unsloth
medgemma 4b it
medgemma 4b it is a 3.88B dense gemma3 model from unsloth. Configured context length is 131,072 tokens. License: other. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters3.9B
Context128K
Attentionhybrid
Licenseother
34 layers · 4 KV heads · head dim 256 · 28 sliding layers · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 7.23 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 4.81 GB | Hugging Face file bytes |
| Q8_0 | 3.85 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 3.32 GB | Hugging Face file bytes |
| Q6_K | 2.97 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 2.64 GB | Hugging Face file bytes |
| Q5_K_M | 2.64 GB | Hugging Face file bytes |
| Q5_K_S | 2.57 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 2.37 GB | Hugging Face file bytes |
| Q4_K_M | 2.32 GB | Hugging Face file bytes |
| Q4_K_S | 2.21 GB | Hugging Face file bytes |
| IQ4_NL | 2.20 GB | Hugging Face file bytes |
| IQ4_XS | 2.11 GB | Hugging Face file bytes |
| Q4_1 | 2.39 GB | Hugging Face file bytes |
| Q4_0 | 2.21 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 2.00 GB | Hugging Face file bytes |
| Q3_K_M | 1.95 GB | Hugging Face file bytes |
| Q3_K_S | 1.80 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 1.59 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 1.65 GB | Hugging Face file bytes |
| Q2_K | 3.22 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 1.25 GB | Hugging Face file bytes |
| UD-IQ1_M | 1.16 GB | Hugging Face file bytes |
| UD-IQ1_S | 1.10 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Runs | UD-Q8_K_XL | 25–34 |
| GeForce RTX 3060 12GB | Runs | BF16 | 22–30 |
| GeForce RTX 4060 Ti 16GB | Runs great | BF16 | 18–24 |
| GeForce RTX 4090 | Runs great | BF16 | 62–85 |
| GeForce RTX 5090 | Runs great | BF16 | 111–151 |
| Apple M4 Pro | Runs great | BF16 | 17–23 |
| Apple M5 | Runs | BF16 | 9.4–13 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 5.5–7.5 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, Q8_0, 18–24 tok/s
- GeForce RTX 2060 6GB — Tight fit, Q8_0, 35–48 tok/s
- GeForce GTX 1660 Super — Tight fit, Q8_0, 35–48 tok/s
- GeForce GTX 1660 Ti — Tight fit, Q8_0, 30–41 tok/s
- Intel Arc A380 — Tight fit, Q8_0, 19–26 tok/s
- GeForce RTX 5060 Ti 8GB — Runs, UD-Q8_K_XL, 39–53 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (BF16).
llama.cpp
llama-server -hf unsloth/medgemma-4b-it-GGUF:BF16 -c 8192 -ngl 34
Ollama
ollama run medgemma
LM Studio
lms get unsloth/medgemma-4b-it-GGUF # Then load quant BF16