vllm-project/fbopt-350m-8bit

This is a copy of yec019/fbopt-350m-8bit at commit 305f804, made on 2026-10-30. vLLM CI uses this checkpoint and the original is on a personal account, so we keep a copy here to reduce supply-chain risk.

This is a model for testing the effect of quantization. Part of the UCSD 23Fall 240D course project. The model is a pretrained model from facebook. Model name is opt-350m. We quantize it with 8 bit quantization provided by bitsandbytes. It is a quantization technique proposed in a paper.