Views
No views yet
| Metric | Score | Notes |
|---|---|---|
| Reasoning Accuracy | 25-50% | Varies by task type |
| Hallucination Rate | 0.0% | Zero confident hallucinations |
| Uncertainty Handling | 100% | Perfect on ambiguous questions |
| Misconception Avoidance | 100% | Avoids common false beliefs |
| Calibration (ECE) | 0.155 | Moderate calibration |
1import torch
2from transformers import AutoTokenizer
3
4# Load model
5model = torch.load("prometheus_model.pt")
6model.eval()
7
8tokenizer = AutoTokenizer.from_pretrained("gpt2")
9tokenizer.pad_token = tokenizer.eos_token
10
11# Generate with reasoning
12prompt = "If all cats are mammals, what can we conclude?"
13inputs = tokenizer(prompt, return_tensors="pt")
14
15with torch.no_grad():
16 output = model.generate(
17 input_ids=inputs['input_ids'],
18 max_length=50,
19 return_reasoning=True,
20 temperature=0.7,
21 repetition_penalty=1.5
22 )
23
24# View reasoning trace
25for step in output['reasoning_trace']:
26 print(f"Step {step['step']}: [{step['type']}] Confidence={step['confidence']:.2f}")
27
28# View generated text
29generated = tokenizer.decode(output['generated_ids'][0], skip_special_tokens=True)
30print(f"Output: {generated}")
31print(f"Final Confidence: {output['confidence'].mean().item():.3f}")1@article{stone2025prometheus,
2 title={Prometheus-1: A Neuro-Symbolic Architecture for Verifiable and Grounded Language Generation},
3 author={Stone, Kent E.},
4 journal={arXiv preprint},
5 year={2025}
6}