From the model card
Fields below are copied from the tags and counters on the HuggingFace repository nvidia/Qwen3.6-35B-A3B-NVFP4 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- nvidia
- Pipeline tag
- text-generation
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Lineage
-
- base model Qwen/Qwen3.6-35B-A3B
- quantized from Qwen/Qwen3.6-35B-A3B
- Downloads (HF counter at last fetch)
- 10,172,652
- Likes (HF counter at last fetch)
- 587
- Model card
- https://huggingface.co/nvidia/Qwen3.6-35B-A3B-NVFP4
Use cases
- Low-latency inference on NVIDIA Hopper-class GPUs
- Serving large MoE models within tighter VRAM budgets
- Production deployment via TensorRT-LLM or vLLM
- Benchmarking FP4 vs BF16 accuracy trade-offs
- Azure AI Studio deployment with NVIDIA hardware
Pros
- Roughly half the VRAM footprint of the BF16 original
- ModelOpt-calibrated quantization preserves conversational quality
- Ready for TensorRT-LLM and vLLM inference engines
- Apache-2.0 licensed, permissive for commercial use
Cons
- Requires NVIDIA Hopper GPU (H100/H200); does not run on older architectures
- FP4 format is not supported in most personal or cloud notebook environments
- Cannot be used for fine-tuning — inference only
- Accuracy may degrade on tasks sensitive to low-bit numeric precision
Tags
Model Optimizersafetensorsqwen3_5_moenvidiaModelOptQwen3.6quantizedFP4fp4text-generationconversationalbase_model:Qwen/Qwen3.6-35B-A3Bbase_model:quantized:Qwen/Qwen3.6-35B-A3Blicense:apache-2.08-bitmodeloptdeploy:azuredeploy:sagemakerregion:us