See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: unsloth/Qwen2-0.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 148a3da1572f97b6_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/148a3da1572f97b6_train_data.json
  type:
    field_instruction: instruction
    field_output: response_8b_instruct
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
device_map:
  ? ''
  : 0,1,2,3,4,5,6,7
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/225892d5-fc4b-4e5d-982b-9b9758f1d465
hub_repo: null
hub_strategy: null
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: 128
lora_dropout: 0.1
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 6883
micro_batch_size: 4
mlflow_experiment_name: /tmp/148a3da1572f97b6_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 10
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: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 0b732679-6afe-438e-9166-ce18203e8392
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 0b732679-6afe-438e-9166-ce18203e8392
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

225892d5-fc4b-4e5d-982b-9b9758f1d465

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

• Loss: 0.8975

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: 6883

Training results

Training Loss	Epoch	Step	Validation Loss
1.1411	0.0004	1	1.1328
0.997	0.0414	100	0.9928
0.9918	0.0828	200	0.9765
0.9758	0.1242	300	0.9664
1.0321	0.1657	400	0.9589
0.9145	0.2071	500	0.9534
0.9221	0.2485	600	0.9484
0.9491	0.2899	700	0.9426
0.9197	0.3313	800	0.9395
0.8908	0.3727	900	0.9348
0.9304	0.4141	1000	0.9311
0.9634	0.4556	1100	0.9282
0.8384	0.4970	1200	0.9245
0.9663	0.5384	1300	0.9215
0.8786	0.5798	1400	0.9180
1.0157	0.6212	1500	0.9154
0.9045	0.6626	1600	0.9139
0.9063	0.7040	1700	0.9108
0.8433	0.7455	1800	0.9071
0.916	0.7869	1900	0.9045
0.911	0.8283	2000	0.9021
0.892	0.8697	2100	0.8995
0.8842	0.9111	2200	0.8974
0.9172	0.9525	2300	0.8951
0.9257	0.9939	2400	0.8918
0.7979	1.0354	2500	0.8967
0.7197	1.0768	2600	0.8975

Framework versions

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