unsloth
Qwen3 Coder 30B A3B Instruct
Qwen3 Coder 30B A3B Instruct is a 30.5B mixture-of-experts qwen3moe model from unsloth, with about 3.0B parameters active per token. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters30.5B / 3.0B active
Context256K
Attentiongqa
Licenseapache-2.0
48 layers · 4 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 56.9 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 33.5 GB | Hugging Face file bytes |
| Q8_0 | 30.3 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 24.5 GB | Hugging Face file bytes |
| Q6_K | 23.4 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 20.2 GB | Hugging Face file bytes |
| Q5_K_M | 20.2 GB | Hugging Face file bytes |
| Q5_K_S | 19.6 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 16.5 GB | Hugging Face file bytes |
| Q4_K_M | 17.3 GB | Hugging Face file bytes |
| Q4_K_S | 16.3 GB | Hugging Face file bytes |
| IQ4_NL | 16.1 GB | Hugging Face file bytes |
| IQ4_XS | 15.3 GB | Hugging Face file bytes |
| Q4_1 | 17.9 GB | Hugging Face file bytes |
| Q4_0 | 16.2 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 12.9 GB | Hugging Face file bytes |
| Q3_K_M | 13.7 GB | Hugging Face file bytes |
| Q3_K_S | 12.4 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 12.0 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 11.0 GB | Hugging Face file bytes |
| Q2_K | 21.0 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 9.62 GB | Hugging Face file bytes |
| UD-IQ1_M | 8.97 GB | Hugging Face file bytes |
| UD-IQ1_S | 8.30 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Expert offload | Q4_K_M | 60–82 |
| GeForce RTX 3060 12GB | Tight fit | UD-IQ2_XXS | 140–190 |
| GeForce RTX 4060 Ti 16GB | Tight fit | UD-Q3_K_XL | 73–100 |
| GeForce RTX 4090 | Tight fit | UD-Q5_K_XL | 188–257 |
| GeForce RTX 5090 | Tight fit | UD-Q6_K_XL | 290–396 |
| Apple M4 Pro | Tight fit | UD-Q4_K_XL | 70–96 |
| Apple M5 | Tight fit | UD-IQ2_XXS | 46–63 |
| CPU only · DDR5-5600 | Runs on CPU | Q6_K | 15–21 |
Smallest devices that fit it fully
- GeForce RTX 3080 10GB — Tight fit, UD-IQ1_S, 428–584 tok/s
- Intel Arc B570 — Tight fit, UD-IQ1_S, 214–292 tok/s
- GeForce RTX 2080 Ti — Tight fit, UD-IQ1_M, 251–343 tok/s
- GeForce GTX 1080 Ti — Tight fit, UD-IQ1_M, 197–269 tok/s
- GeForce RTX 5070 — Tight fit, UD-IQ2_XXS, 261–355 tok/s
- GeForce RTX 4070 Ti — Tight fit, UD-IQ2_XXS, 195–267 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-Q3_K_XL).
llama.cpp
llama-server -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:UD-Q3_K_XL -c 8192 -ngl 48
Ollama
ollama run qwen3
LM Studio
lms get unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF # Then load quant UD-Q3_K_XL