See axolotl config

axolotl version: 0.4.1

adapter: lora
auto_find_batch_size: false
base_model: Qwen/Qwen2.5-1.5B
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - dd39f638a84431cb_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/dd39f638a84431cb_train_data.json
  type:
    field_instruction: instruction
    field_output: prompt
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
early_stopping_threshold: 1.0e-05
eval_max_new_tokens: 128
eval_steps: 17
eval_strategy: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 6
gradient_checkpointing: false
group_by_length: false
hub_model_id: mrferr3t/372f94d6-bf18-4a6e-9232-6e9a04f5d226
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0004
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 17
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
lr_scheduler: cosine
max_steps: 
micro_batch_size: 12
mlflow_experiment_name: /tmp/dd39f638a84431cb_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 100
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: 17
saves_per_epoch: 0
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: .10000000
wandb_entity: null
wandb_mode: 
wandb_name: 618c487b-3dbf-4791-a39a-d50d7294b4c2
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 618c487b-3dbf-4791-a39a-d50d7294b4c2
warmup_steps: 100
weight_decay: 0.0
xformers_attention: null

372f94d6-bf18-4a6e-9232-6e9a04f5d226

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1308

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: 0.0004
• train_batch_size: 12
• eval_batch_size: 12
• seed: 42
• gradient_accumulation_steps: 6
• total_train_batch_size: 72
• 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: 100
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0286	1	2.6912
2.6955	0.4857	17	2.6122
2.5215	0.9714	34	2.3366
2.1726	1.4571	51	1.9901
1.8983	1.9429	68	1.6772
1.5215	2.4286	85	1.3927
1.3127	2.9143	102	1.1804
1.0685	3.4	119	1.0317
0.9234	3.8857	136	0.8665
0.7219	4.3714	153	0.7275
0.5793	4.8571	170	0.6129
0.4619	5.3429	187	0.4976
0.3749	5.8286	204	0.4233
0.2555	6.3143	221	0.3517
0.2266	6.8	238	0.2796
0.1659	7.2857	255	0.2460
0.1285	7.7714	272	0.2215
0.1081	8.2571	289	0.1978
0.0897	8.7429	306	0.1749
0.069	9.2286	323	0.1605
0.0633	9.7143	340	0.1486
0.0604	10.2	357	0.1447
0.0483	10.6857	374	0.1408
0.0474	11.1714	391	0.1403
0.0395	11.6571	408	0.1299
0.0403	12.1429	425	0.1329
0.0338	12.6286	442	0.1329
0.0341	13.1143	459	0.1308

Framework versions

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