Image-Text-to-Text
Transformers
Safetensors
English
Korean
Chinese
qwen3_5_moe
feature-extraction
jev
decision
decision-engine
calibrated
calibration
probability
vision
vision-language
multimodal
vlm
image-classification
zero-shot-image-classification
zero-shot-classification
visual-question-answering
multiple-choice
yes-no
scoring
visual-inspection
quality-control
defect-detection
industrial-inspection
food-inspection
agriculture
egg
robotics
physical-ai
edge-ai
document-understanding
screen-understanding
Mixture of Experts
mixture-of-experts
qwen3.5
multilingual
korean
chinese
conversational
Add GitHub badge
Browse files
README.md
CHANGED
|
@@ -1,206 +1,207 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
base_model: FINAL-Bench/Darwin-35B-A3B-Opus
|
| 4 |
-
library_name: transformers
|
| 5 |
-
pipeline_tag: image-text-to-text
|
| 6 |
-
language:
|
| 7 |
-
- en
|
| 8 |
-
- ko
|
| 9 |
-
- zh
|
| 10 |
-
tags:
|
| 11 |
-
- jev
|
| 12 |
-
- decision
|
| 13 |
-
- decision-engine
|
| 14 |
-
- calibrated
|
| 15 |
-
- calibration
|
| 16 |
-
- probability
|
| 17 |
-
- vision
|
| 18 |
-
- vision-language
|
| 19 |
-
- multimodal
|
| 20 |
-
- vlm
|
| 21 |
-
- image-classification
|
| 22 |
-
- zero-shot-image-classification
|
| 23 |
-
- zero-shot-classification
|
| 24 |
-
- visual-question-answering
|
| 25 |
-
- multiple-choice
|
| 26 |
-
- yes-no
|
| 27 |
-
- scoring
|
| 28 |
-
- visual-inspection
|
| 29 |
-
- quality-control
|
| 30 |
-
- defect-detection
|
| 31 |
-
- industrial-inspection
|
| 32 |
-
- food-inspection
|
| 33 |
-
- agriculture
|
| 34 |
-
- egg
|
| 35 |
-
- robotics
|
| 36 |
-
- physical-ai
|
| 37 |
-
- edge-ai
|
| 38 |
-
- document-understanding
|
| 39 |
-
- screen-understanding
|
| 40 |
-
- moe
|
| 41 |
-
- mixture-of-experts
|
| 42 |
-
- qwen3_5_moe
|
| 43 |
-
- qwen3.5
|
| 44 |
-
- multilingual
|
| 45 |
-
- korean
|
| 46 |
-
- chinese
|
| 47 |
-
---
|
| 48 |
-
|
| 49 |
-
<div align="center">
|
| 50 |
-
|
| 51 |
-
# 👁️ PhysicalEye-Decide-35B
|
| 52 |
-
|
| 53 |
-
### Show it an image. Ask a question. Get a calibrated decision in one forward pass.
|
| 54 |
-
|
| 55 |
-
[](https://huggingface.co/spaces/masterset-ai/PhysicalEye-Decide-35B-demo)
|
| 56 |
-
[](https://huggingface.co/FINAL-Bench/Darwin-35B-A3B-Opus)
|
| 57 |
-
[](https://huggingface.co/masterset-ai/PhysicalEye-35B)
|
| 58 |
-
[](https://huggingface.co/spaces/masterset-ai/PhysicalEye-Decide-35B-demo)
|
| 56 |
+
[](https://huggingface.co/FINAL-Bench/Darwin-35B-A3B-Opus)
|
| 57 |
+
[](https://huggingface.co/masterset-ai/PhysicalEye-35B)
|
| 58 |
+
[](https://github.com/masterset-ai/PhysicalEye-Decide)
|
| 59 |
+
[](https://www.apache.org/licenses/LICENSE-2.0)
|
| 60 |
+
|
| 61 |
+
</div>
|
| 62 |
+
|
| 63 |
+
---
|
| 64 |
+
|
| 65 |
+
## What it is
|
| 66 |
+
|
| 67 |
+
PhysicalEye-Decide-35B is a **decision engine for images**. You give it an image, a question and the possible answers. It returns a probability for every answer.
|
| 68 |
+
|
| 69 |
+
It does not generate text. It reads the input once and scores every option directly. That makes it:
|
| 70 |
+
|
| 71 |
+
- **Fast**: about 0.05 seconds per decision on one GPU.
|
| 72 |
+
- **Predictable**: the answer is always one of your options. No parsing, no free text.
|
| 73 |
+
- **Calibrated**: the probability is meant to be used. Set a threshold, route low-confidence cases to a person, and log every score.
|
| 74 |
+
|
| 75 |
+
## Results
|
| 76 |
+
|
| 77 |
+
| Test | Setting | Result |
|
| 78 |
+
|---|---|---|
|
| 79 |
+
| **Egg inspection** | Held-out test, 100 candling photos, good vs bad | **98.0%** accuracy, mean confidence 0.964 |
|
| 80 |
+
| **AI2D** | 200 science diagram questions, 4 options | **87.0%** (chance 25%) |
|
| 81 |
+
| **Speed** | Median latency per decision, batched server, one B200 GPU | **0.051 s** (p80 0.068 s) |
|
| 82 |
+
|
| 83 |
+
## Where to use it
|
| 84 |
+
|
| 85 |
+
| Area | Example question | Type |
|
| 86 |
+
|---|---|---|
|
| 87 |
+
| Food and farm inspection | "Is this egg good or bad?" | `choice` |
|
| 88 |
+
| Factory quality control | "Is there a crack on this part?" | `noul` |
|
| 89 |
+
| Retail and logistics | "Which product is on the shelf?" | `choice` |
|
| 90 |
+
| Documents and screens | "Is the total on this receipt above 50?" | `noul` |
|
| 91 |
+
| Robots and edge devices | "Is the gripper holding the object?" | `noul` |
|
| 92 |
+
| Rating and triage | "How damaged is this package, from 0 to 10?" | `score` |
|
| 93 |
+
|
| 94 |
+
Try these in the **[live demo](https://huggingface.co/spaces/masterset-ai/PhysicalEye-Decide-35B-demo)**: a real-time inspection line, a sample gallery and a playground in English, Korean and Chinese.
|
| 95 |
+
|
| 96 |
+
## Quick start
|
| 97 |
+
|
| 98 |
+
```bash
|
| 99 |
+
pip install -U torch transformers accelerate safetensors pillow huggingface_hub
|
| 100 |
+
wget https://huggingface.co/masterset-ai/PhysicalEye-Decide-35B/resolve/main/decide.py
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
```python
|
| 104 |
+
from decide import Decider
|
| 105 |
+
|
| 106 |
+
d = Decider("masterset-ai/PhysicalEye-Decide-35B") # downloads about 69 GB on first use
|
| 107 |
+
|
| 108 |
+
result = d.decide(
|
| 109 |
+
images=["egg.jpg"],
|
| 110 |
+
state="Candling photo: a light is shone through the egg in a dark room.",
|
| 111 |
+
question={
|
| 112 |
+
"type": "choice",
|
| 113 |
+
"instructions": "Is this egg good or bad?",
|
| 114 |
+
"criteria": {"good": "Normal, healthy egg", "bad": "Defective, spoiled or damaged egg"},
|
| 115 |
+
},
|
| 116 |
+
)
|
| 117 |
+
print(result)
|
| 118 |
+
# {"type": "choice", "choice": "good", "confidence": ..., "probabilities": {"good": ..., "bad": ...}}
|
| 119 |
+
```
|
| 120 |
+
|
| 121 |
+
From the command line:
|
| 122 |
+
|
| 123 |
+
```bash
|
| 124 |
+
python decide.py --image egg.jpg --instructions "Is this egg good or bad?" --options good bad
|
| 125 |
+
python decide.py --image part.jpg --instructions "Is there a crack on this part?" # yes/no
|
| 126 |
+
```
|
| 127 |
+
|
| 128 |
+
**Hardware**: the weights are about 69 GB in bfloat16. Use one GPU with 80 GB or more memory (for example H100 80GB, H200, B200), or spread the model over several GPUs.
|
| 129 |
+
|
| 130 |
+
## Question types
|
| 131 |
+
|
| 132 |
+
Every call takes up to **4 images**, a context string (`state`) and one question.
|
| 133 |
+
|
| 134 |
+
| Type | Fields | Returns |
|
| 135 |
+
|---|---|---|
|
| 136 |
+
| `choice` | `instructions`, `criteria`: a dict of option name to description (or `null`) | `choice`, `confidence`, `probabilities` |
|
| 137 |
+
| `noul` | `instructions`, optional `criteria`: `{"true": ..., "false": ...}` | `noul`: probability of yes |
|
| 138 |
+
| `score` | `instructions`, `criteria`: a list of level descriptions, lowest first | `score` (expected level), `confidence`, `probabilities` |
|
| 139 |
+
|
| 140 |
+
```python
|
| 141 |
+
# yes / no
|
| 142 |
+
d.decide(images=["part.jpg"], state="Steel bracket on a conveyor.",
|
| 143 |
+
question={"type": "noul", "instructions": "Is there a visible crack?"})
|
| 144 |
+
# {"type": "noul", "noul": ...}
|
| 145 |
+
|
| 146 |
+
# score on a scale
|
| 147 |
+
d.decide(images=["box.jpg"], state="Parcel at the receiving dock.",
|
| 148 |
+
question={"type": "score", "instructions": "How damaged is this package?",
|
| 149 |
+
"criteria": ["no damage", "minor dents", "torn or crushed", "destroyed"]})
|
| 150 |
+
# {"type": "score", "score": ..., "confidence": ..., "probabilities": {...}, "legend": {...}}
|
| 151 |
+
```
|
| 152 |
+
|
| 153 |
+
### Tips
|
| 154 |
+
|
| 155 |
+
- **Describe each option in plain words** in `criteria`. Short, concrete descriptions work best.
|
| 156 |
+
- **Put fixed context in `state`**, such as the camera setup or the product type. Keep `instructions` short.
|
| 157 |
+
- **Use the probability.** Accept high-confidence decisions automatically and send the rest to a person.
|
| 158 |
+
- Up to 255 options per question.
|
| 159 |
+
|
| 160 |
+
## How it works
|
| 161 |
+
|
| 162 |
+
1. A vision-language Mixture-of-Experts backbone reads the images, the context and the question together. It has 35B parameters in total and about 3B active per token, so it runs fast for its size.
|
| 163 |
+
2. A small decision readout on top of the backbone scores every option in the same forward pass.
|
| 164 |
+
3. A calibration temperature turns the scores into probabilities you can threshold.
|
| 165 |
+
|
| 166 |
+
## Files
|
| 167 |
+
|
| 168 |
+
| File | Content |
|
| 169 |
+
|---|---|
|
| 170 |
+
| `model-0000X-of-00015.safetensors` | Backbone weights |
|
| 171 |
+
| `readout.safetensors` | Decision readout |
|
| 172 |
+
| `decision_config.json` | Option codes, token ids and calibration temperature |
|
| 173 |
+
| `decide.py` | Self-contained inference code (Apache-2.0). Examples: [GitHub](https://github.com/masterset-ai/PhysicalEye-Decide) |
|
| 174 |
+
| `config.json`, `processor_config.json`, `tokenizer*`, `chat_template.jinja` | Model, image processor and tokenizer settings |
|
| 175 |
+
|
| 176 |
+
## Model family
|
| 177 |
+
|
| 178 |
+
| Model | Role |
|
| 179 |
+
|---|---|
|
| 180 |
+
| [PhysicalEye-35B](https://huggingface.co/masterset-ai/PhysicalEye-35B) | First-person video understanding |
|
| 181 |
+
| **PhysicalEye-Decide-35B** | Image and text decisions with calibrated probabilities |
|
| 182 |
+
|
| 183 |
+
## Base model and license
|
| 184 |
+
|
| 185 |
+
- Built on [FINAL-Bench/Darwin-35B-A3B-Opus](https://huggingface.co/FINAL-Bench/Darwin-35B-A3B-Opus), a Qwen3.5 MoE based vision-language model.
|
| 186 |
+
- Released under the **Apache-2.0** license. The upstream Qwen license notice applies to the base weights.
|
| 187 |
+
|
| 188 |
+
## Training data credits
|
| 189 |
+
|
| 190 |
+
The decision readout was trained only on openly licensed data. We thank the authors of:
|
| 191 |
+
|
| 192 |
+
- **VQAv2** (CC BY 4.0, annotations) and **COCO** captions and instances 2017 (CC BY 4.0, annotations; images keep their Flickr licenses)
|
| 193 |
+
- **TextVQA 0.5.1** (CC BY 4.0), **ScreenQA**, **RICO-SCA** and **Screen2Words** on RICO (CC BY 4.0)
|
| 194 |
+
- **CLEVR v1.0** (CC BY 4.0), **A-OKVQA** (Apache-2.0), **VSR** (CC BY 4.0), **CORD v2** (CC BY 4.0)
|
| 195 |
+
- **Open-Jev** (CC0 1.0), **MMLU auxiliary train** (MIT), synthetic charts made in-house
|
| 196 |
+
- **Good and Bad Eggs Identification Image Dataset**, Md Mafiul Hasan Matin, Marium Jahan, Md. Saroar Jahan, Mendeley Data, doi:10.17632/mdty358x8m.1 (CC BY 4.0)
|
| 197 |
+
|
| 198 |
+
## Citation
|
| 199 |
+
|
| 200 |
+
```bibtex
|
| 201 |
+
@misc{physicaleye_decide_35b,
|
| 202 |
+
title = {PhysicalEye-Decide-35B: One-Pass Calibrated Image Decisions},
|
| 203 |
+
author = {MASTERSET},
|
| 204 |
+
year = {2026},
|
| 205 |
+
url = {https://huggingface.co/masterset-ai/PhysicalEye-Decide-35B}
|
| 206 |
+
}
|
| 207 |
+
```
|