See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2.5-0.5B
bf16: true
chat_template: llama3
cosine_min_lr_ratio: 0.3
dataset_prepared_path: null
datasets:
- data_files:
  - cd10afd0d2d9b53d_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/cd10afd0d2d9b53d_train_data.json
  type:
    field_input: reasoning
    field_instruction: instruction
    field_output: refined_answer
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 4
eval_max_new_tokens: 128
eval_steps: 200
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: false
hub_model_id: Romain-XV/81bab74b-510e-46d9-9ef5-e1aa43614ada
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: 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
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 16758
micro_batch_size: 4
mlflow_experiment_name: /tmp/cd10afd0d2d9b53d_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
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: 200
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: a8b81e29-b689-475e-9f49-2b7c12effb9e
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: a8b81e29-b689-475e-9f49-2b7c12effb9e
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

81bab74b-510e-46d9-9ef5-e1aa43614ada

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: 0.6156

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: 4
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 2835

Training results

Training Loss	Epoch	Step	Validation Loss
0.8904	0.0011	1	0.8621
0.7232	0.2117	200	0.6833
0.6934	0.4234	400	0.6577
0.7057	0.6351	600	0.6387
0.5769	0.8468	800	0.6261
0.4268	1.0585	1000	0.6224
0.4512	1.2702	1200	0.6236
0.5911	1.4819	1400	0.6145
0.5223	1.6936	1600	0.6081
0.5189	1.9053	1800	0.5980
0.2908	2.1170	2000	0.6211
0.4433	2.3287	2200	0.6228
0.4402	2.5404	2400	0.6200
0.3407	2.7521	2600	0.6156

Framework versions

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