CoolFace
Modelpublic

dhanushmekaka/phi-3-mini-text-to-cypher-supply

sourceHugging Facemitupdated 10mo agoView on Hugging Face
0likes26downloads
Model Card

Phi-3 Mini — Fine-Tuned for Text-to-Cypher (Supply Chain) — GGUF Q8_0

By: M. Dhanush Vardhan (dhanushmekaka)

This is a fine-tuned version of Microsoft Phi-3 Mini 4K Instruct, trained for:

  • —Text → Cypher query generation
  • —Supply chain analytics
  • —Inventory, logistics & forecasting queries
  • —Structured reasoning

The model is exported to GGUF (Q8_0) for high-speed CPU inference using:

  • —llama.cpp
  • —Ollama
  • —LM Studio
  • —llama-cpp-python

📁 Files

FileDescription
phi3_gguf_q8_0.ggufMain model file (Q8_0)
README.mdDocumentation

🚀 Run with llama.cpp (CLI)

bash
./main -m https://huggingface.co/dhanushmekaka/phi-3-mini-text-to-cypher-supply/resolve/main/phi3_gguf_q8_0.gguf \
       -p "<|user|>Top 5 products by total ordered quantity.<|end|><|assistant|>"

🐍 Run with Python (llama-cpp-python)

python
from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="dhanushmekaka/phi-3-mini-text-to-cypher-supply",
    filename="phi3_gguf_q8_0.gguf",
    n_ctx=4096,
)

prompt = """<|user|>
Top 5 products by total ordered quantity.
<|end|>
<|assistant|>
"""

response = llm(prompt, max_tokens=150)
print(response["choices"][0]["text"])

💬 Use with Ollama

Modelfile:

FROM ./phi3_gguf_q8_0.gguf
TEMPLATE """<|system|>{{ .System }}<|end|><|user|>{{ .Prompt }}<|end|><|assistant|>"""

Create:

ollama create phi3-supply -f Modelfile

Run:

ollama run phi3-supply

🧠 Model Training Details

  • —Training method: QLoRA (4-bit)
  • —Dataset size: 5k+ records
  • —Task: text → Cypher / analytics reasoning
  • —Frameworks: TRL, PEFT, BitsAndBytes, HuggingFace
  • —Merged to fp16 → converted to GGUF (Q8_0)

📌 License

  • —Base model: MIT
  • —Finetuned checkpoints: permissive for commercial use

❤️ Author

M. Dhanush Vardhan Data Scientist • Supply Chain AI • LLM Engineer HuggingFace: https://huggingface.co/dhanushmekaka

✅ 3️⃣ Ready Python Load Script

This loads your HF model directly and runs inference:

python
from llama_cpp import Llama

# Load GGUF from your HuggingFace repo
llm = Llama.from_pretrained(
    repo_id="dhanushmekaka/phi-3-mini-text-to-cypher-supply",
    filename="phi3_gguf_q8_0.gguf",
    n_ctx=4096,
)

# Prompt (Phi-3 Chat Template)
prompt = """<|user|>
Top 5 products by total ordered quantity.
<|end|>
<|assistant|>
"""

# Run inference
output = llm(prompt, max_tokens=200)
print("Model Response:\n", output["choices"][0]["text"])

🎉 Everything is ready.

If you want, I can also generate:

Just tell me!