Modelpublic

iapp/openthai2.0-qwen3.8-27b-NVFP4

sourceHugging Faceapache-2.0updated 26d agoView on Hugging Face

2likes3.6kdownloads

8 commits on main

4dd9a6c26d ago

v2.0.2 NVFP4

kobkrit

75b464b27d ago

v2.0.2 NVFP4

kobkrit

98b6e0128d ago

Restore MTP draft head (was missing: acceptance 1.00, 1.6x slower with speculative decoding). Adds model-mtp.safetensors, maps it in the index, ignores re:.*mtp.* in quantization_config. Verified: mean acceptance length 1.68 under vLLM qwen3_5_mtp.

kobkrit

ca3028928d ago

v2.0.1 NVFP4

kobkrit

2b941931mo ago

Re-quantized from release winner

kobkrit

960e3331mo ago

Quant card

kobkrit

b15d9fa1mo ago

NVFP4 (llmcompressor, ultrachat-calibrated; vision+MTP bf16)

kobkrit

5f82cdf1mo ago

initial commit

kobkrit