Expired85% confidenceFactTime unknown
GLM-4.6 tracks active tasks in memory, and all state is lost if the application crashes.
1
Sources
85%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedGLM-4.6's tool calls failed in reasoning mode, and the evaluation team had to disable reasoning mode to get it running properly.61% similarUnverified显存释放通常通过torch.cuda.empty_cache()清理缓存，配合删除模型对象引用并触发gc.collect()垃圾回收，某些框架下需终止推理子进程才能完全释放54% similarUnverifiedThe GLM series uses an Autoregressive Blank Infilling training objective, which randomly masks text spans and requires the model to autoregressively restore them54% similarUnverified寄存器溢出（Register Spilling）会将溢出变量存入本地显存，引入额外全局显存访问导致性能倒退54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/24426API
curl https://kongchang.com/api/v1/knowledge/claims/24426MCP
get_claim(id=24426)