community pool / model answer

LFM2.5-8B-A1B

chat · LiquidAI/LFM2.5-8B-A1B

189.3 tok/s · model median · 35 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
AMD Ryzen 7 8745H w/ Radeon 780M Graphics	4-bit	63.1 tok/s	reported	521	- GB	65,536	llama.cpp
GTX 1060 3GB	4-bit	8.5 tok/s	reported	33,535	- GB	2,048	llama.cpp
M5 Pro 48GB	4-bit	189.3 tok/s	reported	134	4.8 GB	2,048	mlx
Radeon AI Pro R9700 34GB	4-bit	241.1 tok/s	reported	194	5.4 GB	4,096	hipfire
RTX 3060 12GB	4-bit	194.4 tok/s	measured	323	6.9 GB	131,072	llama.cpp
RTX 3060 12GB	8-bit	128.5 tok/s	reported	37	- GB	4,096	llama.cpp
RTX 5060 Ti 16GB	4-bit	254.3 tok/s	reported	22	- GB	4,096	llama.cpp
RTX 5060 Ti 16GB	8-bit	183.1 tok/s	reported	25	- GB	4,096	llama.cpp
RTX 5070 Ti 16GB	4-bit	402.3 tok/s	reported	16	- GB	4,096	llama.cpp
RTX 5070 Ti 16GB	8-bit	307.4 tok/s	reported	20	- GB	4,096	llama.cpp
RX 570 4GB ×2	4-bit	18.6 tok/s	measured	325	- GB	2,048	llama.cpp
RX 6700 10GB	4-bit	172.3 tok/s	reported	2,151	4 GB	4,096	llama.cpp

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.