BharatLLM Tamil -- K-12 Education LoRA

A QLoRA adapter for Mistral-7B, fine-tuned on CBSE/NCERT K-12 curriculum data in Tamil (தமிழ்).

Part of the BharatLLM project: 13 LoRA adapters (12 K-12 languages + 1 BTech Engineering).

Model Details

Property Value
Base Model mistralai/Mistral-7B-Instruct-v0.3
Method QLoRA (4-bit quantization + LoRA, r=64)
Trainable Parameters 167,772,160 (2.26% of 7.4B)
Training Library Unsloth
Language Tamil (தமிழ்)
Domain K-12 Education (CBSE/NCERT, Grades 6-12)
Training Data ~408K curriculum-aligned Q&A pairs
License Apache 2.0

Quick Start (Unsloth -- Fastest)

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="FoundryAILabs/bharat-tamil-7b-lora",
    max_seq_length=2048,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

inputs = tokenizer("[INST] What is photosynthesis? [/INST]", return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Using with HuggingFace Transformers

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3", load_in_4bit=True, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3")
model = PeftModel.from_pretrained(base, "FoundryAILabs/bharat-tamil-7b-lora")

All BharatLLM Models

Website: foundryailabs.io | GitHub: github.com/foundryailabs/BharatLLM

Downloads last month
16
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for FoundryAILabs/bharat-tamil-7b-lora

Adapter
(873)
this model

Dataset used to train FoundryAILabs/bharat-tamil-7b-lora