See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2.5-3B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 3c4c4a932de7414f_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/3c4c4a932de7414f_train_data.json
  type:
    field_instruction: instruction
    field_output: output
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/0c36c7c1-fd7f-46d4-9ca3-caedcf8fa82b
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: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 1428
micro_batch_size: 4
mlflow_experiment_name: /tmp/3c4c4a932de7414f_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: d1eb1639-a5f0-4c79-ab22-9c35eb5781f0
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: d1eb1639-a5f0-4c79-ab22-9c35eb5781f0
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

0c36c7c1-fd7f-46d4-9ca3-caedcf8fa82b

This model is a fine-tuned version of Qwen/Qwen2.5-3B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7002

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: 1428

Training results

Training Loss	Epoch	Step	Validation Loss
0.8095	0.0014	1	1.0926
0.7614	0.1368	100	0.8260
0.8482	0.2736	200	0.7858
0.7933	0.4103	300	0.7641
0.7378	0.5471	400	0.7460
0.892	0.6839	500	0.7341
0.5835	0.8207	600	0.7223
0.7894	0.9574	700	0.7151
0.8442	1.0942	800	0.7134
0.7056	1.2310	900	0.7109
0.7143	1.3678	1000	0.7060
0.7873	1.5045	1100	0.7032
0.5503	1.6413	1200	0.7018
0.7686	1.7781	1300	0.7004
0.7319	1.9149	1400	0.7002

Framework versions

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