UnifiedReward Qwen 32b needs about 22.5 GB of memory at Q4_K_M (a 19.9 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 23 quants, from 7.3 GB (IQ1_S) to 26.9 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
UnifiedReward Qwen 32b – a multimodal reward model based on Qwen2VL
UnifiedReward Qwen 32b is a 32.8 billion‑parameter vision‑enabled reward model. It supports pairwise ranking and pointwise scoring for multimodal content. The model is released under the MIT license and can be used with llama.cpp.
Good at
- Vision‑based reward scoring
- Pairwise ranking of multimodal outputs
- Pointwise preference assessment
Limits
- Requires 128000 context tokens
- No text‑only support
- Large VRAM needed for higher quants
Best for
- Evaluating image caption quality
- Aligning vision models to human preferences
- Benchmarking multimodal generation
Who it’s for. Developers building multimodal evaluation pipelines or researchers needing a vision reward model. The model is suitable for those who can allocate sufficient GPU memory and are comfortable with GGUF and llama.cpp.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 7.3 GB | 9.9 GB | Too big | Get file → | |
| IQ1_M | 7.9 GB | 10.6 GB | Too big | Get file → | |
| IQ2_XXS | 9.0 GB | 11.7 GB | Too big | Get file → | |
| IQ2_XS | 10.0 GB | 12.6 GB | Too big | Get file → | |
| IQ2_S | 10.4 GB | 13.0 GB | Too big | Get file → | |
| IQ2_M | 11.3 GB | 13.9 GB | Too big | Get file → | |
| Q2_K_S | 11.5 GB | 14.1 GB | Too big | Get file → | |
| Q2_K | 12.3 GB | 15.0 GB | Too big | Get file → | |
| IQ3_XXS | 12.8 GB | 15.5 GB | Too big | Get file → | |
| IQ3_XS | 13.7 GB | 16.4 GB | Too big | Get file → | |
| Q3_K_S | 14.4 GB | 17.0 GB | Too big | Get file → | |
| IQ3_S | 14.4 GB | 17.1 GB | Too big | Get file → | |
| IQ3_M | 14.8 GB | 17.5 GB | Too big | Get file → | |
| Q3_K_M | 15.9 GB | 18.6 GB | Too big | Get file → | |
| Q3_K_L | 17.2 GB | 19.9 GB | Too big | Get file → | |
| IQ4_XS | 17.7 GB | 20.3 GB | Too big | Get file → | |
| Q4_0 | 18.7 GB | 21.4 GB | Too big | Get file → | |
| Q4_K_S | 18.8 GB | 21.4 GB | Too big | Get file → | |
| Q4_K_M | 19.9 GB | 22.5 GB | Too big | Get file → | |
| Q4_1 | 20.6 GB | 23.3 GB | Too big | Get file → | |
| Q5_K_S | 22.6 GB | 25.3 GB | Too big | Get file → | |
| Q5_K_M | 23.3 GB | 25.9 GB | Too big | Get file → | |
| Q6_K | 26.9 GB | 29.5 GB | Too big | Get file → |
For 8 GB VRAM use IQ1_S (7.27 GB, 9.9 GB memory at 8k context). For 12 GB VRAM use IQ1_M (7.93 GB, 10.6 GB memory). For 16 GB VRAM use IQ2_XS (9.96 GB, 12.6 GB memory). Higher quants increase accuracy but require more memory.
What GPU do you need for UnifiedReward Qwen 32b?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: the best quant that fits is IQ2_XXS (9.0 GB file, about 11.7 GB in use).
- 16 GB VRAM: the best quant that fits is IQ3_XXS (12.8 GB file, about 15.5 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_1 (20.6 GB file, about 23.3 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K (26.9 GB file, about 29.5 GB in use).
Which UnifiedReward Qwen 32b quant should you choose?
Q4_K_M is the usual starting point for UnifiedReward Qwen 32b: it balances file size and quality well. Here it is a 19.9 GB file.
If memory is tight, the smallest file is IQ1_S at 7.3 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about UnifiedReward Qwen 32b
How much VRAM does UnifiedReward Qwen 32b need?
About 22.5 GB at Q4_K_M with an 8K context window. That is the 19.9 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run UnifiedReward Qwen 32b on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 9.9 GB, so you would need a larger GPU or CPU offloading.
What is the file size of UnifiedReward Qwen 32b GGUF?
It depends on the quant: IQ1_S is 7.3 GB, IQ1_M is 7.9 GB, IQ2_XXS is 9.0 GB, IQ2_XS is 10.0 GB, IQ2_S is 10.4 GB, IQ2_M is 11.3 GB and 17 more.
What context length does UnifiedReward Qwen 32b support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download UnifiedReward Qwen 32b GGUF?
The files are on Hugging Face in the mradermacher/UnifiedReward-qwen-32b-i1-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does UnifiedReward Qwen 32b use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run UnifiedReward Qwen 32b on 8 GB VRAM?
Use the IQ1_S quant, which is 7.27 GB and requires about 9.9 GB of VRAM at 8k context.
Does this model support text-only inference?
No, the model is vision‑enabled and requires image input for reward scoring.
What license applies to this model?
The model is released under the MIT license.