unsloth
Llama 3.2 3B Instruct
Llama 3.2 3B Instruct is a 3.21B dense llama model from unsloth. Configured context length is 131,072 tokens. License: llama3.2. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters3.2B
Context128K
Attentiongqa
Licensellama3.2
28 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 5.99 GB | Hugging Face file bytes |
| F16 | 5.99 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 3.92 GB | Hugging Face file bytes |
| Q8_0 | 3.19 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 2.76 GB | Hugging Face file bytes |
| Q6_K | 2.46 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 2.17 GB | Hugging Face file bytes |
| Q5_K_M | 2.16 GB | Hugging Face file bytes |
| Q5_K_S | 2.11 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 1.92 GB | Hugging Face file bytes |
| Q4_K_M | 1.88 GB | Hugging Face file bytes |
| Q4_K_S | 1.80 GB | Hugging Face file bytes |
| IQ4_NL | 1.79 GB | Hugging Face file bytes |
| IQ4_XS | 1.70 GB | Hugging Face file bytes |
| Q4_1 | 1.95 GB | Hugging Face file bytes |
| Q4_0 | 1.79 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 1.62 GB | Hugging Face file bytes |
| Q3_K_M | 1.57 GB | Hugging Face file bytes |
| Q3_K_S | 1.44 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 1.28 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 1.31 GB | Hugging Face file bytes |
| Q2_K | 2.54 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 0.97 GB | Hugging Face file bytes |
| UD-IQ1_M | 0.89 GB | Hugging Face file bytes |
| UD-IQ1_S | 0.85 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Tight fit | BF16 | 23–31 |
| GeForce RTX 3060 12GB | Runs | BF16 | 27–37 |
| GeForce RTX 4060 Ti 16GB | Runs great | BF16 | 21–29 |
| GeForce RTX 4090 | Runs great | BF16 | 75–103 |
| GeForce RTX 5090 | Runs great | BF16 | 134–182 |
| Apple M4 Pro | Runs great | BF16 | 20–28 |
| Apple M5 | Runs | BF16 | 11–16 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 6.7–9.1 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, UD-Q8_K_XL, 20–27 tok/s
- GeForce RTX 2060 6GB — Tight fit, UD-Q8_K_XL, 40–54 tok/s
- GeForce GTX 1660 Super — Tight fit, UD-Q8_K_XL, 40–54 tok/s
- GeForce GTX 1660 Ti — Tight fit, UD-Q8_K_XL, 34–46 tok/s
- Intel Arc A380 — Tight fit, UD-Q8_K_XL, 22–30 tok/s
- GeForce RTX 5060 Ti 8GB — Tight fit, BF16, 36–49 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (BF16).
llama.cpp
llama-server -hf unsloth/Llama-3.2-3B-Instruct-GGUF:BF16 -c 8192 -ngl 28
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/Llama-3.2-3B-Instruct-GGUF # Then load quant BF16