Unverified50% confidenceFactExact time
v5.20.1 fixes the DeepGEMM Mega MoE expert parallelism issue (#49428).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedDeepSeek-V4.1-Flash adopts a MoE (Mixture of Experts) backbone architecture with 552B parameters.68% similarUnverifiedThe reap576_iq2xxs (478GB) version successfully resolved SWE-Lancer tasks 14294, 15815_1, and 15925.62% similarUnverifiedMoE is a sparse activation architecture; DeepSeek V3 has 256 experts but activates only 8 at a time.61% similarUnverifiedDeepGEMM Ascend was first released on 2026.09.30, supports Ascend 950 devices, is fully compatible with DeepGEMM's API, supports BF16, FP8, and FP4 matrix multiplication, and uses the same package name as the original version.61% similarUnverifiedv5.20.1 fixes the HunYuan-VL batched generation image allocation issue (#49557).60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1044977API
curl https://kongchang.com/api/v1/knowledge/claims/1044977MCP
get_claim(id=1044977)