Qwen3-VL-32B-Instruct-NVFP4

Qwen3-VL-32B-Instruct-NVFP4 (single-file)

This is a single-file repack of RedHatAI/Qwen3-VL-32B-Instruct-NVFP4, which is itself an NVFP4 quantization of Qwen/Qwen3-VL-32B-Instruct.

The original repo ships the model as 5 sharded .safetensors files plus a model.safetensors.index.json. This repo merges those shards into one Qwen3-VL-32B-Instruct-NVFP4.safetensors file for convenience with tools that expect a single-file checkpoint. No weights were modified — tensors were concatenated byte-for-byte from the original shards into one file with a combined header; values, shapes, and dtypes are identical to the source.

This repo intentionally contains only the merged weights file — no config.json, tokenizer, or processor files. Pull those from the original RedHatAI/Qwen3-VL-32B-Instruct-NVFP4 repo if you need to load this as a full transformers model.

Credits

• Base model: Qwen/Qwen3-VL-32B-Instruct by the Qwen team
• Quantization (NVFP4): RedHatAI/Qwen3-VL-32B-Instruct-NVFP4 by Red Hat, produced with llm-compressor
• This repo: shard merge only, no retraining or requantization

License: apache-2.0 (inherited from the base and quantized models).