---
base_model:
- OmniJev/OneJev-4B
tags:
- text-generation-inference
- llama-cpp
- onejev
- system-one
- decision-model
- calibration
- multimodal
- gui-agent
- video
license: apache-2.0
language:
- en
pipeline_tag: image-text-to-text
library_name: transformers
---

# **OneJev-4B-GGUF**

> **OneJev-4B** is the 4-billion-parameter model of OmniJev's OneJev family, a multimodal "System One" decision model built as a full fine-tune of Qwen3.5-4B on 99,193 questions drawn from real agent runs, videos, and images. It accepts a screenshot, photo, video, or plain text along with a few typed questions, such as `Noul` yes/no or `Choice` over labeled options, and returns a calibrated probability for every option in a single forward pass, with no text generation. It is served with the `qev` package, which speaks TypeSafe's System One API and adds a `media` field for images and video. On one H200 with a 1280x720 screenshot, it answers a single question in 64 ms, or 10 questions in one request in 104 ms (10.4 ms per question). Its results are reported against Jev 1.13, Jev-Omni 12B, and Qwen3.8-27B in thinking mode, on a held-out OneJev test set. The card presents these only as a graphic, so the individual scores are not readable from the text, and it notes that Jev 1.13's figures are its published ones and that it reads text only. OneJev-4B sits between OneJev-0.8B (2.2 GB) and OneJev-9B (18.8 GB), alongside OneJev-27B and an FP8 build of it, and it is released under Apache 2.0 (10.4 GB of weights).

## Model Files

| File Name | Quant Type | File Size | File Link | Description |
|-----------|------------|-----------|-----------|-------------|
| OneJev-4B.BF16.gguf | BF16 | 9.7 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.BF16.gguf) | Full BF16 weights. Highest quality, largest file size. |
| OneJev-4B.Q3_K_L.gguf | Q3_K_L | 2.69 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q3_K_L.gguf) | Lower quality but usable, good for low RAM availability. |
| OneJev-4B.Q3_K_M.gguf | Q3_K_M | 2.54 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q3_K_M.gguf) | Low quality. |
| OneJev-4B.Q4_K_M.gguf | Q4_K_M | 3.07 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q4_K_M.gguf) | Good quality, default size for most use cases, *recommended*. |
| OneJev-4B.Q4_K_S.gguf | Q4_K_S | 2.92 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q4_K_S.gguf) | Slightly lower quality with more space savings, *recommended*. |
| OneJev-4B.Q5_K_M.gguf | Q5_K_M | 3.51 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q5_K_M.gguf) | High quality, *recommended*. |
| OneJev-4B.Q5_K_S.gguf | Q5_K_S | 3.43 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q5_K_S.gguf) | High quality, *recommended*. |
| OneJev-4B.Q6_K.gguf | Q6_K | 3.99 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q6_K.gguf) | Very high quality, near perfect, *recommended*. |
| OneJev-4B.Q8_0.gguf | Q8_0 | 5.16 GB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.Q8_0.gguf) | Extremely high quality, generally unneeded but max available quant. |
| OneJev-4B.mmproj-bf16.gguf | mmproj-bf16 | 676 MB | [Link](https://huggingface.co/prithivMLmods/OneJev-4B-GGUF/blob/main/OneJev-4B.mmproj-bf16.gguf) | Multimodal projection file in BF16 format. Used for vision/language models. |

## llama.cpp

LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp