7 ms·
It would be interesting to understand if a ~30B Llama-2 model would be interesting and for what reasons.
by jspisak 3y ago
It would be interesting to understand if a ~30B Llama-2 model would be interesting and for what reasons.
- hnuser123456 3y agoIt would fit on the 24GB top-end consumer graphics cards with quantization.
- brucethemoose2 3y agoLlama 34B is just big enough to fit on a 24GB consumer (or affordable server) GPU. Its also just the right size for llama.cpp inference on machines with 32GB RAM, or 16GB RAM with a 8GB+ GPU. Basically its the most desirable size for AI finetuning hobbyists, and the quality jump from llama v1 13B to llama v1 33B is huge.
- Tostino 3y agoBetter reasoning and general performance than 13b by far (if llama1 was any indication), and like the other user said, can fit on a single 24gb vram gaming card, and can be peft fine-tuned with 2x 24gb cards.
- airgapstopgap 3y agoLlama-1-33B was trained on 40% more tokens than LLama-1-13B; this explained some of the disparity. This time around they both have the same data scale (2T pretraining + 500B code finetune), but 34B is also using GQA which is slightly more noisy than MHA. Furthermore, there have been some weird indications in the original LLama-2 paper that 34B base model is something… even more special, it's been trained on a separate internal cluster with undervolted/underclocked GPUs (though this in itself can't hurt training results), its scores are below expectations, it's been less "aligned". Here, Code-Llama-Instruct-13B is superior to 34B on HumanEval@1. So yes, it's desirable but I wouldn't get my hopes up.