See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2.5-Math-7B-Instruct
bf16: auto
chat_template: llama3
cosine_min_lr_ratio: 0.1
data_processes: 16
dataset_prepared_path: null
datasets:
- data_files:
  - c354ea2b3ae812fe_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/c354ea2b3ae812fe_train_data.json
  type:
    field_instruction: question
    field_output: fluent_answer
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
device_map: auto
do_eval: true
early_stopping_patience: 1
eval_batch_size: 1
eval_sample_packing: false
eval_steps: 25
evaluation_strategy: steps
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 32
gradient_checkpointing: true
group_by_length: true
hub_model_id: sn56m2/290fe7de-a4c8-4622-90bc-8ea6d591f913
hub_strategy: checkpoint
hub_token: null
hub_username: dada22231
learning_rate: 0.0001
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lora_target_modules:
- q_proj
- v_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_memory:
  0: 70GiB
  1: 70GiB
  2: 70GiB
  3: 70GiB
max_steps: 200
micro_batch_size: 1
mlflow_experiment_name: /tmp/c354ea2b3ae812fe_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optim_args:
  adam_beta1: 0.9
  adam_beta2: 0.95
  adam_epsilon: 1e-5
optimizer: adamw_torch
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 25
save_strategy: steps
sequence_len: 2048
strict: false
tf32: false
tokenizer_type: AutoTokenizer
torch_compile: false
train_on_inputs: false
trust_remote_code: true
val_set_size: 50
wandb_entity: sn56-miner
wandb_mode: disabled
wandb_name: 290fe7de-a4c8-4622-90bc-8ea6d591f913
wandb_project: god
wandb_runid: 290fe7de-a4c8-4622-90bc-8ea6d591f913
warmup_ratio: 0.04
weight_decay: 0.01
xformers_attention: null

290fe7de-a4c8-4622-90bc-8ea6d591f913

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: 0.7452

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• total_eval_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=adam_beta1=0.9,adam_beta2=0.95,adam_epsilon=1e-5
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 8
• training_steps: 200

Training results

Training Loss	Epoch	Step	Validation Loss
4.3138	0.0027	1	5.2631
2.1022	0.0669	25	1.4878
1.5052	0.1337	50	0.9975
1.2566	0.2006	75	0.8683
1.1438	0.2675	100	0.8179
1.1605	0.3343	125	0.7787
1.1987	0.4012	150	0.7635
1.1056	0.4681	175	0.7521
1.1422	0.5349	200	0.7452

Framework versions

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