Nekochu's picture
QLoRA merge
719771a
|
raw
history blame
402 Bytes
metadata
license: apache-2.0
pipeline_tag: text-generation
datasets:
  - Nekochu/novel17_train_alpaca_format
language:
  - fr
  - en

Tool used:

  • hiyouga/LLaMA-Efficient-Tuning

  • qwopqwop200/GPTQ-for-LLaMa

Note: QLoRA training os environ Windows, Python 3.11, CUDA 11.8 on 24GB VRAM:

Know issue: load 4bit version in oobabooga/text-generation-webui give gibberish prompt, use ExLlama instead of AutoGPTQ