Merge "Gemini" feature into the main branch (microsoft#2360)

* Start Gemini integration: works ok with Text now

* Gemini notebook lint

* try catch "import" for Gemini

* Debug: id issue for chat completion in Gemini

* Add RAG example

* Update docs for RAG

* Fix missing pydash

* Remove temp folder

* Fix test error in runs/7206014032/job/19630042864

* Fix tqdm warning

* Fix notebook output

* Gemini's vision model is supported now

* Install instructions for the Gemini branch

* Catch and retry when see Interval Server Error 500

* Allow gemini to take more flexible messages
i.e., it can take messages where "user" is not the last role.

* Use int time for Gemini client

* Handle other exceptions in gemini call

* rename to "create" function for gemini

* GeminiClient compatible with ModelClient now

* Lint

* Update instructions in Gemini notebook

* Lint

* Remove empty blocks from Gemini notebook

* Add gemini into example page

* self.create instead of call

* Add py and Py into python execution

* Remove error code from merging

* Remove pydash dependency for gemini

* Add cloud-gemini doc

* Remove temp file

* cache import update

* Add test case for summary with mm input

* Lint: warnings instead of print

* Add test cases for gemini

* Gemini test config

* Disable default model for gemini

* Typo fix in gemini workflow

* Correct grammar in example notebook

* Raise if "model" is not provided in create(...)

* Move TODOs into a roadmap

* Update .github/workflows/contrib-tests.yml

Co-authored-by: Davor Runje <[email protected]>

* Gemini test config update

* Update setup.py

Co-authored-by: Davor Runje <[email protected]>

* Update test/oai/test_gemini.py

Co-authored-by: Davor Runje <[email protected]>

* Update test/oai/test_gemini.py

Co-authored-by: Davor Runje <[email protected]>

* Remove python 3.8 from gemini
No google's generativeai for Windows with Python 3.8

* Update import error handling for gemini

* Count tokens and cost for gemini

---------

Co-authored-by: Li Jiang <[email protected]>
Co-authored-by: Davor Runje <[email protected]>

.github/workflows/contrib-tests.yml

@@ -256,6 +256,44 @@ jobs:
           file: ./coverage.xml
           flags: unittests
 
+  GeminiTest:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-2019]
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          lfs: true
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install packages and dependencies for all tests
+        run: |
+          python -m pip install --upgrade pip wheel
+          pip install pytest
+      - name: Install packages and dependencies for Gemini
+        run: |
+          pip install -e .[gemini,test]
+      - name: Set AUTOGEN_USE_DOCKER based on OS
+        shell: bash
+        run: |
+          if [[ ${{ matrix.os }} != ubuntu-latest ]]; then
+            echo "AUTOGEN_USE_DOCKER=False" >> $GITHUB_ENV
+          fi
+      - name: Coverage
+        run: |
+          coverage run -a -m pytest test/oai/test_gemini.py --skip-openai
+          coverage xml
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v3
+        with:
+          file: ./coverage.xml
+          flags: unittests
+
   ContextHandling:
     runs-on: ${{ matrix.os }}
     strategy:

.gitignore

@@ -172,6 +172,10 @@ test/my_tmp/*
 # Storage for the AgentEval output
 test/test_files/agenteval-in-out/out/
 
+# local cache or coding foler
+local_cache/
+coding/
+
 # Files created by tests
 *tmp_code_*
 test/agentchat/test_agent_scripts/*

.pre-commit-config.yaml

@@ -43,6 +43,7 @@ repos:
               website/static/img/ag.svg |
               website/yarn.lock |
               website/docs/tutorial/code-executors.ipynb |
+              website/docs/topics/non-openai-models/cloud-gemini.ipynb |
               notebook/.*
             )$
   # See https://jaredkhan.com/blog/mypy-pre-commit

autogen/agentchat/conversable_agent.py

@@ -1121,7 +1121,15 @@ def my_summary_method(
     def _last_msg_as_summary(sender, recipient, summary_args) -> str:
         """Get a chat summary from the last message of the recipient."""
         try:
-            summary = recipient.last_message(sender)["content"].replace("TERMINATE", "")
+            content = recipient.last_message(sender)["content"]
+            if isinstance(content, str):
+                summary = content.replace("TERMINATE", "")
+            elif isinstance(content, list):
+                # Remove the `TERMINATE` word in the content list.
+                summary = [
+                    {**x, "text": x["text"].replace("TERMINATE", "")} if isinstance(x, dict) and "text" in x else x
+                    for x in content
+                ]
         except (IndexError, AttributeError) as e:
             warnings.warn(f"Cannot extract summary using last_msg: {e}. Using an empty str as summary.", UserWarning)
             summary = ""

autogen/oai/client.py

@@ -42,6 +42,13 @@
         TOOL_ENABLED = True
     ERROR = None
 
+try:
+    from autogen.oai.gemini import GeminiClient
+
+    gemini_import_exception: Optional[ImportError] = None
+except ImportError as e:
+    gemini_import_exception = e
+
 logger = logging.getLogger(__name__)
 if not logger.handlers:
     # Add the console handler.
@@ -425,6 +432,10 @@ def _register_default_client(self, config: Dict[str, Any], openai_config: Dict[s
                 self._configure_azure_openai(config, openai_config)
                 client = AzureOpenAI(**openai_config)
                 self._clients.append(OpenAIClient(client))
+            elif api_type is not None and api_type.startswith("google"):
+                if gemini_import_exception:
+                    raise ImportError("Please install `google-generativeai` to use Google OpenAI API.")
+                self._clients.append(GeminiClient(**openai_config))
             else:
                 client = OpenAI(**openai_config)
                 self._clients.append(OpenAIClient(client))