PKU-KCL-Meteor-v1.5-0.8B
a fast decision-making and probability estimation model
2 min readupdated
PKU-KCL-Meteor-v1.5-0.8B is a decision-making model developed by Peking University's Knowledge Computing Lab (KCL) and built on Qwen/Qwen3.5-0.8B (0.85B parameters). Instead of generating text, it outputs probabilities for each answer option in a single forward pass. The model supports choice (multiple-choice selection), noul (yes/no probability), and score (level-based rating) question types. Built for structured analysis, classification, and logical selection in a System One format, it delivers high-speed inference without redundant token generation.
Key features
- No text generation: outputs probabilities for every option in a single forward pass.
- Question types: supports multiple choice (up to 255 options), yes/no probabilities, and rating scales from 2 to 10.
- Context window: processes long prompts up to 262,144 tokens without truncation.
- Performance: runs on a single GPU using BF16 weights with FP32 readout.
What you can do
- Service integration: spin up a local server at
/v1/systemonefor rapid option scoring. - Analysis and classification: power structured decision-making inside bots and automated pipelines.
- Criteria evaluation: feed complex system states into the model to get action probabilities.
Example
from meteor import Meteor
model = Meteor()
model.decide(
"The traffic light turned red as the car approached.",
{"action": {"type": "choice", "criteria": {"stop": "Stop", "go": "Keep going"}},
"safe": {"type": "noul"}},
)
Read more
DetailsBenchmarks: 58.67 Decision Index score across 140,178 queries.
Hardware: single NVIDIA GPU, batched text processing (shortest first).
Limitations: text-only; commercial use requires approval from the lab (CC BY-NC 4.0).










Comments