Seed OSS 36B Instruct MPOA v1 needs about 24.4 GB of memory at Q4_K_M (a 21.8 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 24 quants, from 0.0 GB to 29.7 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
Seed OSS 36B Instruct MPOA v1 – 36.2B parameter causal language model
Seed OSS 36B Instruct MPOA v1 is a 36.2B parameter causal language model. It is an abliterated variant of YanLabs/Seed-OSS-36B-Instruct-MPOA, removing safety guardrails for research purposes. The model is licensed under apache-2.0 and supports up to 524288 context tokens.
Good at
- Mechanistic interpretability research
- Analyzing LLM safety mechanisms
- Testing abliteration techniques
Limits
- No safety guardrails – may generate harmful content
- Not suitable for production or user‑facing applications
- Requires large VRAM for full precision
Best for
- Research on refusal behavior removal
- Studying model activation space
- Developing new abliteration methods
Who it’s for. Researchers and developers working on LLM interpretability and safety studies. The model is intended for controlled environments and should not be deployed in production.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 2.7 GB | Fits | Get file → | |
| IQ1_S | 8.1 GB | 10.7 GB | Too big | Get file → | |
| IQ1_M | 8.7 GB | 11.4 GB | Too big | Get file → | |
| IQ2_XXS | 9.9 GB | 12.6 GB | Too big | Get file → | |
| IQ2_XS | 10.9 GB | 13.6 GB | Too big | Get file → | |
| IQ2_S | 11.6 GB | 14.3 GB | Too big | Get file → | |
| IQ2_M | 12.5 GB | 15.2 GB | Too big | Get file → | |
| Q2_K_S | 12.6 GB | 15.3 GB | Too big | Get file → | |
| Q2_K | 13.6 GB | 16.3 GB | Too big | Get file → | |
| IQ3_XXS | 14.1 GB | 16.8 GB | Too big | Get file → | |
| IQ3_XS | 15.1 GB | 17.7 GB | Too big | Get file → | |
| Q3_K_S | 15.9 GB | 18.5 GB | Too big | Get file → | |
| IQ3_S | 15.9 GB | 18.5 GB | Too big | Get file → | |
| IQ3_M | 16.5 GB | 19.1 GB | Too big | Get file → | |
| Q3_K_M | 17.6 GB | 20.3 GB | Too big | Get file → | |
| Q3_K_L | 19.1 GB | 21.8 GB | Too big | Get file → | |
| IQ4_XS | 19.5 GB | 22.1 GB | Too big | Get file → | |
| Q4_0 | 20.6 GB | 23.3 GB | Too big | Get file → | |
| Q4_K_S | 20.7 GB | 23.3 GB | Too big | Get file → | |
| Q4_K_M | 21.8 GB | 24.4 GB | Too big | Get file → | |
| Q4_1 | 22.8 GB | 25.4 GB | Too big | Get file → | |
| Q5_K_S | 25.0 GB | 27.6 GB | Too big | Get file → | |
| Q5_K_M | 25.6 GB | 28.2 GB | Too big | Get file → | |
| Q6_K | 29.7 GB | 32.3 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_M (21.76 GB file, 24.4 GB memory at 8k context). For 16 GB VRAM, Q5_K_M (25.59 GB file, 28.2 GB memory) or Q6_K (29.67 GB file, 32.3 GB memory). Smaller quants such as IQ1_S or IQ1_M require 10–11 GB memory and are suitable for 12–16 GB VRAM setups.
What GPU do you need for Seed OSS 36B Instruct MPOA v1?
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 2.7 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 2.7 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 2.7 GB in use).
- 12 GB VRAM: the best quant that fits is IQ1_M (8.7 GB file, about 11.4 GB in use).
- 16 GB VRAM: the best quant that fits is Q2_K_S (12.6 GB file, about 15.3 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_K_S (20.7 GB file, about 23.3 GB in use).
- 32 GB VRAM: the best quant that fits is Q5_K_M (25.6 GB file, about 28.2 GB in use).
- 48 GB VRAM: the best quant that fits is Q6_K (29.7 GB file, about 32.3 GB in use).
Which Seed OSS 36B Instruct MPOA v1 quant should you choose?
Q4_K_M is the usual starting point for Seed OSS 36B Instruct MPOA v1: it balances file size and quality well. Here it is a 21.8 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 Seed OSS 36B Instruct MPOA v1
How much VRAM does Seed OSS 36B Instruct MPOA v1 need?
About 24.4 GB at Q4_K_M with an 8K context window. That is the 21.8 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 Seed OSS 36B Instruct MPOA v1 on an 8 GB GPU?
Yes, with the smaller files: the 0.0 GB file needs about 2.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Seed OSS 36B Instruct MPOA v1 GGUF?
It depends on the quant: file is 0.0 GB, IQ1_S is 8.1 GB, IQ1_M is 8.7 GB, IQ2_XXS is 9.9 GB, IQ2_XS is 10.9 GB, IQ2_S is 11.6 GB and 18 more.
What context length does Seed OSS 36B Instruct MPOA v1 support?
The GGUF header lists a trained context length of 524,288 tokens (512K). 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 Seed OSS 36B Instruct MPOA v1 GGUF?
The files are on Hugging Face in the mradermacher/Seed-OSS-36B-Instruct-MPOA-v1-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 Seed OSS 36B Instruct MPOA v1 use?
Hugging Face lists the license as apache-2.0. 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 this model on 8 GB VRAM?
The Q4_K_M quant requires 24.4 GB memory at 8k context, so 8 GB VRAM is insufficient. Use a smaller quant or increase VRAM.
Does the model support vision tasks?
No, the model is a purely causal language model with no vision capabilities.
Is the model safe for public deployment?
No. Abliteration removes safety guardrails, so the model may generate harmful content and is intended only for research.