See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2.5-Math-7B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 99457ca2c5b4418a_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/99457ca2c5b4418a_train_data.json
  type:
    field_instruction: instruction
    field_output: chosen_response
    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/b52a2d34-8fc3-44f6-951b-4e958aa9ebf2
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: 2512
micro_batch_size: 4
mlflow_experiment_name: /tmp/99457ca2c5b4418a_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: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 3d04422d-b103-439f-a90a-131f2d245fa3
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 3d04422d-b103-439f-a90a-131f2d245fa3
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

b52a2d34-8fc3-44f6-951b-4e958aa9ebf2

This model is a fine-tuned version of Qwen/Qwen2.5-Math-7B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.3748

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: 2512

Training results

Training Loss	Epoch	Step	Validation Loss
3.0924	0.0005	1	2.9992
2.2515	0.0487	100	1.9017
1.5833	0.0974	200	1.7677
1.9219	0.1460	300	1.6937
1.7724	0.1947	400	1.6421
1.6048	0.2434	500	1.6005
1.6231	0.2921	600	1.5691
1.4775	0.3407	700	1.5428
1.6057	0.3894	800	1.5166
1.4239	0.4381	900	1.4937
1.3718	0.4868	1000	1.4760
1.4327	0.5354	1100	1.4610
1.5657	0.5841	1200	1.4464
1.4628	0.6328	1300	1.4331
1.3774	0.6815	1400	1.4199
1.3677	0.7301	1500	1.4080
1.3432	0.7788	1600	1.3988
1.3434	0.8275	1700	1.3901
1.2272	0.8762	1800	1.3829
1.3983	0.9249	1900	1.3768
1.5008	0.9735	2000	1.3720
1.1392	1.0222	2100	1.3753
1.3585	1.0709	2200	1.3748

Framework versions

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