faisalsns/phi3-finetuned-electric-rates

Phi-3 Mini 4K Instruct – Fine-Tuned on U.S. Electric Utility Rates (2020)

This model is a fine-tuned version of microsoft/phi-3-mini-4k-instruct on a structured dataset containing U.S. electric utility rate data. The training dataset was sourced from data.gov and reformatted into instruction-based examples for instruction-following language modeling.

💡 Use Case

This model is capable of answering natural language questions such as:

"What is the residential electricity rate for PG&E in California?"

It is trained on structured tabular data in natural language format and can be useful for:

• Question answering over regulatory datasets
• Data summarization
• Instruction-tuned downstream reasoning

🧠 Base Model

• Model: microsoft/phi-3-mini-4k-instruct
• Architecture: Transformer-based causal language model
• Context Length: 4K tokens

🏋️ Fine-tuning Details

• Training Dataset: Aggregated from iou_zipcodes_2020.csv and non_iou_zipcodes_2020.csv
• Sample Size: 147 instruction-style records
• Epochs: 1
• Batch Size: 1
• Precision: fp32 (CPU fine-tuning on low-resource device)

📁 Example Format

Each training sample was structured as follows:

{
  "instruction": "What is the residential electricity rate for PG&E in California?",
  "input": "Zip: 94103, Utility: PG&E, State: CA, Service Type: Residential, Ownership: IOU",
  "output": "The residential rate is $0.21 per kWh."
}

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_path = "your-username/phi3-finetuned-electric-rates"

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)

prompt = """### Instruction:
What is the residential electricity rate for PG&E in California?

### Response:"""

inputs = tokenizer(prompt, return_tensors="pt")
with torch.no_grad():
    output = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(output[0], skip_special_tokens=True))


## 👤 Author

Trained and uploaded by [Faisal Syed].

For feedback or questions, contact [[email protected]] or open an issue on the repo.