Unverified50% confidenceFactExact time
Continuous batching counts rows by token position rather than by request count; in one batch, the 120 prefill tokens of a new request and the 4 verification tokens of an existing request were merged into 124 rows, thereby crossing the threshold and selecting GEMM.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring Dispatch queuing, the number of tokens in each expert queue is padded to a multiple of 128 (expert alignment), because the subsequent grouped GEMM only accepts inputs aligned to 128.62% similarUnverifiedMulti-query retrieval uses an LLM to automatically generate 3–5 semantically equivalent but differently phrased variant queries for the same question, retrieves results for each separately, and takes the union to improve recall.60% similarUnverifiedCommon methods for achieving idempotency include assigning a unique request ID (idempotency key) to each write operation, or using conditional writes (compare-and-swap) that only execute when the target data is in the expected state.58% similarUnverifiedEach query first goes through validation: only SELECT or WITH statements are allowed, only whitelisted tables are permitted, a maximum of 100 MB is billed, and a maximum of 200 rows are returned.58% similarUnverifiedThe agent received as input purely numerical coordinate data for 80 boundary configurations, each determined by six paired lines, with the target theorem entirely hidden from the task instructions.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1000273API
curl https://kongchang.com/api/v1/knowledge/claims/1000273MCP
get_claim(id=1000273)