Jev directory
CommunityJev Models & Reimplementations
PPLX Decider Flash v1.2
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
- Reading time
- 2 min
- Published by
- Community
- Author
- Perplexity AI
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locallyRequires API key
Highlights
- About one seventh the size of pplx-decider-v1.2, built on Qwen3.5-4B and distilled from the 27B decision models' probability distributions.
- Self-reported Decision Index 0.3: 58.88 with joint requests and 60.05 with one question per request, versus 68.09 for the 27B model.
- Vision Decision Index 72.53 against 77.43 for the 27B model; it handles text and images up to 1.6 MP.
- Combines several training runs through model souping; for this model one question per request is the safer default.
- Needs Python 3.12+ and a CUDA GPU with room for about 9 GiB of BF16 weights; the repository is private and needs a token.
Quickstart
python
import sys
from pathlib import Path
from huggingface_hub import snapshot_download
checkpoint = Path(snapshot_download("perplexity-ai/pplx-decider-flash-v1.2"))
sys.path.insert(0, str(checkpoint / "source" / "src"))
from pplx_decider import Decider
decider = Decider.from_pretrained(checkpoint, device="cuda")
response = decider.decide({
"state": "My integration keeps failing with a 401 since we rotated our API key. Please help.",
"questions": {"urgent": {"type": "noul", "instructions": "Is the customer blocked right now?"}},
})
print(response["answers"]["urgent"]["noul"])Watch out
Card reviewed 2026-10-17; the Decision Index and vision scores are self-reported on Perplexity's own public evaluation panel, the repository is private and requires an access token despite the Apache-2.0 metadata, and no third-party reproduction of the distillation comparisons is linked.
Questions about this entry
Short answers drawn from the linked project or page.
Perplexity's small decision model: a Qwen3.5-4B checkpoint distilled from the 27B pplx-decider v1.2 models. It answers typed choice, noul and score questions over text and images in one forward pass and returns probabilities instead of generated text.
Same shape: you send a state, one or more typed questions and criteria, and it returns choice, probabilities and confidence for each question with zero output tokens. It also supports joint requests, where several questions share one sequence.
On Decision Index 0.3 public it reports 58.88 overall with joint requests and 60.05 with one question per request, against 68.09 for the 27B model; vision is 72.53 versus 77.43. All figures are Perplexity's own evaluations.
The card says one question per request is the safer default for this model: unlike the 27B model it is more accurate that way on most multi-question tasks. Grouping helps only on tightly coupled facets, at +6.6 on SATA-Bench and +3.4 on the home appliance simulator.
Python 3.12+ plus a CUDA GPU with room for about 9 GiB of BF16 weights and working memory. The repository is private, so authenticate with a Hugging Face token that has access, then install requirements.txt from the checkout.
Perplexity AI built it on Qwen/Qwen3.5-4B as a distilled sibling of pplx-decider-v1.2. The student learns the 27B teachers' full probability distributions on their training requests, not just their answers, and several runs are combined by weight averaging.
The card suggests decider.logits(requests), which returns the raw readout logits of each question's valid options before temperature. For ranking, use the true-minus-false logit margin of a noul question, which does not saturate into ties the way probabilities do.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
X: Standard One SH (StandardThinking)
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 NewsMore Jev resources like this
Perplexity's 27B decision model rebuilt on Qwen3.8-27B with a joint request format that answers a whole request's questions in one sequence, reporting 68.09 on Decision Index 0.3 and 77.43 on vision.
Jev Models & Reimplementations#community#open-models#multimodal
Community
A 4B decision model taking text, JSON, or images up to 262,144 tokens: 52.75 on Decision Index 0.2.1, 204 of 231 on JevBench public, and 26.0 ms for a three-question request on one H200.
Jev Models & Reimplementations#community#open-models#local
Community
InternLM's multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-4B: a shared state, a schema of named questions and optional images go in, and every question returns an answer distribution from one forward pass. The card reports a 90.02 seven-suite average.
Jev Models & Reimplementations#community#open-models#multimodal
Community
InternLM's 0.8B multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-0.8B: a state, named questions and optional images produce an answer distribution per question in one forward pass. Self-reported seven-suite average 79.38, Brier 0.530.
Jev Models & Reimplementations#community#open-models#multimodal
Community
InternLM's 2B multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-2B: a state, named questions and optional images produce an answer distribution per question in one forward pass. Self-reported seven-suite average 84.68, Brier 0.437.
Jev Models & Reimplementations#community#open-models#multimodal
Community
An unofficial cross-encoder fine-tune of ModernBERT-ja-310m for Jev Local: it reads a question-plus-state and one candidate per sequence and hits 92.62% on the JGLUE JNLI test, with CC BY-SA 4.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
From 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.