Instructions to use bigsmoke05/SCOPAI with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use bigsmoke05/SCOPAI with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/CodeQwen1.5-7B") model = PeftModel.from_pretrained(base_model, "bigsmoke05/SCOPAI") - Notebooks
- Google Colab
- Kaggle
File size: 597 Bytes
e08b6b5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | cutoff_len: 512
dataset: scopai
dataset_dir: data
do_train: true
finetuning_type: lora
flash_attn: auto
fp16: true
gradient_accumulation_steps: 4
learning_rate: 0.0002
logging_steps: 1
lora_alpha: 8
lora_dropout: 0.05
lora_rank: 4
lora_target: q_proj,v_proj
lr_scheduler_type: linear
max_grad_norm: 0.5
max_samples: 100
model_name_or_path: Qwen/CodeQwen1.5-7B
num_train_epochs: 2.0
optim: adamw_torch
output_dir: saves/Qwen1.5-Code-7B/lora/train_scopai1
packing: false
per_device_train_batch_size: 2
quantization_bit: 4
report_to: none
save_steps: 100
stage: sft
template: default
warmup_steps: 5
|