JEV 27B VL needs about 19.2 GB of memory at Q4_K_M (a 16.5 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 3 quants, from 16.5 GB (Q4_K_M) to 53.8 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
JEV 27B VL – a 27 B parameter vision‑language model for decision tasks
JEV 27B VL is a 27 B parameter Qwen3.5‑based model with a 262 144 token context window. It supports vision input and is optimized for decision‑making via a patched llama.cpp endpoint. The model is released under Apache‑2.0.
Good at
- Vision‑language decision tasks
- Fast one‑pass decisions
- Low‑latency inference on GPUs
Limits
- Requires patched llama.cpp
- Q4_K_M less accurate, about 1 in 14 decisions differ
- No free‑text generation benchmarked
Best for
- Embedded decision systems
- Real‑time recommendation engines
- Robotics perception pipelines
Who it’s for. Engineers building vision‑enabled decision models who need a large context window and a calibrated decision head. The model is suitable for applications that require quick, single‑pass decisions on image‑rich data.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 16.5 GB | 19.2 GB | Too big | Get file → | |
| Q8_0 | 28.6 GB | 31.2 GB | Too big | Get file → | |
| BF16 | 53.8 GB | 56.5 GB | Too big | Get file → |
For 24 GB VRAM, use Q4_K_M (16.55 GB file, ~19.2 GB VRAM at 8 k context). For 32 GB VRAM, Q8_0 (28.6 GB file, ~31.2 GB VRAM). For 56 GB VRAM, BF16 (53.81 GB file, ~56.5 GB VRAM). Memory estimates are approximate.
What GPU do you need for JEV 27B VL?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: the best quant that fits is Q4_K_M (16.5 GB file, about 19.2 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (28.6 GB file, about 31.2 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (28.6 GB file, about 31.2 GB in use).
Which JEV 27B VL quant should you choose?
Q4_K_M is the usual starting point for JEV 27B VL: it balances file size and quality well. Here it is a 16.5 GB file.
The Q8_0 file is 28.6 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about JEV 27B VL
How much VRAM does JEV 27B VL need?
About 19.2 GB at Q4_K_M with an 8K context window. That is the 16.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run JEV 27B VL on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 19.2 GB, so you would need a larger GPU or CPU offloading.
What is the file size of JEV 27B VL GGUF?
It depends on the quant: Q4_K_M is 16.5 GB, Q8_0 is 28.6 GB, BF16 is 53.8 GB.
What context length does JEV 27B VL support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download JEV 27B VL GGUF?
The files are on Hugging Face in the Atlas3D/JEV-27B-VL-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does JEV 27B VL use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run JEV 27B VL on an RTX 6000 with 24 GB VRAM?
Yes, using the Q4_K_M quantization. It uses about 22.7 GB VRAM at 4k context, which fits the RTX 6000. Accuracy is slightly lower than Q8_0.
Does the model support free‑text generation?
Free‑text generation was not benchmarked. The model is primarily optimized for decision tasks via the patched llama.cpp endpoint.