Unverified90% confidenceFactTime unknown
GPT-OSS原生支持FP4混合精度，无需额外量化即可大幅压缩显存占用
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGPT-2（355M）在混合精度（FP16）下微调显存占用约2-4GB，单卡T4即可完成全流程82% similarUnverifiedGPT-4o 在保持 128K 上下文能力的同时实现了更快的推理速度和更低的成本76% similarUnverifiedGPT-4o-mini has fast inference speed and low cost, making it suitable for everyday lightweight coding and code completion.76% similarUnverifiedThe GPT-OSS 20B version activates only 3.6B parameters during inference and achieves performance close to O3-mini.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/23207API
curl https://kongchang.com/api/v1/knowledge/claims/23207MCP
get_claim(id=23207)