5a743d1dc4
git-subtree-dir: apps/core git-subtree-mainline: a8963e11e7a5a0059acbc849ce768e1eee80df61 git-subtree-split: ea2a7395f4023f5b9fab03e6273db3b64a1181d5
83 lines
2.7 KiB
YAML
83 lines
2.7 KiB
YAML
name: OpenAI GPT-4 Capabilities Test
|
|
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
COMPOSIO_API_KEY: ${{ secrets.COMPOSIO_API_KEY }}
|
|
|
|
on:
|
|
push:
|
|
branches: [ main ]
|
|
pull_request:
|
|
branches: [ main ]
|
|
|
|
jobs:
|
|
test:
|
|
runs-on: ubuntu-latest
|
|
timeout-minutes: 15
|
|
steps:
|
|
- name: Checkout
|
|
uses: actions/checkout@v4
|
|
|
|
- name: "Setup Python, Poetry and Dependencies"
|
|
uses: packetcoders/action-setup-cache-python-poetry@main
|
|
with:
|
|
python-version: "3.12"
|
|
poetry-version: "1.8.2"
|
|
install-args: "-E dev -E external-tools"
|
|
|
|
- name: Test first message contains expected function call and inner monologue
|
|
id: test_first_message
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_returns_valid_first_message
|
|
|
|
- name: Test model sends message with keyword
|
|
id: test_keyword_message
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_returns_keyword
|
|
|
|
- name: Test model uses external tool correctly
|
|
id: test_external_tool
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_uses_external_tool
|
|
|
|
- name: Test model recalls chat memory
|
|
id: test_chat_memory
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_recall_chat_memory
|
|
|
|
- name: Test model uses 'archival_memory_search' to find secret
|
|
id: test_archival_memory_search
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_archival_memory_retrieval
|
|
|
|
- name: Test model uses 'archival_memory_insert' to insert archival memories
|
|
id: test_archival_memory_insert
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_archival_memory_insert
|
|
|
|
- name: Test model can edit core memories
|
|
id: test_core_memory
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_openai_gpt_4o_edit_core_memory
|
|
|
|
- name: Test embedding endpoint
|
|
id: test_embedding_endpoint
|
|
env:
|
|
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
|
|
run: |
|
|
poetry run pytest -s -vv tests/test_model_letta_perfomance.py::test_embedding_endpoint_openai
|