DeepSeek GRM 27B needs about 20.2 GB of memory at Q4_K_M (a 16.6 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 24 quants, from 0.0 GB to 22.3 GB (Q6_K).
DeepSeek GRM 27B – a 27.2B‑parameter generative reward model
DeepSeek GRM 27B is a 27.2B‑parameter generative reward model based on Gemma‑2‑27B. It evaluates query‑response pairs by writing a principle, generating a critique, and scoring from 1 to 10. The model supports a 8192‑token context window and tool usage.
Good at
- Generating detailed critiques
- Producing transparent 1‑10 scores
- Supporting tool calls
Limits
- No vision support
- Requires 8192‑token context
- License is gemma
Best for
- Evaluating LLM responses
- Offline data collection
- Online RL reward generation
Who it’s for. Researchers and developers needing a local reward model for LLM evaluation. The model is suitable for users who can run GGUF files with llama.cpp and have sufficient VRAM for the chosen quant.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 3.6 GB | Fits | Get file → | |
| IQ1_S | 6.1 GB | 9.7 GB | Too big | Get file → | |
| IQ1_M | 6.7 GB | 10.3 GB | Too big | Get file → | |
| IQ2_XXS | 7.6 GB | 11.2 GB | Too big | Get file → | |
| IQ2_XS | 8.4 GB | 12.0 GB | Too big | Get file → | |
| IQ2_S | 8.7 GB | 12.2 GB | Too big | Get file → | |
| IQ2_M | 9.4 GB | 13.0 GB | Too big | Get file → | |
| Q2_K_S | 9.7 GB | 13.3 GB | Too big | Get file → | |
| Q2_K | 10.4 GB | 14.0 GB | Too big | Get file → | |
| IQ3_XXS | 10.8 GB | 14.3 GB | Too big | Get file → | |
| IQ3_XS | 11.6 GB | 15.1 GB | Too big | Get file → | |
| IQ3_S | 12.2 GB | 15.8 GB | Too big | Get file → | |
| Q3_K_S | 12.2 GB | 15.8 GB | Too big | Get file → | |
| IQ3_M | 12.5 GB | 16.0 GB | Too big | Get file → | |
| Q3_K_M | 13.4 GB | 17.0 GB | Too big | Get file → | |
| Q3_K_L | 14.5 GB | 18.1 GB | Too big | Get file → | |
| IQ4_XS | 14.8 GB | 18.4 GB | Too big | Get file → | |
| Q4_0 | 15.7 GB | 19.3 GB | Too big | Get file → | |
| Q4_K_S | 15.7 GB | 19.3 GB | Too big | Get file → | |
| Q4_K_M | 16.6 GB | 20.2 GB | Too big | Get file → | |
| Q4_1 | 17.3 GB | 20.8 GB | Too big | Get file → | |
| Q5_K_S | 18.9 GB | 22.5 GB | Too big | Get file → | |
| Q5_K_M | 19.4 GB | 23.0 GB | Too big | Get file → | |
| Q6_K | 22.3 GB | 25.9 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (6.13 GB, 9.7 GB memory at 8k context). For 12 GB VRAM, Q4_K_M (16.65 GB, 20.2 GB memory) is a good balance. Larger VRAM allows Q5_K_M or Q6_K for higher precision.
What GPU do you need for DeepSeek GRM 27B?
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: the best quant that fits is the main file (0.0 GB file, about 3.6 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 3.6 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 3.6 GB in use).
- 12 GB VRAM: the best quant that fits is IQ2_XS (8.4 GB file, about 12.0 GB in use).
- 16 GB VRAM: the best quant that fits is IQ3_M (12.5 GB file, about 16.0 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_K_M (19.4 GB file, about 23.0 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K (22.3 GB file, about 25.9 GB in use).
Which DeepSeek GRM 27B quant should you choose?
Q4_K_M is the usual starting point for DeepSeek GRM 27B: it balances file size and quality well. Here it is a 16.6 GB file.
If memory is tight, the smallest file is the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about DeepSeek GRM 27B
How much VRAM does DeepSeek GRM 27B need?
About 20.2 GB at Q4_K_M with an 8K context window. That is the 16.6 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 DeepSeek GRM 27B on an 8 GB GPU?
Yes, with the smaller files: the 0.0 GB file needs about 3.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of DeepSeek GRM 27B GGUF?
It depends on the quant: file is 0.0 GB, IQ1_S is 6.1 GB, IQ1_M is 6.7 GB, IQ2_XXS is 7.6 GB, IQ2_XS is 8.4 GB, IQ2_S is 8.7 GB and 18 more.
What context length does DeepSeek GRM 27B support?
The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 3.1 GB at 8K and 3.1 GB at 8K.
Where can I download DeepSeek GRM 27B GGUF?
The files are on Hugging Face in the mradermacher/DeepSeek-GRM-27B-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 DeepSeek GRM 27B use?
Hugging Face lists the license as gemma. 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 DeepSeek GRM 27B on 8 GB VRAM?
Yes, using IQ1_S quant, which is 6.13 GB and requires about 9.7 GB memory at 8k context.
Does the model support image inputs?
No, the model has no vision capabilities.
What is the context length of this model?
The model supports 8192 tokens of context.