- Change default `SMART_LLM` from `gpt-4` to `gpt-4-turbo-preview`
- Change default `FAST_LLM` from `gpt-3.5-turbo-16k` to `gpt-3.5-turbo-0125`
- Change default `EMBEDDING_MODEL` from `text-embedding-ada-002` to `text-embedding-3-small`
- Update .env.template, azure.yaml.template, and documentation accordingly
- Add `text-embedding-3-small` and `text-embedding-3-large` as `EMBEDDING_v3_S` and `EMBEDDING_v3_L` respectively
- Add `gpt-3.5-turbo-0125` as `GPT3_v4`
- Add `gpt-4-1106-vision-preview` as `GPT4_v3_VISION`
- Add GPT-4V models to info map
- Change chat model info mapping to derive info for aliases (e.g. `gpt-3.5-turbo`) from specific versions instead of the other way around
* Add `_sideload_chrome_extensions` subroutine to `open_page_in_browser` in web_selenium.py
* Sideloads uBlock Origin and I Still Don't Care About Cookies, downloading them if necessary
* Add 2-second delay to `open_page_in_browser` to allow time for handling cookie walls
Commit 956cdc7 "fix(agent/json_utils): Decode as JSON rather than Python objects" broke these unit tests because they generated "JSON" by stringifying a Python object.
* Compress steps in the prompt to reduce token usage, and to increase longevity when using models with limited context windows
* Move multiple copies of step formatting code to `Episode.format` method
* Add `EpisodicActionHistory.handle_compression` method to handle compression of new steps
* Implement `extract_information` function in `autogpt.processing.text` module. This function extracts pieces of information from a body of text based on a list of topics of interest.
* Add `topics_of_interest` and `get_raw_content` parameters to `read_webpage` commmand
* Limit maximum content length if `get_raw_content=true` is specified
* Replace `ast.literal_eval` with `json.loads` in `extract_dict_from_response`
This fixes a bug where boolean values could not be decoded because of their required capitalization in Python.
LLMs are probabilistic systems. Reproducibility of completions is not guaranteed. It only makes sense to account for this, by running challenges multiple times to obtain a success ratio rather than a boolean success/failure result.
Changes:
- Add `-N`, `--attempts` option to CLI and `attempts_per_challenge` parameter to `main.py:run_benchmark`.
- Add dynamic `i_attempt` fixture through `pytest_generate_tests` hook in conftest.py to achieve multiple runs per challenge.
- Modify `pytest_runtest_makereport` hook in conftest.py to handle multiple reporting calls per challenge.
- Refactor report_types.py, reports.py, process_report.ty to allow multiple results per challenge.
- Calculate `success_percentage` from results of the current run, rather than all known results ever.
- Add docstrings to a number of models in report_types.py.
- Allow `None` as a success value, e.g. for runs that did not render any results before being cut off.
- Make SingletonReportManager thread-safe.
* feat(benchmark): Add JungleGym WebArena challenges
- Add `WebArenaChallenge`, `WebArenaChallengeSpec`, and other logic to make these challenges work
- Add WebArena challenges to Pytest collection endpoint generate_test.py
* feat(benchmark/webarena): Add hand-picked selection of WebArena challenges
- Pydantic shallow-copies models when they are passed into a parent model, meaning they can't be updated through the original reference. This commit adds a fix for the resulting cost persistence issue.
- The `extract_dict_from_response` function, which is supposed to reliably extract a JSON object from an LLM's response, positively discriminated objects defined on a single line, causing issues.
- `summarize_text` and `QueryLanguageModel.__call__` still tried to access `response["content"]`, which isn't possible since upgrading to the OpenAI v1 client library.
- When an Artifact's file is modified by the agent, set its `agent_created` attribute to `True` instead of registering a new Artifact
- Update the `autogpt-forge` dependency to the newest version, in which `AgentDB.update_artifact` has been implemented
Squashed commit of the following:
commit 7d6476d329
Author: Reinier van der Leer <pwuts@agpt.co>
Date: Tue Jan 9 18:10:45 2024 +0100
refactor(benchmark/challenge): Set up structure to support more challenge providers
- Move `Challenge`, `ChallengeData`, `load_challenges` to `challenges/builtin.py` and rename to `BuiltinChallenge`, `BuiltinChallengeSpec`, `load_builtin_challenges`
- Create `BaseChallenge` to serve as interface and base class for different challenge implementations
- Create `ChallengeInfo` model to serve as universal challenge info object
- Create `get_challenge_from_source_uri` function in `challenges/__init__.py`
- Replace `ChallengeData` by `ChallengeInfo` everywhere except in `BuiltinChallenge`
- Add strong typing to `task_informations` store in app.py
- Use `call.duration` in `finalize_test_report` and remove `timer` fixture
- Update docstring on `challenges/__init__.py:get_unique_categories`
- Add docstring to `generate_test.py`
commit 5df2aa7939
Author: Reinier van der Leer <pwuts@agpt.co>
Date: Tue Jan 9 16:58:01 2024 +0100
refactor(benchmark): Refactor & rename functions in agent_interface.py and agent_api_interface.py
- `copy_artifacts_into_temp_folder` -> `copy_challenge_artifacts_into_workspace`
- `copy_agent_artifacts_into_folder` -> `download_agent_artifacts_into_folder`
- Reorder parameters of `run_api_agent`, `copy_challenge_artifacts_into_workspace`; use `Path` instead of `str`
commit 6a256fef4c
Author: Reinier van der Leer <pwuts@agpt.co>
Date: Tue Jan 9 16:02:25 2024 +0100
refactor(benchmark): Refactor & typefix report generation and handling logic
- Rename functions in reports.py and ReportManager.py to better reflect what they do
- `get_previous_test_results` -> `get_and_update_success_history`
- `generate_single_call_report` -> `initialize_test_report`
- `finalize_reports` -> `finalize_test_report`
- `ReportManager.end_info_report` -> `SessionReportManager.finalize_session_report`
- Modify `pytest_runtest_makereport` hook in conftest.py to finalize the report immediately after the challenge finishes running instead of after teardown
- Move result processing logic from `initialize_test_report` to `finalize_test_report` in reports.py
- Use `Test` and `Report` types from report_types.py where possible instead of untyped dicts: reports.py, utils.py, ReportManager.py
- Differentiate `ReportManager` into `SessionReportManager`, `RegressionTestsTracker`, `SuccessRateTracker`
- Move filtering of optional challenge categories from challenge.py (`Challenge.skip_optional_categories`) to conftest.py (`pytest_collection_modifyitems`)
- Remove unused `scores` fixture in conftest.py
commit 370d6dbf5d
Author: Reinier van der Leer <pwuts@agpt.co>
Date: Tue Jan 9 15:16:43 2024 +0100
refactor(benchmark): Simplify models in report_types.py
- Removed ForbidOptionalMeta and BaseModelBenchmark classes.
- Changed model attributes to optional: `Metrics.difficulty`, `Metrics.success`, `Metrics.success_percentage`, `Metrics.run_time`, and `Test.reached_cutoff`.
- Added validator to `Metrics` model to require `success` and `run_time` fields if `attempted=True`.
- Added default values to all optional model fields.
- Removed duplicate imports.
- Added condition in process_report.py to prevent null lookups if `metrics.difficulty` is not set.
- Update `openai` dependency from ^v0.27.10 to ^v1.7.2
- Update poetry.lock
- Update code for changed endpoints and new output types of OpenAI library
- Replace uses of `AssistantChatMessageDict` by `AssistantChatMessage`
- Update `PromptStrategy`, `BaseAgent`, and all of their subclasses accordingly
- Update `OpenAIProvider`, `OpenAICredentials`, azure.yaml.template, .env.template and test_config.py to work with new separate `AzureOpenAI` client
- Remove `_OpenAIRetryHandler` and implement retry mechanism with `tenacity`
- Rewrite pytest fixture `cached_openai_client` (renamed from `patched_api_requestor`) for OpenAI v1 library
* Update `openai` dependency from `^0.27.8` to `^1.7.2`
* Update `litellm` dependency from `^0.1.821` to `^1.17.9`
* Migrate llm.py from OpenAI module-level client to client instance
* Update return types in llm.py for new OpenAI and LiteLLM versions
* Also remove `Exception` as a return type because they are raised, not returned
* Update tutorials/003_crafting_agent_logic.md accordingly
Note: this changes the output types of the functions in `forge.llm`: `chat_completion_request`, `create_embedding_request`, `transcribe_audio`
* docs: Add documentation on how to use Agent Protocol in the template
- Added documentation on how to use Agent Protocol and its settings in the `.env.template` file.
- An explanation is provided for the `AP_SERVER_PORT` and `AP_SERVER_DB_URL` settings.
- This change aims to improve the understanding and usage of Agent Protocol in the project.
* docs: Update usage.md with information about configuring the API port
- Update the documentation for the `serve` mode in `usage.md`
- Add information about configuring the port for the API server using the `AP_SERVER_PORT` environment variable.
---------
Co-authored-by: Reinier van der Leer <pwuts@agpt.co>
When executing `port = os.getenv("PORT", 8000)` if the port is being fetched from a `.env` file it is fetched as a string.
This caused an error: `TypeError: 'str' object cannot be interpreted as an integer`
* refactor(benchmark): Deduplicate configuration loading logic
- Move the configuration loading logic to a separate `load_agbenchmark_config` function in `agbenchmark/config.py` module.
- Replace the duplicate loading logic in `conftest.py`, `generate_test.py`, `ReportManager.py`, `reports.py`, and `__main__.py` with calls to `load_agbenchmark_config` function.
* fix(benchmark): Fix type errors, linting errors, and clean up CLI validation in __main__.py
- Fixed type errors and linting errors in `__main__.py`
- Improved the readability of CLI argument validation by introducing a separate function for it
* refactor(benchmark): Lint and typefix app.py
- Rearranged and cleaned up import statements
- Fixed type errors caused by improper use of `psutil` objects
- Simplified a number of `os.path` usages by converting to `pathlib`
- Use `Task` and `TaskRequestBody` classes from `agent_protocol_client` instead of `.schema`
* refactor(benchmark): Replace `.agent_protocol_client` by `agent-protcol-client`, clean up schema.py
- Remove `agbenchmark.agent_protocol_client` (an offline copy of `agent-protocol-client`).
- Add `agent-protocol-client` as a dependency and change imports to `agent_protocol_client`.
- Fix type annotation on `agent_api_interface.py::upload_artifacts` (`ApiClient` -> `AgentApi`).
- Remove all unused types from schema.py (= most of them).
* refactor(benchmark): Use pathlib in agent_interface.py and agent_api_interface.py
* refactor(benchmark): Improve typing, response validation, and readability in app.py
- Simplified response generation by leveraging type checking and conversion by FastAPI.
- Introduced use of `HTTPException` for error responses.
- Improved naming, formatting, and typing in `app.py::create_evaluation`.
- Updated the docstring on `app.py::create_agent_task`.
- Fixed return type annotations of `create_single_test` and `create_challenge` in generate_test.py.
- Added default values to optional attributes on models in report_types_v2.py.
- Removed unused imports in `generate_test.py`
* refactor(benchmark): Clean up logging and print statements
- Introduced use of the `logging` library for unified logging and better readability.
- Converted most print statements to use `logger.debug`, `logger.warning`, and `logger.error`.
- Improved descriptiveness of log statements.
- Removed unnecessary print statements.
- Added log statements to unspecific and non-verbose `except` blocks.
- Added `--debug` flag, which sets the log level to `DEBUG` and enables a more comprehensive log format.
- Added `.utils.logging` module with `configure_logging` function to easily configure the logging library.
- Converted raw escape sequences in `.utils.challenge` to use `colorama`.
- Renamed `generate_test.py::generate_tests` to `load_challenges`.
* refactor(benchmark): Remove unused server.py and agent_interface.py::run_agent
- Remove unused server.py file
- Remove unused run_agent function from agent_interface.py
* refactor(benchmark): Clean up conftest.py
- Fix and add type annotations
- Rewrite docstrings
- Disable or remove unused code
- Fix definition of arguments and their types in `pytest_addoption`
* refactor(benchmark): Clean up generate_test.py file
- Refactored the `create_single_test` function for clarity and readability
- Removed unused variables
- Made creation of `Challenge` subclasses more straightforward
- Made bare `except` more specific
- Renamed `Challenge.setup_challenge` method to `run_challenge`
- Updated type hints and annotations
- Made minor code/readability improvements in `load_challenges`
- Added a helper function `_add_challenge_to_module` for attaching a Challenge class to the current module
* fix(benchmark): Fix and add type annotations in execute_sub_process.py
* refactor(benchmark): Simplify const determination in agent_interface.py
- Simplify the logic that determines the value of `HELICONE_GRAPHQL_LOGS`
* fix(benchmark): Register category markers to prevent warnings
- Use the `pytest_configure` hook to register the known challenge categories as markers. Otherwise, Pytest will raise "unknown marker" warnings at runtime.
* refactor(benchmark/challenges): Fix indentation in 4_revenue_retrieval_2/data.json
* refactor(benchmark): Update agent_api_interface.py
- Add type annotations to `copy_agent_artifacts_into_temp_folder` function
- Add note about broken endpoint in the `agent_protocol_client` library
- Remove unused variable in `run_api_agent` function
- Improve readability and resolve linting error
* feat(benchmark): Improve and centralize pathfinding
- Search path hierarchy for applicable `agbenchmark_config`, rather than assuming it's in the current folder.
- Create `agbenchmark.utils.path_manager` with `AGBenchmarkPathManager` and exporting a `PATH_MANAGER` const.
- Replace path constants defined in __main__.py with usages of `PATH_MANAGER`.
* feat(benchmark/cli): Clean up and improve CLI
- Updated commands, options, and their descriptions to be more intuitive and consistent
- Moved slow imports into the entrypoints that use them to speed up application startup
- Fixed type hints to match output types of Click options
- Hid deprecated `agbenchmark start` command
- Refactored code to improve readability and maintainability
- Moved main entrypoint into `run` subcommand
- Fixed `version` and `serve` subcommands
- Added `click-default-group` package to allow using `run` implicitly (for backwards compatibility)
- Renamed `--no_dep` to `--no-dep` for consistency
- Fixed string formatting issues in log statements
* refactor(benchmark/config): Move AgentBenchmarkConfig and related functions to config.py
- Move the `AgentBenchmarkConfig` class from `utils/data_types.py` to `config.py`.
- Extract the `calculate_info_test_path` function from `utils/data_types.py` and move it to `config.py` as a private helper function `_calculate_info_test_path`.
- Move `load_agent_benchmark_config()` to `AgentBenchmarkConfig.load()`.
- Changed simple getter methods on `AgentBenchmarkConfig` to calculated properties.
- Update all code references according to the changes mentioned above.
* refactor(benchmark): Fix ReportManager init parameter types and use pathlib
- Fix the type annotation of the `benchmark_start_time` parameter in `ReportManager.__init__`, was mistyped as `str` instead of `datetime`.
- Change the type of the `filename` parameter in the `ReportManager.__init__` method from `str` to `Path`.
- Rename `self.filename` with `self.report_file` in `ReportManager`.
- Change the way the report file is created, opened and saved to use the `Path` object.
* refactor(benchmark): Improve typing surrounding ChallengeData and clean up its implementation
- Use `ChallengeData` objects instead of untyped `dict` in app.py, generate_test.py, reports.py.
- Remove unnecessary methods `serialize`, `get_data`, `get_json_from_path`, `deserialize` from `ChallengeData` class.
- Remove unused methods `challenge_from_datum` and `challenge_from_test_data` from `ChallengeData class.
- Update function signatures and annotations of `create_challenge` and `generate_single_test` functions in generate_test.py.
- Add types to function signatures of `generate_single_call_report` and `finalize_reports` in reports.py.
- Remove unnecessary `challenge_data` parameter (in generate_test.py) and fixture (in conftest.py).
* refactor(benchmark): Clean up generate_test.py, conftest.py and __main__.py
- Cleaned up generate_test.py and conftest.py
- Consolidated challenge creation logic in the `Challenge` class itself, most notably the new `Challenge.from_challenge_spec` method.
- Moved challenge selection logic from generate_test.py to the `pytest_collection_modifyitems` hook in conftest.py.
- Converted methods in the `Challenge` class to class methods where appropriate.
- Improved argument handling in the `run_benchmark` function in `__main__.py`.
* refactor(benchmark/config): Merge AGBenchmarkPathManager into AgentBenchmarkConfig and reduce fragmented/global state
- Merge the functionality of `AGBenchmarkPathManager` into `AgentBenchmarkConfig` to consolidate the configuration management.
- Remove the `.path_manager` module containing `AGBenchmarkPathManager`.
- Pass the `AgentBenchmarkConfig` and its attributes through function arguments to reduce global state and improve code clarity.
* feat(benchmark/serve): Configurable port for `serve` subcommand
- Added `--port` option to `serve` subcommand to allow for specifying the port to run the API on.
- If no `--port` option is provided, the port will default to the value specified in the `PORT` environment variable, or 8080 if not set.
* feat(benchmark/cli): Add `config` subcommand
- Added a new subcommand `config` to the AGBenchmark CLI, to display information about the present AGBenchmark config.
* fix(benchmark): Gracefully handle incompatible challenge spec files in app.py
- Added a check to skip deprecated challenges
- Added logging to allow debugging of the loading process
- Added handling of validation errors when parsing challenge spec files
- Added missing `spec_file` attribute to `ChallengeData`
* refactor(benchmark): Move `run_benchmark` entrypoint to main.py, use it in `/reports` endpoint
- Move `run_benchmark` and `validate_args` from __main__.py to main.py
- Replace agbenchmark subprocess in `app.py:run_single_test` with `run_benchmark`
- Move `get_unique_categories` from __main__.py to challenges/__init__.py
- Move `OPTIONAL_CATEGORIES` from __main__.py to challenge.py
- Reduce operations on updates.json (including `initialize_updates_file`) outside of API
* refactor(benchmark): Remove unused `/updates` endpoint and all related code
- Remove `updates_json_file` attribute from `AgentBenchmarkConfig`
- Remove `get_updates` and `_initialize_updates_file` in app.py
- Remove `append_updates_file` and `create_update_json` functions in agent_api_interface.py
- Remove call to `append_updates_file` in challenge.py
* refactor(benchmark/config): Clean up and update docstrings on `AgentBenchmarkConfig`
- Add and update docstrings
- Change base class from `BaseModel` to `BaseSettings`, allow extras for backwards compatibility
- Make naming of path attributes on `AgentBenchmarkConfig` more consistent
- Remove unused `agent_home_directory` attribute
- Remove unused `workspace` attribute
* fix(benchmark): Restore mechanism to select (optional) categories in agent benchmark config
* fix(benchmark): Update agent-protocol-client to v1.1.0
- Fixes issue with fetching task artifact listings
gitpython was installed as an indirect dependency via agbenchmark. The release builds don't contain agbenchmark and thus also lack the gitpython package, which breaks the image.
- Move `auto-gpt-plugin-template` from dev dependencies to regular dependencies in `pyproject.toml`.
- Fixes#6566 - No module named 'auto_gpt_plugin_template'.
- Add logging to capture errors raised during execution of actions in the Agent
- Move `fmt_kwargs` function from `agent_protocol_server.py` to `logs/utils.py`
- Update the `read_file` function in `file_operations.py` to pass the file's extension to the `decode_textual_file` function.
- Modify the `decode_textual_file` function in `file_operations_utils.py` to accept the file extension as an argument.
- Update the `content` property in the `FileContextItem` class in `context_item.py` to pass the file's extension to the `decode_textual_file` function.
- Update the `test_parsers` function in `test_text_file_parsers.py` to pass the file extension to the `decode_textual_file` function.
- Prevent the Agent from treating `AgentTerminated` like it would any other exception raised by a command.
- The agent should raise AgentTerminated exception to exit the loop.
- Fix the parsing of invalid LLM responses by appending an error message to the prompt and allowing the LLM to fix its mistakes.
- Update the `OpenAIProvider` to handle the self-correction process and limit the number of attempts to fix parsing errors.
- Update the `BaseAgent` to profit from the new pasing and parse-fixing mechanism.
This change ensures that the system can handle and recover from errors in parsing LLM responses.
Hopefully this fixes#1407 once and for all.
Mahdi Karami
Reinier van der Leer
Fernando Navarro Páez
Nicholas Tindle
Or Arbel
Himanshu Mittal