--- title: StartLux-Decision-4B canonical_url: "https://www.modelscope.cn/models/StartLuxAI/StartLux-Decision-4B" md_url: "https://www.modelscope.cn/models/StartLuxAI/StartLux-Decision-4B.md" repository: StartLuxAI/StartLux-Decision-4B last_updated: 2026-10-04 license: cc-by-nc-4.0 model_type: - qwen3_5 architectures: - Qwen3_5ForConditionalGeneration parameters: 4.7B tensor_type: - BF16 - F32 library_name: - safetensors language: - en downloads: 58 stars: 0 tags: - decision-model - typed-decisions - classification --- # StartLux-Decision-4B > StartLux-Decision-4B - StartLuxAI 在 ModelScope 开源的模型。StartLux-Decision-4B StartLuxAI/StartLux-Decision-4B 是 ModelScope 魔搭社区上的 4.7B 参数机器学习模型，采用 cc-by-nc-4.0 许可。 - **Repository**: StartLuxAI/StartLux-Decision-4B - **License**: cc-by-nc-4.0 - **Parameters**: 4.7B - **Tags**: decision-model, typed-decisions, classification - **Downloads**: 58 - **Stars**: 0 - **Last updated**: 2026-10-04 Source: https://www.modelscope.cn/models/StartLuxAI/StartLux-Decision-4B --- # StartLux-Decision-4B

StartLux-Decision: a probability for every option

StartLux-Decision-4B answers typed questions about a state: pick one of several options, yes or no, or a rating on a scale. The state can be text, JSON or images, up to 262,144 tokens (256K). Every question comes back with a probability for each option. Requests and responses use the TypeSafe `/v1/systemone` format, so clients written for Jev work unchanged. Sizes: [0.8B](https://huggingface.co/startlux-models/StartLux-Decision-0.8B) · [2B](https://huggingface.co/startlux-models/StartLux-Decision-2B) · **4B** · [9B](https://huggingface.co/startlux-models/StartLux-Decision-9B) · [27B](https://huggingface.co/startlux-models/StartLux-Decision-27B) · [35B-A3B](https://huggingface.co/startlux-models/StartLux-Decision-35B-A3B) ## Results
| StartLux-Decision-4B | | |:---:|:---:| | Decision Index 0.2.1 | 52.75 | | Decision Index 0.2 | 48.38 | | JevBench public, correct of 231 | 204 | | Intern-Decision, average accuracy over seven suites | 91.17 | | Latency, one request with three questions | 26.0 ms | | Latency, one yes/no question | 14.7 ms | | Input | text, JSON or images, up to 262,144 tokens |
Latency is end to end over HTTP on one H200 in bf16, one request at a time.

Decision Index 0.2.1

Decision Index at every size

### Compared with other decision models
| Model | JevBench public, of 231 | Intern avg | DI 0.2 / 0.2.1 | Latency, 3 questions | |:---:|:---:|:---:|:---:|:---:| | StartLux-Decision-35B-A3B | 210 | 92.29 | 57.24 / 61.55 | 52.5 ms | | StartLux-Decision-27B | 208 | 91.82 | 59.54 / 63.88 | 102.3 ms | | StartLux-Decision-9B | 201 | 91.08 | 54.37 / 58.63 | 35.7 ms | | **StartLux-Decision-4B** | 204 | 91.17 | 48.38 / 52.75 | 26.0 ms | | Intern-Decision-4B | 201 | 90.02 | 35.90 / 37.81 | 44.2 ms ¹ | | JevK5 | 200 | 85.16 | 36.44 / 38.81 | | | Jev 1.13 | 199 | 88.74 | 51.67 / 57.91 | 64.0 ms ² | | StartLux-Decision-2B | 196 | 88.46 | 40.72 / 44.19 | 15.5 ms | | SemIf | 187 | 84.23 | 25.70 / 25.94 | | | Intern-Decision-2B | 180 | 84.68 | 19.49 / 19.38 | 33.3 ms ¹ | | StartLux-Decision-0.8B | 179 | 85.03 | 35.57 / 38.86 | 12.2 ms | | Intern-Decision-0.8B | 163 | 79.38 | 11.32 / 11.94 | 34.0 ms ¹ | | Laya | 130 | 57.77 | 5.51 / 6.04 | |
JevBench public counts the correct answers on the 231 public items in the Intern-Decision bundle; Intern avg is the average accuracy over its seven suites; DI is the Decision Index under both editions, the public board's values for the other systems. A blank cell means the number is not published. StartLux-Decision latencies are for one H200, with the three questions answered in one forward pass. ¹ Intern-Decision's own measurement on an RTX 4090. ² The server time the TypeSafe API gateway reports for the same request, mean of 100, network left out as in ours; Intern-Decision reports 109.7 ms end to end.

Latency on one H200

## Fast inference The folder ships its own inference package, `startlux_decision/`, which is the fast path: - `requirements.txt` installs the fast kernels, `flash-linear-attention` and `causal-conv1d`, and `python -m startlux_decision.check .` confirms they are active. Without them transformers falls back to a path more than ten times slower, and the server refuses to start on a GPU. - All questions of a request run in one forward pass, and the server records CUDA graphs at start-up and replays them for short requests: on one H200 a request with three questions takes 26.0 ms end to end and a single yes/no question 14.7 ms. - For bulk work, `decide_batch` batches the questions of many requests together, which is several times faster than sending them one at a time. ## Images and long inputs The weights include a vision tower, and the inference package uses it: a request can carry images as part of its evidence, `images=[...]` in Python (PIL images, file paths, encoded bytes, base64 strings or data URIs) or `"images": [...]` over HTTP (base64 strings or data URIs). `` in a string state marks where each image goes. Prompts can run to 262,144 tokens (256K), the model's native context: a long state is read once, in chunks, and every question of the request branches off it. The MLX backend and the GGUF files read text only. `confidence` follows TypeSafe's definitions (for a choice, (p_max − 1/n) / (1 − 1/n)); the top probability is in `probabilities`. ## Usage ```bash hf download startlux-models/StartLux-Decision-4B --local-dir StartLux-Decision-4B cd StartLux-Decision-4B pip install -r requirements.txt python -m startlux_decision.check . # must print "fast kernels: active" python -m startlux_decision.server --model . --port 8090 ``` ```bash curl -s localhost:8090/v1/systemone -H 'Content-Type: application/json' -d '{ "state": {"ticket": "I was charged twice for order #4411 and the app still shows it as unpaid."}, "questions": { "team": {"type": "choice", "instructions": "Which team should handle this ticket?", "criteria": {"billing": "Payments, refunds and invoices", "shipping": "Delivery and tracking", "technical": "App, login and account problems"}}, "urgent": {"type": "noul", "instructions": "Should this ticket be answered today?"}, "severity": {"type": "score", "instructions": "How severe is the impact?", "criteria": ["cosmetic", "annoying", "blocks the customer"]} } }' ``` Or in Python, from the same folder: ```python from startlux_decision import StartLuxDecision m = StartLuxDecision(".") answers, usage = m.decide(state, questions) many = m.decide_batch([(state, questions), ...]) ``` With images, from the same folder: ```python answers, usage = m.decide("Photo taken at delivery: ", {"damaged": {"type": "noul", "instructions": "Is the parcel damaged?"}}, images=["parcel.jpg"]) ``` ## License The model weights are released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/): free for research and other non-commercial use, with attribution. Commercial use requires a separate license from StartLux Labs; contact [contact@startlux.com](mailto:contact@startlux.com). The inference code in `startlux_decision/` is Apache-2.0. See [LICENSE](https://huggingface.co/startlux-models/StartLux-Decision-4B/blob/main/LICENSE) and [NOTICE](https://huggingface.co/startlux-models/StartLux-Decision-4B/blob/main/NOTICE).