nvidia
/

GLM-5.3-Flash-NVFP4

Image-Text-to-Text

Model Optimizer

8-bit precision

Model card Files Files and versions

Resources

View closed (6)

MTP works after adding layer 45 to the quantization exclusion lists

#15 opened 1 day ago by

Why don't the commits in this repo have the "verified" badge?

#14 opened 1 day ago by

MTP layer is unquantized but missing from exclude_modules, so vLLM cannot load it for speculative decoding

#13 opened 1 day ago by

Error in function 'TllmGenFmhaRunner' at .sglang/lib/python3.12/site-packages/flashinfer/data/include/flashinfer/trtllm/fmha/fmhaRunner.cuh:37: Unsupported architecture

#12 opened 8 days ago by

fix: declare NextN (MTP) layer 45 as BF16 in quantization ignore lists

#11 opened 11 days ago by

Serving Recipe: GLM-5.3-Flash-NVFP4 with vLLM at 1M Context on Dual DGX Spark

#8 opened 16 days ago by

[FIX] Reasoning Loop on Trivial Task same as FP8

#7 opened 17 days ago by

ignore / exclude_modules omits the MTP layer (45): checkpoint fails to load with speculative decoding

#6 opened 18 days ago by

Where's sglang? Do you have a plan for adding Sglang support?

#2 opened 19 days ago by