Unverified50% confidenceFactExact time
GPTQ (2022) achieved the first usable performance of large models at INT4 by quantizing layer by layer and minimizing quantization error using Hessian information.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGPTQ achieved high-quality model compression at 4-bit precision in 2022 through layer-wise optimal second-order error compensation.78% similarUnverifiedGPTQ（2022年）和AWQ（2023年）能将70B参数的LLM量化到INT4精度，困惑度损失不到1%76% similarUnverifiedQuantization methods such as GPTQ, AWQ, and SqueezeLLM can still maintain over 95% of the original model's performance at 4-bit precision.71% similarUnverified在GPT-4发布初期，主流模型在GPQA上的准确率约为35-40%，人类领域专家基准线约为65%，随机猜测为25%69% similarUnverifiedQ4量化可将模型体积缩减至原来的约四分之一，MLX原生支持分组量化（Group Quantization）67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/828100API
curl https://kongchang.com/api/v1/knowledge/claims/828100MCP
get_claim(id=828100)