Upload folder using huggingface_hub

.gitattributes

@@ -3,3 +3,4 @@ adapters/adapters.safetensors filter=lfs diff=lfs merge=lfs -text
 model-00001-of-00002.safetensors filter=lfs diff=lfs merge=lfs -text
 model-00002-of-00002.safetensors filter=lfs diff=lfs merge=lfs -text
 tokenizer.model filter=lfs diff=lfs merge=lfs -text
+finance-extractor-v8-f16.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -10,40 +10,55 @@ tags:
 - email
 - bank-statement
 - payment-apps
+- pytorch
+- transformers
+- gguf
+- llama-cpp
 - mlx
 - lora
 - phi-3
 - indian-banking
 - multi-bank
 - structured-output
-- pytorch
-- transformers
 library_name: transformers
 ---
 
 # 🧠 Finance Entity Extractor v0.8.0 (Universal)
 
-> **Now supports Linux/NVIDIA/PyTorch!** Production-ready LLM for structured financial extraction.
+> **Production-ready LLM** for structured financial entity extraction. Works on **Linux/NVIDIA**, **macOS/MLX**, and **any platform via GGUF**.
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97-Hugging%20Face-blue)](https://huggingface.co/Ranjit0034/finance-entity-extractor)
 [![PyTorch](https://img.shields.io/badge/PyTorch-Supported-red)](https://pytorch.org/)
+[![GGUF](https://img.shields.io/badge/GGUF-llama.cpp-green)](https://github.com/ggerganov/llama.cpp)
+
+## 🌟 Platform Support
 
-## 🌟 Features
+| Platform | Framework | Status |
+|----------|-----------|--------|
+| Linux + NVIDIA GPU | PyTorch/Transformers | ✅ Full Support |
+| Linux + CPU | PyTorch/GGUF | ✅ Full Support |
+| Windows | GGUF/llama.cpp | ✅ Full Support |
+| macOS Apple Silicon | MLX | ✅ Full Support |
+| Cloud (AWS/GCP/Azure) | PyTorch/Transformers | ✅ Full Support |
 
-- **Universal Support**: Runs on Linux (NVIDIA/CPU) and Mac (MLX).
-- **Multi-Bank**: HDFC, ICICI, SBI, Axis, Kotak.
+## 🎯 Features
+
+- **Universal Support**: Runs on Linux, Windows, macOS (any hardware).
+- **Multi-Bank**: HDFC, ICICI, SBI, Axis, Kotak + PhonePe, GPay, Paytm.
 - **Structured JSON**: Validated, parseable output.
-- **Accuracy**: 94.5% (Multi-bank), 100% (Real HDFC).
+- **Accuracy**: 94.5% (Multi-bank), 100% (Real HDFC emails).
+
+---
 
-## 📦 Installation
+## ⚡ Quick Start (PyTorch / Linux / NVIDIA)
+
+Recommended for production servers with NVIDIA GPUs.
 
 ```bash
-pip install transformers torch
+pip install transformers torch accelerate
 ```
 
-## ⚡ Quick Start (PyTorch / Linux / NVIDIA)
-
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
@@ -54,7 +69,7 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(
     model_id, 
     torch_dtype=torch.float16, 
-    device_map="auto"
+    device_map="auto"  # Automatically uses GPU
 )
 
 prompt = """Extract financial entities from this email:
@@ -69,18 +84,56 @@ outputs = model.generate(**inputs, max_new_tokens=200)
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
+---
+
+## 🦙 Quick Start (GGUF / llama.cpp)
+
+Recommended for CPU inference / cross-platform deployment / edge devices.
+
+```bash
+pip install llama-cpp-python
+```
+
+```python
+from llama_cpp import Llama
+
+# Download the GGUF file from this repo first
+llm = Llama(model_path="finance-extractor-v8-f16.gguf")
+
+output = llm(
+    "Extract financial entities from: Rs.500 debited from A/c 1234 on 01-01-25\nOutput JSON:",
+    max_tokens=200
+)
+print(output["choices"][0]["text"])
+```
+
+**llama.cpp CLI:**
+```bash
+./main -m finance-extractor-v8-f16.gguf \
+    -p "Extract financial entities from: Rs.500 debited from A/c 1234 on 01-01-25"
+```
+
+---
+
 ## 🍏 Quick Start (Apple Silicon / MLX)
 
+Recommended for Mac developers using Apple Silicon.
+
 ```bash
 pip install mlx-lm
 ```
 
 ```python
 from mlx_lm import load, generate
+
 model, tokenizer = load("Ranjit0034/finance-entity-extractor", adapter_path="adapters")
-# ... use as normal
+prompt = "Extract financial entities from: Rs.500 debited from A/c 1234 on 01-01-25"
+response = generate(model, tokenizer, prompt=prompt, max_tokens=200)
+print(response)
 ```
 
+---
+
 ## 📊 Evaluation
 
 | Bank | Accuracy | Status |
@@ -88,14 +141,19 @@ model, tokenizer = load("Ranjit0034/finance-entity-extractor", adapter_path="ada
 | ICICI | 100% | ✅ |
 | HDFC | 95% | ✅ |
 | SBI | 93.3% | ✅ |
+| Axis | 93.3% | ✅ |
+| Kotak | 92% | ✅ |
 | **Overall** | **94.5%** | 🏆 |
 
-## 📁 Files
+## 📁 Repository Files
 
-- `model.safetensors`: Full 7GB PyTorch model (Dequantized).
-- `adapters/`: LoRA adapters (25MB) for bandwidth constrained users.
-- `inference.py`: Production API wrapper.
-- `train.py`: Reproducible training script.
+| File | Size | Description |
+|------|------|-------------|
+| `model-*.safetensors` | ~7.1GB | Full PyTorch model (bfloat16) |
+| `finance-extractor-v8-f16.gguf` | ~7.1GB | GGUF for llama.cpp (F16) |
+| `adapters/` | ~24MB | LoRA adapters for MLX |
+| `inference.py` | - | Production API wrapper |
+| `train.py` | - | Reproducible training script |
 
 ---
 **Made with ❤️ by Ranjit Behera**

adapters/adapter_config.json

@@ -1,12 +1,12 @@
 {
-    "adapter_path": "models/adapters/finance-lora-v8",
+    "adapter_path": "models/adapters/finance-lora-v6",
     "batch_size": 1,
     "config": null,
     "data": "data/training",
     "fine_tune_type": "lora",
     "grad_accumulation_steps": 1,
     "grad_checkpoint": false,
-    "iters": 800,
+    "iters": 1500,
     "learning_rate": 1e-05,
     "lora_parameters": {
         "rank": 8,

adapters/adapters.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af742ba0c9de0119ea1f0a667a8dca8f42d105d67f52401bb514e79f6e59937c
+oid sha256:1c98bb44acb9bdf50180c215020c6b93ffa42e1b9626bc3b2eac8522b2e6bf03
 size 25179794

finance-extractor-v8-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a130047e6928fe3bd2eedc2ff9fc556263325d1bc9ee7de59634bd7a9b601b2
+size 7643296448