Unverified50% confidenceBenchmarkExact time
In the evict_read control experiment, the speed was 3405 tok/s when the weights resided in L3, and dropped to 1681 tok/s when L3 was polluted by random reads, a penalty factor of 2.03x.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedA third-party same-VRAM comparison on Zhihu shows that Stratavie writes responses at 62.9 tokens/second, while Llama.cpp achieves 23.4 tokens/second, with long-document reading being nearly 10 times faster.62% similarUnverifiedIn a model comparison test, qwen2.5:3b correctly understood pace, produced 0 hallucinated place names, had a length error of 0.3%, peak memory of 259 MB, and took 106 seconds.59% similarUnverifiedIn the InferenceX public benchmark, Jalapeño achieved 1.5 to 1.9 times the tasks per watt at peak throughput compared to NVIDIA's reference system, with end-to-end latency reduced by 1.7 to 3.6 times.59% similarUnverified以 GPT-3 为例，r=4 时 LoRA 可将可训练参数从 175B 降至约 4.7M，减少约 37,000 倍58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1006037API
curl https://kongchang.com/api/v1/knowledge/claims/1006037MCP
get_claim(id=1006037)