96 lines
3.7 KiB
Markdown
96 lines
3.7 KiB
Markdown
---
|
|
library_name: transformers
|
|
tags:
|
|
- text-generation-inference
|
|
- code
|
|
- fact
|
|
- math
|
|
- short-context-reasoning
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
- zh
|
|
base_model:
|
|
- Qwen/Qwen2.5-0.5B-Instruct
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
|
|
# **TESS-QwenRe-Fact-0.5B**
|
|
|
|
> **TESS-QwenRe-Fact-0.5B** is a **compact fact-checking and short reasoning model** built upon **Qwen2.5 0.5B**. Designed for rapid response, real-world fact verification, and concise logical reasoning, this lightweight model is ideal for digital assistants, quick-response tools, and misinformation detection systems in **English** and **Chinese**.
|
|
|
|
## **Key Features**
|
|
|
|
1. **Fact Verification & Correction**
|
|
Trained to analyze factual accuracy in statements and offer corrected or clarified responses, making it ideal for real-time verification tasks and misinformation mitigation.
|
|
|
|
2. **Concise Reasoning**
|
|
Specializes in **short-form reasoning**, capable of analyzing and explaining claims, decisions, or statements in just a few logical steps — perfect for Q&A bots and assistant systems.
|
|
|
|
3. **Multilingual Support (EN + ZH)**
|
|
Supports fact-checking tasks in both **English** and **Simplified Chinese**, enhancing accessibility for bilingual or regional use cases.
|
|
|
|
4. **Built on Qwen2.5 0.5B**
|
|
Combines the latest architectural improvements from **Qwen2.5** with a small parameter footprint (0.5B), optimized for **speed**, **efficiency**, and **edge-device compatibility**.
|
|
|
|
5. **Prompt-Friendly Output**
|
|
Responds well to well-structured queries, returning clean, interpretable answers — especially for true/false classification, source-based fact validation, and yes/no reasoning.
|
|
|
|
## **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/TESS-QwenRe-Fact-0.5B"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Is the capital of Australia Sydney? Explain briefly."
|
|
messages = [
|
|
{"role": "system", "content": "You are a concise and accurate fact-checking assistant."},
|
|
{"role": "user", "content": prompt}
|
|
]
|
|
text = tokenizer.apply_chat_template(
|
|
messages,
|
|
tokenize=False,
|
|
add_generation_prompt=True
|
|
)
|
|
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
|
|
|
generated_ids = model.generate(
|
|
**model_inputs,
|
|
max_new_tokens=256
|
|
)
|
|
generated_ids = [
|
|
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
|
]
|
|
|
|
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
```
|
|
|
|
## **Intended Use**
|
|
|
|
- **Fact-Checking Assistants**: Quickly verify factual claims in conversation or content.
|
|
- **Digital Truth Detectors**: Misinformation and rumor detection in social feeds or news summaries.
|
|
- **Micro-Reasoning Bots**: Smart agents for short-form logic and rationale generation.
|
|
- **Multilingual Knowledge Tools**: Fact reasoning in **EN/ZH**, ideal for diverse platforms.
|
|
|
|
## **Limitations**
|
|
|
|
1. **Limited Depth**
|
|
Focused on **short-form reasoning** — may not perform well on multi-step or abstract logic tasks.
|
|
|
|
2. **Compact Model Scale**
|
|
At 0.5B parameters, it prioritizes **efficiency over complexity** — best for straightforward fact-based tasks.
|
|
|
|
3. **Language & Topic Bias**
|
|
Inherits limitations and biases from its base model Qwen2.5 0.5B. Use carefully in sensitive contexts.
|
|
|
|
4. **Prompt Clarity Required**
|
|
Structured prompts result in higher factual accuracy and shorter response latency. |