Qwen3-4B-Instruct-2507-Decision

Qwen3-4B-Instruct-2507-Decision

🤗 Model · 🎮 Live Demo · 💻 Source

A prompt-programmable low-latency decision model built on Qwen3.

v0.1 is an inference interface over Qwen/Qwen3-4B-Instruct-2507. It does not ship new weights.

Live Demo: https://huggingface.co/spaces/dsif2012/Qwen3-Decision-v1

context + question + candidates
        ↓
   Qwen logits
        ↓
 candidate softmax
        ↓
 selected + confidence

Code and serving: github.com/dsif2012/Qwen3-4B-Instruct-2507-Decision

Supported

• BOOLEAN → true / false
• CHOICE → one label from a dynamic candidate set
• Full candidate scores, confidence, margin
• Prefix-cache optimized inference (vLLM)

Example

{
  "context": "...",
  "questions": [
    {
      "id": "can_attack",
      "type": "BOOLEAN",
      "question": "Can the actor attack now?"
    },
    {
      "id": "next_action",
      "type": "CHOICE",
      "question": "What should the actor do next?",
      "choices": ["ATTACK", "CHASE", "WAIT", "RETREAT"]
    }
  ]
}

{
  "can_attack": {"value": true, "confidence": 0.96, "margin": 0.92},
  "next_action": {
    "value": "CHASE",
    "confidence": 0.81,
    "margin": 0.69,
    "ranking": [["CHASE", 0.81], ["ATTACK", 0.12], ["WAIT", 0.05], ["RETREAT", 0.02]]
  }
}

Use the base weights

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "Qwen/Qwen3-4B-Instruct-2507"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="bfloat16", device_map="auto")

Decision scoring lives in the GitHub package (pcd). Load Qwen3, then POST /decision via pcd-cuda.

License

Apache-2.0. Qwen3-4B-Instruct-2507 weights remain under the original Qwen Apache-2.0 license.