Unverified50% confidenceSolutionExact time
Fixing max-model-len at 16384 halves the KV Cache upper bound; the original team's measured real prompt p99 was 11k tokens, so 16k both leaves headroom and avoids doubling the pool.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
VerifiedKV Cache大小约等于 2 × 层数 × 注意力头数 × 头维度 × 序列长度 × 精度字节数70% similarUnverifiedQwen2.5-Max-0902 has a maximum output of 130,000 tokens, with the limit rising to 260,000 tokens in reasoning mode.65% similarUnverifiedOn the held-out set, the 8k–16k vocabularies had the lowest bpb across all three model sizes, while the 2k vocabulary consistently performed worst.65% similarUnverifiedAfter identifying the matching 112-shape tuning table (16 shapes × 7 T-buckets) and enabling MMQ and Prefill chunking, Fresh Prefill for 4–9K jumped from 405 to 701–775 tokens/s, approximately a 1.8× improvement, with decode stabilizing at 52–55.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063643API
curl https://kongchang.com/api/v1/knowledge/claims/1063643MCP
get_claim(id=1063643)