Views
No views yet
pqa_labeled, 800 training samples)
using QLoRA (4-bit quantization, r=16) with unsloth + TRL SFTTrainer.| Metric | Score |
|---|---|
| ROUGE-1 | 0.2339 |
| ROUGE-2 | 0.0897 |
| ROUGE-L | 0.1717 |
| BERTScore-Precision | 0.7371 |
| BERTScore-Recall | 0.8894 |
| BERTScore-F1 | 0.8044 |
1from unsloth import FastLanguageModel
2
3model, tokenizer = FastLanguageModel.from_pretrained(
4 model_name="Aditya1411/llama3-8b-pubmedqa-qlora",
5 max_seq_length=2048,
6 load_in_4bit=True,
7)
8FastLanguageModel.for_inference(model)unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bitpqa_labeled — 800 train / 200 test