output = llm(
"Once upon a time,",
max_tokens=512,
echo=True
)
print(output)Finance Earnings-Call Q&A Assistant
A purpose-built LLM fine-tuned for answering finance questions and recreating earnings-call dialogues
Created by Dhruv Bajaji
🚀 Model Snapshot
This release is a quantized version of the Qwen-2.5 B model, refined on authentic earnings-call transcripts and analyst exchanges from S&P 500 firms and other major markets. It excels at:
- Distilling lengthy calls into clear, actionable takeaways
- Emulating the back-and-forth between analysts and corporate leadership
🏆 Ideal Applications
- Students – Observe the language and framing used by real analysts and executives
- Investors – Request quick call summaries or run mock Q&A sessions
- Researchers & Builders – Integrate a finance-savvy chatbot or mine structured insights from transcripts
🗂 Training Corpus
- Source – Hand-selected Q&A snippets from publicly available earnings-call datasets (e.g., Kaggle)
- Structure – JSONL where each record pairs an analyst’s query with a CFO/CEO reply
- Scale – 1 000 + Q&A examples covering a wide range of scenarios
💡 Sample Prompts
| Analyst Query | Model Reply (sample) |
|---|---|
| “What fueled the 20 % year-over-year revenue jump?” | “Growth stemmed mainly from new subscription tiers and price lifts.” |
| “Why did EU margins narrow this quarter?” | “Logistics inflation and unfavorable FX rates pressured margins.” |
| “Top risks you foresee for next quarter?” | “We see supply-chain uncertainty and currency swings as key threats.” |
🔬 Training Setup
- Base – Qwen-2.5 B (GGUF, 4-bit)
- Environment – Google Colab; low-memory 4-bit quantization
- Method – Fine-tuned with Unsloth + PEFT on the curated dataset
🤝 Credits
- Hugging Face and Kaggle for infrastructure and datasets
- The broader open-source community for continual inspiration
🌐 License
Released under the Apache 2.0 license.
- Downloads last month
- 33
Hardware compatibility
Log In to add your hardware
4-bit
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="DhruvBajaji/finance-analyzer", filename="unsloth.Q4_K_M.gguf", )