Hugging Face's logo

nguyenthanhdo
/
ViMath-Qwen2-7B-LORA

PEFT
Safetensors
qwen2
Generated from Trainer
Model card Files Community
ViMath-Qwen2-7B-LORA / README.md
nguyenthanhdo's picture
nguyenthanhdo
Upload folder using huggingface_hub
0ebd7b6 verified
preview code
|
raw
history blame contribute delete
4.73 kB
metadata
base_model: Qwen/Qwen2-7B-Instruct
library_name: peft
license: apache-2.0
tags:
  - generated_from_trainer
model-index:
  - name: workspace/axolotl/vinh/Qwen_Qwen2-7B-Instruct-lora-2024-06-29-17-30-14
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.1

base_model: Qwen/Qwen2-7B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false
 
datasets:
  - path: /workspace/axolotl/vinh/PAL/input_output_qwen.json
    type: input_output
  - path: /workspace/axolotl/vinh/INSTRUCT/input_output_qwen.json
    type: input_output
dataset_prepared_path:
val_set_size: 0.05
eval_sample_packing: false
output_dir: /workspace/axolotl/vinh/Qwen_Qwen2-7B-Instruct-lora-2024-06-29-17-30-14

sequence_len: 2048
sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir: 
lora_r: 64
lora_alpha: 128
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 128
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 2e-4

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention: 
flash_attention: true
s2_attention:

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 10
eval_table_size:
eval_max_new_tokens: 512
saves_per_epoch: 2
save_total_limit: 20
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:

workspace/axolotl/vinh/Qwen_Qwen2-7B-Instruct-lora-2024-06-29-17-30-14

This model is a fine-tuned version of Qwen/Qwen2-7B-Instruct on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0911

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: 128
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 10
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss
0.5558 0.0063 1 0.5296
0.1574 0.1011 16 0.1632
0.1279 0.2023 32 0.1379
0.1166 0.3034 48 0.1265
0.1335 0.4045 64 0.1188
0.1145 0.5056 80 0.1134
0.1036 0.6068 96 0.1082
0.0937 0.7079 112 0.1063
0.0934 0.8090 128 0.1029
0.0975 0.9101 144 0.1008
0.0657 1.0113 160 0.0980
0.0671 1.1124 176 0.0990
0.0664 1.2135 192 0.0986
0.0735 1.3146 208 0.0965
0.0694 1.4158 224 0.0944
0.0555 1.5169 240 0.0923
0.0719 1.6180 256 0.0914
0.071 1.7191 272 0.0894
0.073 1.8203 288 0.0876
0.0543 1.9214 304 0.0869
0.043 2.0225 320 0.0866
0.0333 2.1236 336 0.0934
0.0392 2.2248 352 0.0924
0.0453 2.3259 368 0.0919
0.0488 2.4270 384 0.0920
0.0361 2.5281 400 0.0915
0.0357 2.6293 416 0.0912
0.0364 2.7304 432 0.0912
0.0365 2.8315 448 0.0912
0.0338 2.9326 464 0.0911

Framework versions

  • PEFT 0.11.1
  • Transformers 4.41.1
  • Pytorch 2.1.2+cu118
  • Datasets 2.19.1
  • Tokenizers 0.19.1