tais_coder_mini_36b_a3b_core_generalist-merged NVFP4
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:
NVFP4
- ModelOpt qformat:
nvfp4
- 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.
- The retry flow first prefers
/dev/shm when there is enough free space, then falls back to /content, then enables ModelOpt low-memory mode if needed.
- NVFP4 export is intended for Blackwell-class deployment stacks that support ModelOpt checkpoints.
