Unverified50% confidenceBenchmarkExact time
Under the SDFT setting on Qwen2.5-7B-Instruct, Muon-RT improves average accuracy by 2.6 percentage points over AdamW while reducing persistent memory by 42%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedIQ2_XXS以约2.06位的平均比特率对权重编码，相比IQ3_S（约3.44位）或Q4_K_M（约4.83位）压缩率更高但信息损失更大70% similarUnverifiedOn the Qwen2.5-32B model, merely increasing AdamW's first-order momentum coefficient β₁ from its conventional value brought nearly a 2 percentage point improvement in average validation accuracy, without modifying other hyperparameters.70% similarUnverifiedIn accuracy-sensitive scenarios, 7B-scale large-model VLMs still have value relative to GUI-JEPA (an 86% to 92% accuracy retention rate implies an 8% to 14% accuracy loss).69% similarVerifiedINT8 quantization loses approximately 0.5% accuracy while halving VRAM usage; INT4 quantization compresses VRAM to 1/4 of the original with an accuracy loss of approximately 1–3%.69% similarUnverifiedThe accuracy after same-category substitution differs from the original accuracy by no more than 5.2 percentage points.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125155API
curl https://kongchang.com/api/v1/knowledge/claims/1125155MCP
get_claim(id=1125155)