Apropos
Learning algorithms for production language model programs.
Spinning Up - Usage
Get started by running
pip install apropos-ai
Hello World
To get started with simple bootstrapped fewshot random search on the Hendryks Math benchmark, run:
python examples/hello_world.py
Custom Dataset + Prompt
To get started with a custom dataset and/or language model program, if your program happens to be a single-prompt program, we have a simple demo here
Simply replace
messages_examples, gold_outputs, custom_math_metric = (
get_hendryks_messages_and_gold_outputs()
)
with your own data (in the form of system/user prompt pairs and possibly gold outputs) and a metric of your choosing. Then, run away!
Nota Bene: the logic involved in converting this data to the appropriate DAG / benchmark is very new and experimental, please file an issue if you run into any trouble.
Spinning Up - Dev
1. UV
uv venv apropos-dev source apropos-dev/bin/activate
1. Pyenv
/bash
pyenv install 3.11.0
pyenv virtualenv 3.11.0 apropos-dev
pyenv activate apropos-dev
2. Poetry
/bash
curl -sSL https://install.python-poetry.org | python3 -
poetry install
Usage
If you use the following idea(s):
- Optimizing over variations of specific substrings within a prompt
please cite this repo and forthcoming paper when released
If you use the MIPROv2 optimizer in your academic work, please cite the following paper:
@misc{opsahlong2024optimizinginstructionsdemonstrationsmultistage,
title={Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs},
author={Krista Opsahl-Ong and Michael J Ryan and Josh Purtell and David Broman and Christopher Potts and Matei Zaharia and Omar Khattab},
year={2024},
eprint={2406.11695},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2406.11695},
}
Moreover, if you find the notion of learning from bootstrapped demonstrations useful, or have used algorithms such as the breadth-first random search optimizer, consider citing the following paper
@misc{khattab2023dspycompilingdeclarativelanguage,
title={DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines},
author={Omar Khattab and Arnav Singhvi and Paridhi Maheshwari and Zhiyuan Zhang and Keshav Santhanam and Sri Vardhamanan and Saiful Haq and Ashutosh Sharma and Thomas T. Joshi and Hanna Moazam and Heather Miller and Matei Zaharia and Christopher Potts},
year={2023},
eprint={2310.03714},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2310.03714},
}
1 maintainer
Release files for apropos-ai 0.1.13
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| apropos_ai-0.1.13.tar.gz | 3.6 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| apropos_ai-0.1.13-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 6.1 MB
Release files / apropos_ai-0.1.13.tar.gz
| Download URL | apropos_ai-0.1.13.tar.gz |
|---|---|
| Size | 3.6 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
d748b39050b6377003dc516137aa446aa5902ab665302dd0ee070bd2f762c454
|
|
BLAKE2b-256 checksum How to use checksums |
d6d9b83ad2a453197fb31942091271e4cd2b4dd336362b35ccb6c6786add15d3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/5.1.1 CPython/3.12.5
|
Release files / apropos_ai-0.1.13-py3-none-any.whl
| Download URL | apropos_ai-0.1.13-py3-none-any.whl |
|---|---|
| Size | 2.5 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
3b1ecc24cc56c7791f4c72138fa2eb2a0e2ce486522d369d34443ee5e6c513d4
|
|
BLAKE2b-256 checksum How to use checksums |
dc30bf29151480aa36e679717e3070b4792d728290ceeee6d374c30998af7d38
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/5.1.1 CPython/3.12.5
|
