prakhar146/financial-llm-cpu

Financial LLM - CPU Ready

Fine-tuned Qwen/Qwen2.5-1.5B-Instruct on 74,998 financial samples.

Capabilities

• 📈 Technical analysis with step-by-step reasoning
• 💹 Trading decision making
• 📊 Portfolio allocation and risk management
• 🏦 Macro economic analysis
• 🇮🇳 India-specific financial context

Training

• Dataset: 74,998 samples (FinGPT + FinQA + Synthetic CoT)
• Method: QLoRA → Merged to full model
• Training loss: 0.1089
• Eval loss: 0.5173

CPU Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model = AutoModelForCausalLM.from_pretrained(
    "prakhar146/financial-llm-cpu",
    torch_dtype=torch.float32,
    device_map="cpu"
)
tokenizer = AutoTokenizer.from_pretrained("prakhar146/financial-llm-cpu")

Benchmarks

• Sentiment Analysis: 80.0%
• Trading Decisions: 75.0% (beats GPT-4's 72%)
• Reasoning Quality: 70/100