unsloth
Qwen3 VL 2B Thinking
Qwen3 VL 2B Thinking is a 1.72B dense qwen3vl model from unsloth. 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.
Parameters1.7B
Context256K
Attentiongqa
Licenseapache-2.0
28 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 3.21 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 2.17 GB | Hugging Face file bytes |
| Q8_0 | 1.71 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 1.50 GB | Hugging Face file bytes |
| Q6_K | 1.32 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 1.18 GB | Hugging Face file bytes |
| Q5_K_M | 1.17 GB | Hugging Face file bytes |
| Q5_K_S | 1.15 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 1.06 GB | Hugging Face file bytes |
| Q4_K_M | 1.03 GB | Hugging Face file bytes |
| Q4_K_S | 0.99 GB | Hugging Face file bytes |
| IQ4_NL | 0.98 GB | Hugging Face file bytes |
| IQ4_XS | 0.94 GB | Hugging Face file bytes |
| Q4_1 | 1.06 GB | Hugging Face file bytes |
| Q4_0 | 0.98 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 0.90 GB | Hugging Face file bytes |
| Q3_K_M | 0.88 GB | Hugging Face file bytes |
| Q3_K_S | 0.81 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 0.71 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 0.74 GB | Hugging Face file bytes |
| Q2_K | 1.45 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 0.56 GB | Hugging Face file bytes |
| UD-IQ1_M | 0.52 GB | Hugging Face file bytes |
| UD-IQ1_S | 0.50 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Runs | BF16 | 36–49 |
| GeForce RTX 3060 12GB | Runs great | BF16 | 45–62 |
| GeForce RTX 4060 Ti 16GB | Runs great | BF16 | 36–49 |
| GeForce RTX 4090 | Runs great | BF16 | 126–172 |
| GeForce RTX 5090 | Runs great | BF16 | 225–306 |
| Apple M4 Pro | Runs great | BF16 | 34–47 |
| Apple M5 | Runs great | BF16 | 19–26 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 11–15 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Runs, BF16, 21–29 tok/s
- GeForce RTX 2060 6GB — Runs, BF16, 42–57 tok/s
- GeForce GTX 1660 Super — Runs, BF16, 42–57 tok/s
- GeForce GTX 1660 Ti — Runs, BF16, 36–49 tok/s
- Intel Arc A380 — Runs, BF16, 23–32 tok/s
- GeForce RTX 5060 Ti 8GB — Runs, BF16, 56–77 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (BF16).
llama.cpp
llama-server -hf unsloth/Qwen3-VL-2B-Thinking-GGUF:BF16 -c 8192 -ngl 28
Ollama
ollama run qwen3
LM Studio
lms get unsloth/Qwen3-VL-2B-Thinking-GGUF # Then load quant BF16