Llama-3.1-8B-Table-Finetuned

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on table-based question answering tasks.

Model Details

Base Model: meta-llama/Meta-Llama-3.1-8B-Instruct
Fine-tuning Method: QLoRA (4-bit Quantized Low-Rank Adaptation)
Context Length: 16K tokens
Training Data: Table-based question answering dataset

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("pandoradox/llama-3.1-8B-table-finetuned_1.2k")
tokenizer = AutoTokenizer.from_pretrained("pandoradox/llama-3.1-8B-table-finetuned_1.2k")

# Format your input with a table
prompt = '''
<|system|>
You are an expert at analyzing tables and answering questions about them.
<|end|>
<|user|>
Based on the following table:

Table title: Example Table

Headers: Name, Age, City

Row 1: John, 30, New York
Row 2: Jane, 25, Boston
Row 3: Bob, 35, Chicago

Question: Who is the oldest person in the table?
<|end|>
<|assistant|>
'''

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_length=2048)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Limitations

The model may struggle with extremely complex tables or ambiguous questions
Performance may vary based on how tables are formatted in the input prompt