tianrui6641/tais_coder_mini_36b_a3b_core_generalist-merged-MXFP4

tais_coder_mini_36b_a3b_core_generalist-merged MXFP4

Source model: tianrui6641/tais_coder_mini_36b_a3b_core_generalist-merged

This repository contains a Hugging Face checkpoint generated by NVIDIA ModelOpt from a locally downloaded source checkpoint.

Notes

• Export format: MXFP4
• ModelOpt qformat: w4a8_mxfp4_fp8
• This artifact is not a GGUF file.
• Export attempt profile: default
• The original source checkpoint was downloaded to /content first and then reused for PTQ.
• The export directory is selected automatically: /dev/shm is used only when it has enough free space; otherwise /content is used.
• Deployment support depends on ModelOpt-compatible runtimes such as TensorRT-LLM, vLLM, or SGLang.
• The original source model remains subject to its original license and terms.
• This export uses the supported ModelOpt qformat w4a8_mxfp4_fp8.
• The retry flow first prefers /dev/shm when there is enough free space, then falls back to /content, then reduces MoE calibration coverage if needed.