GP

Gpt Oss 20b Offload

gpt-ossGGUF by mradermacherUpdated 2025-12-04
Download on Hugging Face
Parameters20.9B
Context128K
Layers24
Architecturegpt-oss
Smallest quant0.0 GB
Quick answer

Gpt Oss 20b Offload needs about 16.7 GB of memory at Q4_K_M (a 15.8 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 22.2 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 1.6 GB.

GPT‑OSS‑20B Offload, 20.9B‑parameter Mixture‑of‑Experts model with CPU+GPU offload

GPT‑OSS‑20B Offload is a 20.9B‑parameter decoder‑only transformer from OpenAI, released in 2025. It supports 4 active experts per token out of 32 total. The model is licensed MIT and can be run with CPU+GPU offload for low‑VRAM environments.

Good at

  • Text generation
  • Summarization
  • Question answering

Limits

  • Requires 131072 context tokens
  • No vision support
  • Mixture‑of‑Experts limits memory usage

Best for

  • Low‑VRAM inference
  • Fine‑tuning for domain assistants
  • Chatbot integration

Who it’s for. Developers and researchers needing a large language model that can run in GPU‑constrained settings. The model is suitable for building chatbots, summarizers, or other generative applications where CPU+GPU offload is acceptable.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—0.0 GB 0.9 GB
Fits Get file →
IQ1_M12.0 GB 12.9 GB
Too big Get file →
IQ1_S12.0 GB 12.9 GB
Too big Get file →
IQ2_XXS12.0 GB 12.9 GB
Too big Get file →
IQ2_XS12.0 GB 12.9 GB
Too big Get file →
Q3_K_S12.1 GB 13.0 GB
Too big Get file →
IQ2_M12.1 GB 13.0 GB
Too big Get file →
IQ2_S12.1 GB 13.0 GB
Too big Get file →
IQ3_S12.1 GB 13.0 GB
Too big Get file →
IQ3_XS12.1 GB 13.0 GB
Too big Get file →
IQ3_XXS12.1 GB 13.0 GB
Too big Get file →
Q2_K12.1 GB 13.0 GB
Too big Get file →
IQ4_XS12.1 GB 13.0 GB
Too big Get file →
Q2_K_S12.1 GB 13.0 GB
Too big Get file →
Q4_012.1 GB 13.1 GB
Too big Get file →
IQ3_M12.2 GB 13.1 GB
Too big Get file →
Q3_K_M12.9 GB 13.8 GB
Too big Get file →
Q3_K_L13.3 GB 14.2 GB
Too big Get file →
Q4_113.4 GB 14.3 GB
Too big Get file →
Q4_K_S14.7 GB 15.6 GB
Too big Get file →
Q4_K_M15.8 GB 16.7 GB
Too big Get file →
Q5_K_S15.9 GB 16.8 GB
Too big Get file →
Q5_K_M16.9 GB 17.8 GB
Too big Get file →
Q6_K22.2 GB 23.1 GB
Too big Get file →

For 8 GB VRAM use IQ1_M or IQ1_S (12.02 GB file, ~12.9 GB memory at 8 k context). For 16 GB VRAM use Q3_K_S (12.06 GB file, ~13 GB memory). For 24 GB VRAM use Q4_K_M (15.81 GB file, ~16.7 GB memory). Estimates are approximate.

What GPU do you need for Gpt Oss 20b Offload?

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 0.9 GB in use).
  • 6 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 0.9 GB in use).
  • 8 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 0.9 GB in use).
  • 12 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 0.9 GB in use).
  • 16 GB VRAM: the best quant that fits is Q4_K_S (14.7 GB file, about 15.6 GB in use).
  • 24 GB VRAM: the best quant that fits is Q6_K (22.2 GB file, about 23.1 GB in use).

Which Gpt Oss 20b Offload quant should you choose?

Q4_K_M is the usual starting point for Gpt Oss 20b Offload: it balances file size and quality well. Here it is a 15.8 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

Architecturegpt-oss
Parameters20.9B (approximate, mixture of experts)
Layers24
Trained context131,072 tokens
KV cache (fp16)48 KB per token

Source

Quantized bymradermacher
Familygpt-oss
Licensemit
Downloads (HF)695
Last updated2025-12-04

Frequently asked questions about Gpt Oss 20b Offload

How much VRAM does Gpt Oss 20b Offload need?

About 16.7 GB at Q4_K_M with an 8K context window. That is the 15.8 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 Gpt Oss 20b Offload on an 8 GB GPU?

Yes, with the smaller files: the 0.0 GB file needs about 0.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Gpt Oss 20b Offload GGUF?

It depends on the quant: file is 0.0 GB, IQ1_M is 12.0 GB, IQ1_S is 12.0 GB, IQ2_XXS is 12.0 GB, IQ2_XS is 12.0 GB, Q3_K_S is 12.1 GB and 18 more.

What context length does Gpt Oss 20b Offload support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.4 GB at 8K and 1.6 GB at 32K.

Where can I download Gpt Oss 20b Offload GGUF?

The files are on Hugging Face in the mradermacher/gpt-oss-20b-offload-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 Gpt Oss 20b Offload 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 this model on 8 GB VRAM?

Yes, use IQ1_M or IQ1_S quant, which are 12.02 GB files and need about 12.9 GB memory at 8 k context.

What does Mixture‑of‑Experts mean for inference?

Each token activates 4 experts out of 32, reducing memory per token but requiring more compute. Offload helps manage memory usage.

How do I enable CPU+GPU offload?

Load the model with device_map="auto" and set max_memory for GPU and CPU in your framework, as shown in the provided code snippet.

Other GGUFs of gpt-oss-20b-offload

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More gpt-oss models

All gpt-oss →

Other medium models

All medium models →