See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: unsloth/Qwen2.5-Math-1.5B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - cb39c063f14002d5_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/cb39c063f14002d5_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/68f80a80-b5c8-447a-92a3-d9433fb29cde
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: 2520
micro_batch_size: 4
mlflow_experiment_name: /tmp/cb39c063f14002d5_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: 0.037878213966455056
wandb_entity: null
wandb_mode: online
wandb_name: 5a1d5196-a942-457d-9fd4-72f486ecfb9f
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 5a1d5196-a942-457d-9fd4-72f486ecfb9f
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

68f80a80-b5c8-447a-92a3-d9433fb29cde

This model is a fine-tuned version of unsloth/Qwen2.5-Math-1.5B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4261

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: 2520

Training results

Training Loss	Epoch	Step	Validation Loss
0.8902	0.0003	1	0.7814
0.4751	0.0252	100	0.4923
0.6004	0.0504	200	0.4770
0.4982	0.0756	300	0.4688
0.4731	0.1008	400	0.4625
0.4162	0.1260	500	0.4577
0.4351	0.1512	600	0.4540
0.4932	0.1764	700	0.4506
0.5687	0.2016	800	0.4482
0.5139	0.2268	900	0.4452
0.426	0.2520	1000	0.4426
0.3966	0.2772	1100	0.4406
0.4305	0.3024	1200	0.4380
0.3815	0.3275	1300	0.4363
0.4487	0.3527	1400	0.4344
0.3897	0.3779	1500	0.4328
0.3705	0.4031	1600	0.4312
0.3748	0.4283	1700	0.4300
0.4569	0.4535	1800	0.4289
0.4704	0.4787	1900	0.4281
0.3975	0.5039	2000	0.4273
0.4603	0.5291	2100	0.4268
0.3964	0.5543	2200	0.4264
0.4499	0.5795	2300	0.4262
0.3652	0.6047	2400	0.4261
0.5563	0.6299	2500	0.4261

Framework versions

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