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	English
Domain	BTech Engineering (11 departments, 552 subjects)
Training Data	815,906 Q&A pairs
Difficulty Levels	Easy (369K), Medium (275K), Hard (172K)
License	Apache 2.0

Departments Covered

Code	Department	Entries	Subjects
CSE	Computer Science & Engineering	81,301	54
ME	Mechanical Engineering	79,369	52
CE	Civil Engineering	78,373	50
ECE	Electronics & Communication	76,377	50
EEE	Electrical & Electronics	76,306	49
IT	Information Technology	72,739	51
CH	Chemical Engineering	71,265	46
CSBS	CS & Business Systems	71,138	48
CSE_DS	CSE (Data Science)	70,799	50
CSE_IOT	CSE (Internet of Things)	69,237	51
CSE_AIML	CSE (AI & Machine Learning)	69,002	51
Total	11 Departments	815,906	552

Quick Start (Unsloth)

python

from unsloth import FastLanguageModel

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

prompt = "[INST] <<SYS>>\nYou are BharatLLM, an expert engineering tutor.\n<</SYS>>\n\nExplain Dijkstra's shortest path algorithm with time complexity. [/INST]"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Using with HuggingFace Transformers

python

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-btech-7b-lora")

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