Model Card for letxbe/mistral-7b-v03-BoundingDocs-rephrased

letxbe/mistral-7b-v03-BoundingDocs-rephrased is a fine-tuned Mistral-7B-v0.3 for the Document Question Answering task. It was trained on BoundingDocs using the rephrased version of the questions.

Model Details

Model Description

• Developed by: LetXBe
• Model Type: LLM
• Languages: Multilingual
• License: CC BY 4.0
• Finetuned From: Mistral-7B-v0.3
• Input Format: Text using custom prompt
• Output Format: JSON

🚀 How to Use

Prompt

The model should be prompted with this prompt:

TEMPLATE_PROMPT = '''<|startdocument|>
{DOCUMENT}
<|enddocument|>

<|starttask|>
Answer the following question about the document:
Question: "{QUESTION}"
Answer completing the following format:

'''json
{"value": ""}
'''

<|endtask|>
'''

where DOCUMENT is the textual content of the document page.

Inference Example

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("letxbe/mistral-7b-v03-BoundingDocs-rephrased")
model = AutoModelForCausalLM.from_pretrained("letxbe/mistral-7b-v03-BoundingDocs-rephrased")

# Encode input
input_text = "Your prompt"
inputs = tokenizer(input_text, return_tensors="pt")

# Generate response
outputs = model.generate(**inputs)

# Decode and print
print(tokenizer.decode(outputs[0], skip_special_tokens=True))