burmesegpt 0.1.8

BurmeseGPT is a Burmese-first AI ecosystem package, including Padauk, an agentic small language model built for tool use, function calling, and local deployment.

BurmeseGPT

BurmeseGPT is an open Burmese-first AI ecosystem created by Dr. Wai Yan Nyein Naing. This Python package is the main package for BurmeseGPT projects, including Padauk, a practical agentic small language model built for Burmese language understanding, tool use, function calling, and local or edge deployment.

Why BurmeseGPT

Burmese is still a low-resource language in AI, and practical Burmese AI tooling remains limited for many real developer workflows. BurmeseGPT focuses on useful, production-minded, and developer-friendly Burmese AI systems that can run locally and integrate with modern agent and API patterns.

Related projects

• Padauk - Burmese-first agentic language model.
• Burmese-Coder - Burmese coding and technical AI direction.
• Future tools and integrations under the BurmeseGPT ecosystem.

End-User Install And Use

This section is the user flow only: install, download, run, and test the local API.

Requires Python >=3.10.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install "burmesegpt[full]"

If your default python3 is below 3.10, use an explicit interpreter:

/opt/homebrew/bin/python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install "burmesegpt[full]"

See docs/guides/pypi-install-quickstart.md for the same copy-paste flow.

Package extras remain:

• burmesegpt[server] installs the packaged llama-cpp-python server runtime.
• burmesegpt[openai] installs the OpenAI client integration only.
• burmesegpt[full] installs both server and OpenAI extras.
• burmesegpt[dev] adds maintainer tooling.

Gemma 4 caveat: the current PyPI llama-cpp-python dependency should not be treated as sufficient for Gemma 4 shared-KV serving or release validation. burmesegpt serve and burmesegpt run now support --backend auto|python|standalone. auto is the default: it keeps the packaged Python server for the normal Padauk flow, but it will prefer standalone llama-server for Gemma 4 shared-KV GGUFs when one is available. Standalone discovery order is --llama-server-path, then BURMESEGPT_LLAMA_SERVER, then llama-server on PATH. Use --backend standalone when you want a hard failure instead of silently continuing on the Python backend. For release work, use a standalone llama.cpp source build pinned to commit b8833, with b8751 as the minimum supported floor.

Download Model And Run API

burmesegpt download --quant q8_0
burmesegpt serve --quant q8_0 --host 127.0.0.1 --port 8000

Keep that terminal running while you test requests.

If you have a standalone llama-server binary installed, you can force that backend explicitly:

export BURMESEGPT_LLAMA_SERVER="$HOME/llama.cpp/build/bin/llama-server"
burmesegpt serve --quant q8_0 --backend standalone --host 127.0.0.1 --port 8000

The same backend flags also work with the interactive chat helper:

burmesegpt run --quant q8_0 --backend standalone --llama-server-path "$HOME/llama.cpp/build/bin/llama-server"

Test With API Call (curl)

curl http://127.0.0.1:8000/v1/models

curl http://127.0.0.1:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-no-key-required" \
  -d '{
    "model": "padauk-agent",
    "messages": [{"role": "user", "content": "မင်္ဂလာပါ"}]
  }'

Test With OpenAI Agents SDK

python -m pip install openai-agents

import asyncio

from openai import AsyncOpenAI
from agents import Agent, Runner, set_default_openai_api, set_default_openai_client, set_tracing_disabled

set_tracing_disabled(True)
set_default_openai_api("chat_completions")
set_default_openai_client(
    AsyncOpenAI(base_url="http://127.0.0.1:8000/v1", api_key="sk-no-key-required")
)

padauk = Agent(
    name="Padauk Assistant",
    model="padauk-agent",
    instructions="Reply in Burmese by default.",
)

async def main() -> None:
    result = await Runner.run(padauk, input="မင်္ဂလာပါ")
    print(result.final_output)

asyncio.run(main())

Maintainer Release Validation

Use this maintainer-only flow before any package or GGUF release:

• docs/guides/pypi-release-guide.md
• docs/guides/hf-to-gguf-quantization-guide.md
• docs/guides/gemma4-shared-kv-rca.md

Links

• Website: https://www.waiyannyeinnaing.com/
• Padauk: https://www.waiyannyeinnaing.com/projects/padauk
• Source: https://github.com/WaiYanNyeinNaing/burmese-gpt-pypi-api
• Hugging Face source repo: https://huggingface.co/WYNN747/Burmese-GPT-Padauk
• Hugging Face GGUF repo: https://huggingface.co/WYNN747/padauk-burmese-agentic-llm