Unverified50% confidenceBenchmarkExact time
In a 10M token context test on GLM-5.3-FLASH (QWEN-3.8-27B), D-LSA achieves a prefill speedup of up to 72.4x and a decoding speedup of up to 42.5x compared with full attention.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/2/2026
First Seen
Valid until: 11/16/2026
Sources
Related Entities
Related Claims
UnverifiedThe paper presents comparison data at a 10M token context on GLM-5.3-FLASH (QWEN-3.8-27B).71% similarUnverified在覆盖九大维度共403道题的测试中，Muse-Glimmer-30B综合均分为76.04，比Corpus 3.6 27B的75.57高出0.47分69% similarUnverifiedIn an FP16-precision 4K×4K×4K transposed matrix multiplication task, the M5 Ultra with Neural Accelerator enabled surpassed 107 TFLOPS, which is 3.51 times its own non-accelerated performance and 4.79 times that of the M3 Ultra.68% similarUnverifiedIn a model comparison test, qwen2.5:3b correctly understood pace, produced 0 hallucinated place names, had a length error of 0.3%, peak memory of 259 MB, and took 106 seconds.68% similarVerifiedThe GGUF quantization format (e.g., Q4_K_M) compresses weight precision from 32-bit floating point to 4-bit integers, typically incurring only about 1–5% performance loss.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1126147API
curl https://kongchang.com/api/v1/knowledge/claims/1126147MCP
get_claim(id=1126147)