Expired50% confidenceEventExact time
The UP host fine-tuned Qwen3-4B with approximately 100 to 200 additional identity-stability-related data samples specifically to resolve the character confusion issue.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
8/24/2026
First Seen
Valid until: 9/7/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedQwen3.6 35B模型包含256个专家，每个Token仅激活8个路由专家和1个共享专家，实际推理时激活参数约为3B71% similarUnverifiedQ4_K_M采用分组量化策略，将权重张量划分为64或256维子块并独立计算缩放因子与零点，能将困惑度损失控制在1–3%以内65% similarUnverifiedQwen3.6 27B NVFP4 achieved a perfect score of 100 in the tool-calling (ToolCore) test.62% similarUnverifiedQwen3 27B在OpenAI HumanEval 164道Python题上取得85%通过率和99%应答率62% similarUnverifiedIn the Q4_K_XL scheme, MoE layers use MXFP4 and the remaining layers are compressed to Q8_0, achieving 96% same on top-1% prediction consistency62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/796826API
curl https://kongchang.com/api/v1/knowledge/claims/796826MCP
get_claim(id=796826)