LLM Fine-Tuning Misadventures (with QLoRA)

I tried fine-tuning a 7B model on internal company data, and QLoRA is truly innovative. It runs fine on a single 24GB GPU.

However, training was too slow, so I needed to optimize. Reducing batch size and using gradient accumulation helped a bit. The results exceeded expectations, but inference latency is a bit of an issue, so I'm considering switching to a lighter model.

by 스타트업러444

9 answers

Agreed, QLoRA is amazing.

by 카페인중독534 · ▲0

I also tried it with 24GB and it worked better than expected.

by AI덕후876 · ▲0

But isn't the learning speed really slow?

by 알고리즘고수183 · ▲0

Thank you for the tips on adjusting batch size and gradient accumulation. I used 8-bit Adam and found I could reduce memory further. For inference latency, consider lowering the quantization bit or using frameworks like vLLM.

by 문과출신개발자48 · ▲0

I'm not sure QLoRA is that innovative... It doesn't seem much different from LoRA? In fact, I noticed performance degradation due to quantization.

by 밤샘코더327 · ▲0

Haha, I'm also in the middle of a rabbit hole.

by 초보개발자9 · ▲0

Inference latency issues seem to be more affected by prompt length than model size; would caching strategies help improve it?

by 뉴비탈출10 · ▲0

I also went through a similar struggle, but eventually I put a 7B model with 4-bit quantization on an edge device. I trained it with QLoRA and served it with vLLM, and the latency came out to around 200ms. But data quality turned out to be more important...

by 무한도전러905 · ▲0

Do you have a source? According to the QLoRA paper, it says 65B is possible with 24GB.

by 클라우드러버317 · ▲0