saicharan1010/ SmolLM2-FT-legal-india:latest

91 Downloads Updated 1 year ago

This is a fine-tuned version of the SmolLM2-135M-Instruct model, trained on legal texts from the Indian-Law dataset by vishnun0027 on Hugging Face.

ollama run saicharan1010/SmolLM2-FT-legal-india

curl http://localhost:11434/api/chat \
  -d '{
    "model": "saicharan1010/SmolLM2-FT-legal-india",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='saicharan1010/SmolLM2-FT-legal-india',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'saicharan1010/SmolLM2-FT-legal-india',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 year ago

1 year ago

e4af0090fbcb · 271MB ·

model

archllama

·

parameters135M

·

quantizationF16

271MB

template

{{- range .Messages }}<|im_start|>{{ .Role }} {{ .Content }}<|im_end|> {{ end }}<|im_start|>assistan

102B

params

{ "stop": [ "<|im_start|>", "<|im_end|>" ] }

59B

Readme

Legal Language Model

This is a fine-tuned version of the SmolLM2-135M-Instruct model, trained on legal texts from the Indian-Law dataset by vishnun0027 on Hugging Face.

🤗 Model Availability

The model is publicly available on Hugging Face: saicharan1010/SmolLM2-FT-legal-india

The model is publicly available on Ollama: saicharan1010/SmolLM2-FT-legal-india

Model Information

Base Model: HuggingFaceTB/SmolLM2-135M-Instruct
Dataset: Indian-Law (25,600 instruction-response pairs after cleaning)
Training: Fine-tuned using SFT (Supervised Fine-Tuning)

Training Details

Training Steps: 1,000 steps
Batch Size: 16
Learning Rate: 5e-5
Final Training Loss: 1.086

Performance Evaluation

Evaluation on 1,280 test samples showed improved legal reasoning compared to the base model: - BLEU Score: 0.126 (compared to base model’s 0.121) - ROUGE-L F-Score: 0.304

Usage

Using Ollama

Run the model directly with Ollama:

ollama run saicharan1010/SmolLM2-FT-legal-india

This will start an interactive session where you can query the model about Indian legal concepts.

Using Transformers Library

Alternatively, run the model using the transformers library:

from transformers import pipeline, AutoTokenizer
# Load tokenizer and create pipeline
tokenizer = AutoTokenizer.from_pretrained("saicharan1010/SmolLM2-FT-legal-india")
pipe = pipeline("text-generation", model="saicharan1010/SmolLM2-FT-legal-india")
# Format with chat template
prompt = "Can a Vakalatnama be revoked or withdrawn in India?"
messages = [{"role": "user", "content": prompt}]
formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False)
# Generate response
response = pipe(formatted_prompt, max_new_tokens=200)
print(response[0]['generated_text'])

This model is specifically optimized for legal language tasks in the Indian context. It shows improved understanding of Indian legal terminology and concepts compared to the base model.

# Legal Language Model
This is a fine-tuned version of the **SmolLM2-135M-Instruct** model, trained on legal texts from the **Indian-Law** dataset by [vishnun0027](https://huggingface.co/datasets/vishnun0027/Indian-Law) on Hugging Face.

## 🤗 Model Availability
**The model is publicly available on Hugging Face:** [saicharan1010/SmolLM2-FT-legal-india](https://ollama.com/saicharan1010/SmolLM2-FT-legal-india)

**The model is publicly available on Ollama:** [saicharan1010/SmolLM2-FT-legal-india](https://huggingface.co/saicharan1010/SmolLM2-FT-legal-india)

## Model Information
- **Base Model**: HuggingFaceTB/SmolLM2-135M-Instruct
- **Dataset**: Indian-Law (25,600 instruction-response pairs after cleaning)
- **Training**: Fine-tuned using SFT (Supervised Fine-Tuning)

## Training Details
- **Training Steps**: 1,000 steps
- **Batch Size**: 16
- **Learning Rate**: 5e-5
- **Final Training Loss**: 1.086

## Performance Evaluation
Evaluation on 1,280 test samples showed improved legal reasoning compared to the base model:
- **BLEU Score**: 0.126 (compared to base model's 0.121)
- **ROUGE-L F-Score**: 0.304

## Usage

### Using Ollama
Run the model directly with Ollama:

```bash
ollama run saicharan1010/SmolLM2-FT-legal-india
```

This will start an interactive session where you can query the model about Indian legal concepts.

### Using Transformers Library
Alternatively, run the model using the `transformers` library:

```python
from transformers import pipeline, AutoTokenizer
# Load tokenizer and create pipeline
tokenizer = AutoTokenizer.from_pretrained("saicharan1010/SmolLM2-FT-legal-india")
pipe = pipeline("text-generation", model="saicharan1010/SmolLM2-FT-legal-india")
# Format with chat template
prompt = "Can a Vakalatnama be revoked or withdrawn in India?"
messages = [{"role": "user", "content": prompt}]
formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False)
# Generate response
response = pipe(formatted_prompt, max_new_tokens=200)
print(response[0]['generated_text'])
```

This model is specifically optimized for legal language tasks in the Indian context. It shows improved understanding of Indian legal terminology and concepts compared to the base model.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)