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:
  - e0c41a65c97fb0ab_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/e0c41a65c97fb0ab_train_data.json
  type:
    field_instruction: prompt
    field_output: org_response
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 5
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: Romain-XV/b6a99682-3529-4be8-b0fb-cb265b79043f
hub_repo: null
hub_strategy: checkpoint
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: 32
lora_dropout: 0.05
lora_fan_in_fan_out: true
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
max_steps: 1762
micro_batch_size: 4
mlflow_experiment_name: /tmp/e0c41a65c97fb0ab_train_data.json
model_type: AutoModelForCausalLM
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: bc469934-f65d-4554-a373-c57006d470f3
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: bc469934-f65d-4554-a373-c57006d470f3
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

b6a99682-3529-4be8-b0fb-cb265b79043f

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: 1.2141

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: 16
• total_train_batch_size: 64
• 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: 1762

Training results

Training Loss	Epoch	Step	Validation Loss
2.1222	0.0002	1	2.3931
1.7759	0.0112	50	1.6894
1.7087	0.0225	100	1.5614
1.5066	0.0337	150	1.4999
1.6564	0.0449	200	1.4552
1.2208	0.0562	250	1.4239
1.3663	0.0674	300	1.3977
1.5511	0.0786	350	1.3783
1.4065	0.0899	400	1.3604
1.3633	0.1011	450	1.3459
1.4855	0.1124	500	1.3321
1.5217	0.1236	550	1.3173
1.3671	0.1348	600	1.3077
1.1679	0.1461	650	1.2967
1.3639	0.1573	700	1.2875
1.3644	0.1685	750	1.2790
1.1246	0.1798	800	1.2719
1.3098	0.1910	850	1.2646
1.2754	0.2022	900	1.2575
1.2915	0.2135	950	1.2512
1.3131	0.2247	1000	1.2458
1.0848	0.2359	1050	1.2410
1.3334	0.2472	1100	1.2370
1.4238	0.2584	1150	1.2331
1.1619	0.2697	1200	1.2288
1.2892	0.2809	1250	1.2258
1.0178	0.2921	1300	1.2233
1.1591	0.3034	1350	1.2208
1.29	0.3146	1400	1.2192
1.0718	0.3258	1450	1.2172
1.0717	0.3371	1500	1.2160
1.1195	0.3483	1550	1.2151
1.0664	0.3595	1600	1.2146
1.3966	0.3708	1650	1.2143
1.3138	0.3820	1700	1.2141
1.267	0.3932	1750	1.2141

Framework versions

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