3 ms·
Native MCP: For Qwen 35B enabling native MCP on MLX models slows it down by 10%. For Qwen 27B enabling native MCP on MLX models speeds token generation up alm
by egorfine 5mo ago
Native MCP:
For Qwen 35B enabling native MCP on MLX models slows it down by 10%.
For Qwen 27B enabling native MCP on MLX models speeds token generation up almost exactly 1.5x.
(all tested on M5 pro).
- mlvljr 5mo ago[dead]