simpleai-sdk 0.1.1

A simple, high-level Python package for the Lille language model.

SimpleAI

A simple, high-level API for running the Lille language model.

This package provides an easy-to-use interface for both the Hugging Face Transformers and ONNX Runtime backends, automatically downloading the required model files on first use.

Installation

pip install simpleai-sdk

A Note on ONNX Runtime and GPU Support

The package will automatically try to install the correct version of onnxruntime for your system:

• If you have a CUDA 12.x+ compatible GPU and torch is installed, it will automatically install onnxruntime-gpu.
• If you have a CPU-only system, it will install onnxruntime.

Important for CUDA 11.x Users: If you are using CUDA 11.x, pip cannot handle the special installation source automatically. The setup will default to the CPU version and print a warning. To get GPU acceleration, you must run the following command after installing simpleai-sdk:

pip install onnxruntime-gpu --index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-11/pypi/simple/

Quick Start

The lille() function is the main entry point. You can specify either the "huggingface" or "onnx" backend.

Hugging Face Backend (Default)

This is the recommended backend for flexibility and ease of use, especially on GPUs.

from simple_ai import lille

# This will download and cache the model on first run
# You can specify the model version, e.g., "130m-instruct" (default) or "130m-base"
model = lille("huggingface", "130m-instruct") 

# For text completion
prompt = "Artificial Intelligence is"
response = model.generate(prompt, max_new_tokens=50, temperature=0.9)
print(response)

# For chat
print("\n--- Chat Example ---")
response1 = model.chat("What is the capital of France?", max_new_tokens=50, top_k=200)
print(f"Bot: {response1}")

response2 = model.chat("And what is its population?", max_new_tokens=50, top_p=0.90)
print(f"Bot: {response2}")

# Reset the conversation history
model.reset_chat()

ONNX Backend

The ONNX backend is optimized for fast CPU and GPU inference.

from simple_ai import lille

# This will download and cache the model on first run
# You can specify the model version, e.g., "130m-instruct" (default) or "130m-base"
model = lille("onnx", "130m-instruct") 

# For text completion
prompt = "Artificial Intelligence is"
response = model.generate(prompt, max_new_tokens=50, temperature=0.9)
print(response)

# For chat
print("\n--- Chat Example ---")
response1 = model.chat("What is the capital of France?", max_new_tokens=50, top_k=200)
print(f"Bot: {response1}")

response2 = model.chat("And what is its population?", max_new_tokens=50, top_p=0.90)
print(f"Bot: {response2}")

# Reset the conversation history
model.reset_chat()

Advanced Usage (Direct Transformers Access)

If you use the "huggingface" backend, you can still access the underlying transformers model and tokenizer for advanced use cases, just like any standard Hugging Face model.

from simple_ai import lille

# Load the model and tokenizer
lille_hf = lille("huggingface")
model = lille_hf.model
tokenizer = lille_hf.tokenizer

# Example from Hugging Face docs
messages = [{"role": "user", "content": "What is gravity?"}]
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer.encode(input_text, return_tensors="pt").to(model.device)

outputs = model.generate(inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))