Jev directory
CommunityJev Models & Reimplementations
Solar-Decision-JEV-27B-V3-NVFP4-FP8
A Hugging Face model repository from SargeDev publishing Solar-Decision-JEV-27B-V3 in NVFP4 and FP8 quantization. Its published chat template handles text, image, and video content, exposes reasoning-effort levels, and emits XML-style tool calls.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- Model repository SargeDev/Solar-Decision-JEV-27B-V3-NVFP4-FP8 hosted on Hugging Face
- Chat template accepts text, image, and video content items in messages
- Reasoning effort accepts xhigh (the default), medium, and low; other values raise an exception
- Tool calls use <tool_call> tags wrapping a <function=name> block with <parameter=...> values
- Tokenizer configuration sets the pad token to <|endoftext|> and unk_token to null
Questions about this entry
Short answers drawn from the linked project or page.
It is a model page for SargeDev/Solar-Decision-JEV-27B-V3-NVFP4-FP8, whose name indicates a 27B Jev-related decision model distributed in NVFP4 and FP8 formats. The available material covers the page's tokenizer configuration and chat template rather than training details.
Messages can carry plain strings, text items, image or image_url items, and video items. Images and videos are rendered as vision placeholders, and the template raises an exception if a system message contains an image or a video.
Thinking is enabled unless enable_thinking is explicitly false. When enabled, reasoning_effort accepts xhigh (the default), medium, or low; passing any other value raises an exception. xhigh and low carry their own instruction text, while medium relies on the default.
Tools are listed in a <tools> block inside the system message, and the model is told to reply with a <tool_call> element containing a nested <function=name> block with <parameter=...> values. Function calls must follow that format, required parameters must be present, and no suffix may follow the call.
The tokenizer configuration shows a pad token of <|endoftext|> and an unk_token of null, and the chat template is supplied as a Jinja template with reasoning, tools, and vision-placeholder handling.
The provided material does not state a license for this repository. Check the model page itself before redistributing or using the weights commercially.
No. The available material only describes the model page's template and tokenizer configuration; it contains no accuracy, calibration, latency, or benchmark results, so those properties need independent verification before the weights are relied on.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.