Unverified50% confidenceFactExact time
DeepInfer uses SWA bounded replay technology to reduce the last 2,560 tokens recomputed when rebuilding the cache to 128, cutting total prefill computation nearly in half and decoupling long-context cost from length.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
Related Entities
Related Claims
UnverifiedSome of DeepSeek's model layers use sliding window attention, attending only to the most recent 128 tokens; SWA bounded replay replays only a short window at the end of the prompt, which is said to nearly eliminate total prefill work.72% similarUnverifiedIn DeepSeek-V4 optimizations, skipping empty c128 launches achieves approximately 2x operator improvement, and removing redundant full operators achieves 1.88x operator improvement.66% similarUnverifiedDeepSeek's first-generation model consumed 389,000 bytes per context token, whereas the new model released this month consumes only 890 bytes per token, a 437-fold reduction.64% similarUnverifiedDeepInfer's core mechanism is context caching, which keeps already-computed mathematical paths in memory, with a cache validity period exceeding 72 hours.64% similarVerifiedDeepSeek released DeepSeek-V4.1-Flash, a method that significantly compresses the memory required for KV cache.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088587API
curl https://kongchang.com/api/v1/knowledge/claims/1088587MCP
get_claim(id=1088587)