Unverified50% confidenceFactExact time
The GC-MoE compute partition runs operand-swapped tcgen05 block-scaled MMA on Blackwell, and on Hopper runs register-A WGMMA with FP4 decoding completed in registers; both fuse SwiGLU, requantization, and optional shared experts.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Related Entities
Related Claims
Unverified主流量化算法GPTQ、GGUF、AWQ在INT4精度下仍能保持较高能力保留率，是HuggingFace和llama.cpp生态的主流格式63% similarUnverifiedgguf_cve_map.json and GGUF_CVE_MAP.md (CC BY 4.0 license) map each GGUF vulnerability advisory to the specific overflowing header field, the CWE number, and the scan rule number that can detect the vulnerability.62% similarUnverifiedggml是llama.cpp、whisper.cpp等高性能本地推理项目背后的底层张量计算库60% similarVerifiedThe mainstream quantization formats in the community include GGUF, AWQ, and GPTQ, among which GGUF was popularized by the llama.cpp project and supports CPU inference.58% similarUnverifiedGQA reduces the size of the KV Cache by allowing multiple Query heads to share the same set of Key-Value heads, which is particularly critical for WebLLM operating in memory-constrained environments.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1133734API
curl https://kongchang.com/api/v1/knowledge/claims/1133734MCP
get_claim(id=1133734)