You can deploy this model and start earning money today!

ArpitJha/
Indian-FinBert

Name: Indian-FinBert
Availability: InStock

text-embeddings-inference

endpoints_compatible

Views

No views yet

Indian FinBERT — LoRA Fine-tuned Adapter

FinBERT (ProsusAI/finbert) fine-tuned on Indian financial news headlines using LoRA adapters.
Optimised for Indian market sentiment — Nifty 50 stocks, NSE/BSE news, RBI announcements, and Indian business headlines. This repo contains the model itself, you don't need to download file , just go to the How to use model section of this read me and check how to do inference.

What's in this repo

├── config.json
├── labeled_dataset.csv             # training dataset
├── result.png                      # Result image
├── model.safetensor                # Whole merged model
├── tokenizer.json
├── tokenizer_config.json           # Tokenizer settings
└── README.md

Labels

ID	Label	Meaning
0	POSITIVE	Bullish sentiment
1	NEGATIVE	Bearish sentiment
2	NEUTRAL	No clear directional signal

How to use the model

Option A — Quick inference example code

pip install torch==2.6.0 transformers==4.47.0 peft==0.13.0 safetensors==0.4.3

Have an NVIDIA GPU? Install the CUDA build of torch for faster inference:

bash
1pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cu124 2pip install transformers==4.47.0 peft==0.13.0 safetensors==0.4.3

python

import torch
from transformers import pipeline

LABEL_MAP = {"LABEL_0": "POSITIVE", "LABEL_1": "NEGATIVE", "LABEL_2": "NEUTRAL"}

# HuggingFace downloads and caches the model automatically on first run
nlp = pipeline(
    "sentiment-analysis",
    model     = "ArpitJha/Indian-FinBert",
    tokenizer = "ArpitJha/Indian-FinBert",
    device    = 0 if torch.cuda.is_available() else -1
)

headlines = [
    "Reliance Industries posts record quarterly profit.",
    "Adani Group stocks crash amid fraud allegations.",
    "RBI keeps interest rates unchanged in policy meeting.",
    "Infosys wins $2 billion AI transformation deal.",
]

for h in headlines:
    result    = nlp(h)[0]
    sentiment = LABEL_MAP.get(result["label"], result["label"])
    print(f"{sentiment}  ({result['score']*100:.1f}%)  —  {h}")

Batch inference on a CSV

Install pandas using the command : pip install pandas

python

import torch
import pandas as pd
from transformers import pipeline

LABEL_MAP = {"LABEL_0": "POSITIVE", "LABEL_1": "NEGATIVE", "LABEL_2": "NEUTRAL"}

# 1. Initialize the pipeline
print("Loading model...")
nlp = pipeline(
    "sentiment-analysis",
    model="ArpitJha/Indian-FinBert",
    tokenizer="ArpitJha/Indian-FinBert",
    device=0 if torch.cuda.is_available() else -1
)

# 2. Load the CSV data
csv_file_path = "input_data.csv" # Change this to your file's path
print(f"Loading data from {csv_file_path}...")
df = pd.read_csv(csv_file_path)

# Ensure the column exists (replace 'headline' with your actual column name)
text_column = "headline" 
if text_column not in df.columns:
    raise ValueError(f"Column '{text_column}' not found in the CSV. Available columns: {df.columns.tolist()}")

# Convert the column to a standard Python list
texts_to_analyze = df[text_column].astype(str).tolist()

# 3. Run Batch Inference
print(f"Processing {len(texts_to_analyze)} rows. This might take a moment...")
# Adjust batch_size based on your GPU memory (e.g., 8, 16, 32, 64)
results = nlp(texts_to_analyze, batch_size=16)

# 4. Extract labels and scores
mapped_sentiments = []
confidence_scores = []

for result in results:
    mapped_sentiments.append(LABEL_MAP.get(result["label"], result["label"]))
    confidence_scores.append(round(result["score"] * 100, 2)) # Score as a percentage

# 5. Add the results back to the DataFrame
df["predicted_sentiment"] = mapped_sentiments
df["confidence_score_%"] = confidence_scores

# 6. Save the results to a new CSV
output_path = "output_results.csv"
df.to_csv(output_path, index=False)
print(f"Batch inference complete! Results saved to {output_path}")

Requirements

torch>=2.6.0
transformers==4.47.0
peft==0.13.0
safetensors==0.4.3
pandas>=2.2.0          # only needed for batch CSV inference

Model details

Property	Value
Base model	`ProsusAI/finbert`
Fine-tuning method	LoRA (PEFT)
LoRA rank	32
LoRA alpha	64
Target modules	`query`, `value`
Training data	Indian financial news headlines
Task	3-class sentiment classification
Labels	Positive / Negative / Neutral
Trainable parameters	~1% of total

License

The adapter weights in this repository are released under the MIT License.
The base model (ProsusAI/finbert) is subject to its own Apache 2.0 license.