d01392d1e4_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/c64ff0d01392d1e4_train_data.json
  type:
    field_instruction: prompt_type
    field_output: prompt_text
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/c31a0826-774c-48af-86bb-629bd7ef2583
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 32
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 8832
micro_batch_size: 4
mlflow_experiment_name: /tmp/c64ff0d01392d1e4_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 2
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 100
sequence_len: 1024
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.04
wandb_entity: null
wandb_mode: online
wandb_name: fb46a0d2-7710-4f02-ba9b-a717c0c8c0cd
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: fb46a0d2-7710-4f02-ba9b-a717c0c8c0cd
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

c31a0826-774c-48af-86bb-629bd7ef2583

This model is a fine-tuned version of Qwen/Qwen2-0.5B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.8095

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• training_steps: 3084

Training results

Training Loss	Epoch	Step	Validation Loss
5.3671	0.0006	1	5.3415
3.1325	0.0649	100	3.3948
3.3967	0.1297	200	3.2506
2.9834	0.1946	300	3.1461
2.6376	0.2594	400	3.0808
3.4206	0.3243	500	2.9966
3.0461	0.3892	600	2.9330
2.2323	0.4540	700	2.8605
2.7525	0.5189	800	2.7908
2.8358	0.5838	900	2.7131
2.8312	0.6486	1000	2.6433
2.4678	0.7135	1100	2.5774
2.5489	0.7783	1200	2.5080
2.3458	0.8432	1300	2.4473
2.3761	0.9081	1400	2.3796
2.0236	0.9729	1500	2.3125
1.9383	1.0378	1600	2.2606
1.8816	1.1026	1700	2.1880
1.8313	1.1675	1800	2.1419
1.9847	1.2324	1900	2.0941
1.8436	1.2972	2000	2.0431
1.6931	1.3621	2100	2.0023
1.593	1.4269	2200	1.9543
1.8932	1.4918	2300	1.9170
2.0395	1.5567	2400	1.8831
1.7951	1.6215	2500	1.8598
1.5655	1.6864	2600	1.8419
1.2855	1.7513	2700	1.8256
1.4709	1.8161	2800	1.8161
1.7402	1.8810	2900	1.8116
1.7177	1.9458	3000	1.8095

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1