Expired50% confidenceBenchmarkExact time
On an AMD W7900 GPU using llama.cpp with the ROCm backend, the Q8 quantized version of LFM2.5-24B-A2B sustains an inference speed of about 95 tokens/second.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/15/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/29/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedOn 8 AMD Instinct MI355X accelerator cards, using the vLLM inference framework, the GLM-5.3 model achieved a single-user decoding speed of 469 tok/s.69% similarUnverifiedAMD's RX 9700 series GPUs have hardware-level acceleration support for mxfp4, and upgrading to 128GB of memory and using the mxfp4 quantized version of Flash Next can increase QFN speed by approximately 2 times.66% similarUnverifiedThe AMD platform enables the Vulkan backend by default, achieving a 20% performance improvement over ROCm in both the prefill and decoding stages.66% similarUnverified索尼WH-1000XM5搭载自研V1处理器与QN1降噪芯片，支持LDAC编解码（最高990kbps码率）及多设备同时连接63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038618API
curl https://kongchang.com/api/v1/knowledge/claims/1038618MCP
get_claim(id=1038618)