search

artidoro / qlora

QLoRA: Efficient Finetuning of Quantized LLMs
https://arxiv.org/abs/2305.14314
MIT License
9.74k stars 800 forks source link

Table 4 and Table 5 have different results #285

Open dekoponTree opened 6 months ago

dekoponTree commented 6 months ago

image

image

Llama-7b with QLoRA finetuned on Alpaca has different results in Table 4 and Table 5.

The former is 39.0, while the latter is 38.8