unsloth
DeepSeek V4 Flash Vision Exp
DeepSeek V4 Flash Vision Exp is a 284.3B mixture-of-experts deepseek4 model from unsloth, with about 49.3B parameters active per token. Configured context length is 1,048,576 tokens. License: mit. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters284B / 49.3B active
Context1024K
Attentiongqa
Licensemit
43 layers · 1 KV heads · head dim 512 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| UD-Q8_K_XL | 150.8 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 144.4 GB | Hugging Face file bytes |
| UD-IQ4_XS | 127.3 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 119.4 GB | Hugging Face file bytes |
| UD-IQ3_S | 106.5 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 95.9 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 90.2 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 84.5 GB | Hugging Face file bytes |
| UD-IQ1_M | 80.9 GB | Hugging Face file bytes |
| UD-IQ1_S | 76.8 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Won't fit | UD-IQ1_S | — |
| GeForce RTX 3060 12GB | Won't fit | UD-IQ1_S | — |
| GeForce RTX 4060 Ti 16GB | Won't fit | UD-IQ1_S | — |
| GeForce RTX 4090 | Won't fit | UD-IQ1_S | — |
| GeForce RTX 5090 | Won't fit | UD-IQ1_S | — |
| Apple M4 Pro | Won't fit | UD-IQ1_S | — |
| Apple M5 | Won't fit | UD-IQ1_S | — |
| CPU only · DDR5-5600 | Won't fit | UD-IQ1_S | — |
Smallest devices that fit it fully
- NVIDIA H200 SXM — Tight fit, UD-IQ4_XS, 108–147 tok/s
- NVIDIA B200 — Runs, UD-Q8_K_XL, 153–208 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ1_S).
llama.cpp
llama-server -hf unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF:UD-IQ1_S -c 2048 -ngl 0
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF # Then load quant UD-IQ1_S