unsloth
gpt oss 20b
gpt oss 20b is a 20.9B mixture-of-experts gpt-oss model from unsloth, with about 5.8B parameters active per token. Configured context length is 131,072 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters20.9B / 5.8B active
Context128K
Attentionhybrid
Licenseapache-2.0
24 layers · 8 KV heads · head dim 64 · 12 sliding layers · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| F16 | 12.8 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 12.3 GB | Hugging Face file bytes |
| Q8_0 | 11.3 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 11.2 GB | Hugging Face file bytes |
| Q6_K | 11.2 GB | Hugging Face file bytes |
| Q5_K_M | 10.9 GB | Hugging Face file bytes |
| Q5_K_S | 10.9 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 11.1 GB | Hugging Face file bytes |
| Q4_K_M | 10.8 GB | Hugging Face file bytes |
| Q4_K_S | 10.8 GB | Hugging Face file bytes |
| Q4_1 | 10.8 GB | Hugging Face file bytes |
| Q4_0 | 10.7 GB | Hugging Face file bytes |
| Q3_K_M | 10.7 GB | Hugging Face file bytes |
| Q3_K_S | 10.7 GB | Hugging Face file bytes |
| Q2_K | 21.6 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Expert offload | Q4_K_M | 47–64 |
| GeForce RTX 3060 12GB | Expert offload | Q4_K_M | 58–79 |
| GeForce RTX 4060 Ti 16GB | Tight fit | UD-Q8_K_XL | 39–54 |
| GeForce RTX 4090 | Runs great | UD-Q8_K_XL | 138–188 |
| GeForce RTX 5090 | Runs great | UD-Q8_K_XL | 244–333 |
| Apple M4 Pro | Runs | UD-Q8_K_XL | 37–51 |
| Apple M5 | Won't fit | Q3_K_S | — |
| CPU only · DDR5-5600 | Runs on CPU | UD-Q8_K_XL | 12–17 |
Smallest devices that fit it fully
- GeForce RTX 5080 — Tight fit, UD-Q8_K_XL, 131–179 tok/s
- GeForce RTX 5070 Ti — Tight fit, UD-Q8_K_XL, 122–167 tok/s
- GeForce RTX 5060 Ti 16GB — Tight fit, UD-Q8_K_XL, 61–83 tok/s
- GeForce RTX 4080 Super — Tight fit, UD-Q8_K_XL, 100–137 tok/s
- GeForce RTX 4080 — Tight fit, UD-Q8_K_XL, 98–133 tok/s
- GeForce RTX 4070 Ti Super — Tight fit, UD-Q8_K_XL, 92–125 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-Q8_K_XL).
llama.cpp
llama-server -hf unsloth/gpt-oss-20b-GGUF:UD-Q8_K_XL -c 8192 -ngl 24
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/gpt-oss-20b-GGUF # Then load quant UD-Q8_K_XL