See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2.5-0.5B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 6212221035c10067_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/6212221035c10067_train_data.json
  type:
    field_instruction: problem
    field_output: qwq
    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/c493bf66-e248-4378-8666-90e7b8bd33bc
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: 13885
micro_batch_size: 4
mlflow_experiment_name: /tmp/6212221035c10067_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: 0.04060583911966541
wandb_entity: null
wandb_mode: online
wandb_name: 5684560b-1c7a-4061-99f1-c4168d7aa948
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 5684560b-1c7a-4061-99f1-c4168d7aa948
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

c493bf66-e248-4378-8666-90e7b8bd33bc

This model is a fine-tuned version of Qwen/Qwen2.5-0.5B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5727

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: 13885

Training results

Training Loss	Epoch	Step	Validation Loss
0.8822	0.0003	1	0.9058
0.7206	0.0271	100	0.6582
0.6481	0.0542	200	0.6433
0.6888	0.0813	300	0.6354
0.7084	0.1083	400	0.6275
0.5956	0.1354	500	0.6232
0.6001	0.1625	600	0.6184
0.5065	0.1896	700	0.6151
0.6325	0.2167	800	0.6128
0.6525	0.2438	900	0.6105
0.5587	0.2709	1000	0.6077
0.5955	0.2980	1100	0.6057
0.5601	0.3250	1200	0.6030
0.594	0.3521	1300	0.6014
0.6321	0.3792	1400	0.6001
0.6179	0.4063	1500	0.5978
0.6089	0.4334	1600	0.5959
0.6292	0.4605	1700	0.5943
0.6272	0.4876	1800	0.5936
0.672	0.5147	1900	0.5919
0.55	0.5417	2000	0.5902
0.5409	0.5688	2100	0.5885
0.5965	0.5959	2200	0.5879
0.585	0.6230	2300	0.5866
0.6607	0.6501	2400	0.5851
0.679	0.6772	2500	0.5839
0.6806	0.7043	2600	0.5825
0.7091	0.7314	2700	0.5813
0.5744	0.7584	2800	0.5813
0.4991	0.7855	2900	0.5797
0.6094	0.8126	3000	0.5782
0.6356	0.8397	3100	0.5772
0.6231	0.8668	3200	0.5763
0.5833	0.8939	3300	0.5752
0.564	0.9210	3400	0.5743
0.6711	0.9481	3500	0.5732
0.5064	0.9751	3600	0.5719
0.4811	1.0022	3700	0.5736
0.5398	1.0293	3800	0.5727

Framework versions

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