Unverified50% confidenceBenchmarkExact time
Under the SDFT setting on Qwen2.5-7B-Instruct, Muon-RT improves average accuracy by 5.4 percentage points over Muon with momentum while reducing persistent memory by 27%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedOn the NinaPro DB1 dataset, experience replay achieves a final accuracy of 83.3% in the class-incremental drift scenario and limits the accuracy drop caused by forgetting to between 1.8% and 5.8%.69% similarUnverifiedMore aggressive quantization yields faster speed and smaller memory footprint but lower precision, while IQ3_XXS offers higher precision but requires close to 47 GB of memory and runs more slowly.66% similarUnverifiedFixed-precision source estimation as persistence l approaches zero requires Θ(l⁻⁴) records.66% similarUnverifiedThrough post-training methods, Swift-Qwen3.8-27B reduces thinking tokens by 58.3% and improves inference speed by 1.95×, while keeping accuracy loss within 1%.65% similarUnverifiedAfter CACPU is integrated into the SGLang inference framework, time to first token (TTFT) for Llama-3.1-8B is reduced by up to 1.74x while model accuracy remains unchanged.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125156API
curl https://kongchang.com/api/v1/knowledge/claims/1125156MCP
get_claim(id=1125156)