Unverified50% confidenceFactExact time
Recursive used a model ladder starting at about 400 million parameters to validate architectural changes; the hyper-connections variant worked well at small scale but failed completely above 4 billion parameters, so it was abandoned.
1
Sources
50%
Confidence
Long-term
Relevance
11/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedReAct框架在任务链超过5-7个工具调用节点时存在明显的错误累积效应,每个节点的输出误差会在后续节点中被放大65% similarUnverifiedPreviously, modifying reasoning-related parameters (such as chain-of-thought length or reasoning mode toggles) would invalidate the cache key, causing shared context to be fully recomputed.63% similarUnverifiedv2.1.286 refactors the retry logic for API request failures, with one limit covering the entire model call; under default settings, a single failed call sends at most 14 requests.63% similarUnverifiedIn continuous batching frameworks such as vLLM and SGLang, the decoding steps of multiple requests are packed into large batches, and the verification stage of speculative decoding competes for compute with other requests in the batch, causing relative gains to decline.63% similarUnverifiedThe gateway-fallbacks example verifies fallback policies locally: the fallback chain depth cannot exceed three levels, max_attempts must be between 1 and 5, the fallback chain cannot reference itself or contain duplicates, and every fallback model must appear in allowed_models.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1166493API
curl https://kongchang.com/api/v1/knowledge/claims/1166493MCP
get_claim(id=1166493)