license: mit

Mistral Malayalam Colloquial Model

Overview

Model Details

• Base Model: Mistral 7B (fine-tuned)
• Training Data: Malayalam colloquial dataset
• Fine-Tuning: The model has been fine-tuned specifically for generating colloquial Malayalam, which includes regional variations and informal expressions.
• Purpose: Text generation in colloquial Malayalam for applications like chatbots, virtual assistants, content generation, etc.

Features

• Text Generation: Use the model to generate Malayalam text based on a prompt.
• Customizable Parameters: Set parameters like max_length, temperature, and top_p to control the style and length of the output.
• Optimized for Performance: The model is quantized to run efficiently even on limited hardware, while still providing high-quality outputs.

Installation

Usage

1. Load the model:
   Load the fine-tuned Mistral model with the following code:
   python

   1from transformers import AutoModelForCausalLM, AutoTokenizer 2 3# Load the fine-tuned model and tokenizer 4model = AutoModelForCausalLM.from_pretrained("your_hf_username/mistral-malayalam-colloquial") 5tokenizer = AutoTokenizer.from_pretrained("your_hf_username/mistral-malayalam-colloquial")
2. Text Generation:
   Use the model to generate Malayalam colloquial text:
   python

   1from transformers import pipeline 2 3# Create the text generation pipeline 4pipe = pipeline("text-generation", model=model, tokenizer=tokenizer) 5 6# Test with a prompt 7input_text = "എന്റെ പേര് അരുൺ ആണ്. ഞാൻ കേരളത്തിൽ താമസിക്കുന്നു." 8output = pipe(input_text, max_length=50, do_sample=True) 9 10print(output[0]['generated_text'])
   This will output the generated text in colloquial Malayalam based on the provided input.

Fine-Tuning Details

• The model was fine-tuned using a Malayalam colloquial dataset, which includes informal and conversational Malayalam text.
• Fine-tuning process: The model was trained using LoRA and quantization techniques to improve efficiency without compromising too much on accuracy.

Limitations

• The model may still have some inconsistencies in output, as fine-tuning large language models is an evolving process.
• It is optimized for colloquial Malayalam but may not perform well on formal text.

License