Configuration Parsing Warning:Config file config.json cannot be fetched (too big)
MiMo V2.5 Pro NVFP4 + MXFP8 Attention TP8
Experimental checkpoint produced from Xiaomi MiMo V2.5 Pro for SGLang/B12X testing.
- Source serving path used locally:
/data/models/MiMo-V2.5-Pro-NVFP4-MXFP8-attn - Base weights: NVFP4
- Attention/QKV path: MXFP8/FP8, deinterleaved for TP=8
- Includes
model-mtp.safetensors - Current known serving baseline: TP=8, non-MTP, SGLang B12X,
--quantization modelopt_mixed - MTP/EAGLE is still under investigation and was not treated as validated at upload time.
Uploaded at 2026-05-09T21:04:14.398244+00:00.
- Downloads last month
- 70
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support