adapter: lora
base_model: Qwen/Qwen2.5-0.5B-Instruct
bf16: false
chat_template: chatml
dataset_prepared_path: last_run_prepared
datasets:
- field_messages: messages
  path: shenghuanjiuer/sft-train
  type: chat_template
eval_sample_packing: true
evals_per_epoch: 1
flash_attention: true
fp16: true
gradient_accumulation_steps: 2
gradient_checkpointing: true
group_by_length: false
hub_model_id: shenghuanjiuer/fr-lora-mini
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: false
logging_steps: 1
lora_alpha: 8
lora_dropout: 0.05
lora_r: 4
lora_target_modules:
- q_proj
- v_proj
loss_watchdog_patience: 4
loss_watchdog_threshold: 8
lr_scheduler: cosine
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_8bit
output_dir: /runpod-volume/fine-tuning/test-run
pad_to_sequence_len: true
push_to_hub: true
run_name: test-run
runpod_job_id: 0343c0e6-ca53-4c21-bdbf-f5df3f5d92a0-e1
sample_packing: true
saves_per_epoch: 1
sequence_len: 2048
special_tokens:
  pad_token: <|end_of_text|>
strict: true
tf32: false
train_on_inputs: false
val_set_size: 0.1
warmup_steps: 10
weight_decay: 0

fr-lora-mini

This model is a fine-tuned version of Qwen/Qwen2.5-0.5B-Instruct on the shenghuanjiuer/sft-train dataset.

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: 1
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 2
optimizer: Use OptimizerNames.ADAMW_8BIT 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: 14
mixed_precision_training: Native AMP

Training results

Framework versions

PEFT 0.15.2
Transformers 4.52.4
Pytorch 2.6.0+cu124
Datasets 3.6.0
Tokenizers 0.21.1