bharat-tiny-llm 0.1.1

Bharat-Tiny-LLM - India's first native edge AI for Hinglish and Hindi, running on-device on ₹8,000 phones.

🇮🇳 Bharat-Tiny-LLM

India's first native edge AI for Hinglish & Hindi — running fully offline on ₹8,000 phones.

bharat-tiny-llm is the official Python package for Bharat-Tiny-LLM: a 1.5B-parameter, LoRA-fine-tuned language model that speaks fluent Hinglish (Romanized Hindi) and Devanagari Hindi, and runs entirely on-device — no API, no cloud, no internet.

Built by eulogik · 🤗 Model: eulogik/Bharat-Tiny-LLM · 🚀 Live demo: spaces/eulogik/Bharat-Tiny-LLM · 💻 Code: github.com/eulogik/Bharat-Tiny-LLM

---

Why Bharat-Tiny-LLM?

• 🌐 Truly bilingual — Hinglish and Devanagari Hindi in one model (most Indic models do only one).
• 📱 Edge-native — 880 MB 4-bit build runs offline on ₹8,000 Android phones & Apple Silicon.
• 🆓 Open & free — Apache-2.0 weights, no vendor lock-in, self-hostable.
• 💸 $0 training cost — fine-tuned on a Mac Mini M4, zero cloud compute.
• 🪶 Small — 1.5B params, ~57 tok/s on a Mac Mini M4.

Install

# Apple Silicon (recommended — MLX 4-bit, fastest)
pip install bharat-tiny-llm[mlx]

# Other platforms (CPU / CUDA, transformers backend)
pip install bharat-tiny-llm[torch]

Quick start

from bharat_tiny_llm import chat

reply = chat([
    {"role": "user", "content": "Chai peete hain?"},
])
print(reply)

The package ships with a canonical generation config (temperature=0.3, top_p=0.85, repetition_penalty=1.25, no_repeat_ngram_size=3) so output is clean out of the box — no garbled scripts, no degenerate loops. You normally never have to tune these.

Apple Silicon, low-level MLX

from bharat_tiny_llm import load
from mlx_lm import generate

model, tokenizer = load()  # pulls eulogik/Bharat-Tiny-LLM (MLX 4-bit)
prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "Biryani kaise banate hain?"}],
    tokenize=False, add_generation_prompt=True,
)
print(generate(model, tokenizer, prompt=prompt, max_tokens=128))

Model variants

Repo	Format	Size	Use
eulogik/Bharat-Tiny-LLM	MLX 4-bit	~880 MB	Edge / Apple Silicon (default)
eulogik/Bharat-Tiny-LLM-GGUF	GGUF Q4_K_M	~1.06 GB	Cross-platform (llama.cpp, Android, Pi, CPU)
eulogik/Bharat-Tiny-LLM-fused	PyTorch fp16	~3.3 GB	Server / fine-tuning base

Links

• 🤗 Model card: https://huggingface.co/eulogik/Bharat-Tiny-LLM
• 🤗 GGUF (llama.cpp): https://huggingface.co/eulogik/Bharat-Tiny-LLM-GGUF
• 🤗 fp16 fused: https://huggingface.co/eulogik/Bharat-Tiny-LLM-fused
• 🚀 Demo space: https://huggingface.co/spaces/eulogik/Bharat-Tiny-LLM
• 💻 Source: https://github.com/eulogik/Bharat-Tiny-LLM
• 🏢 Built by eulogik

License

Apache-2.0 (base Qwen2.5-1.5B weights Apache-2.0; LoRA adapter Apache-2.0).