trismik 0.9.12__py3-none-any.whl → 1.0.0__py3-none-any.whl

trismik/client_async.py

@@ -1,405 +0,0 @@
-"""
-Trismik async client for interacting with the Trismik API.
-
-This module provides an asynchronous client for interacting with the Trismik
-API. It uses httpx for making HTTP requests.
-"""
-
-from typing import List, Optional
-
-import httpx
-
-from trismik._mapper import TrismikResponseMapper
-from trismik._utils import TrismikUtils
-from trismik.exceptions import (
-    TrismikApiError,
-    TrismikPayloadTooLargeError,
-    TrismikValidationError,
-)
-from trismik.settings import client_settings, environment_settings
-from trismik.types import (
-    TrismikClassicEvalRequest,
-    TrismikClassicEvalResponse,
-    TrismikDataset,
-    TrismikMeResponse,
-    TrismikProject,
-    TrismikReplayRequest,
-    TrismikReplayResponse,
-    TrismikRunMetadata,
-    TrismikRunResponse,
-    TrismikRunSummary,
-)
-
-
-class TrismikAsyncClient:
-    """
-    Asynchronous client for the Trismik API.
-
-    This class provides an asynchronous interface to interact with the Trismik
-    API, handling authentication, dataset runs, and responses.
-    """
-
-    def __init__(
-        self,
-        service_url: Optional[str] = None,
-        api_key: Optional[str] = None,
-        http_client: Optional[httpx.AsyncClient] = None,
-    ) -> None:
-        """
-        Initialize the Trismik async client.
-
-        Args:
-            service_url (Optional[str]): URL of the Trismik service.
-            api_key (Optional[str]): API key for the Trismik service.
-            http_client (Optional[httpx.AsyncClient]): HTTP client to use for
-                requests.
-
-        Raises:
-            TrismikError: If service_url or api_key are not provided and not
-                found in environment.
-            TrismikApiError: If API request fails.
-        """
-        self._service_url = TrismikUtils.option(
-            service_url,
-            client_settings["endpoint"],
-            environment_settings["trismik_service_url"],
-        )
-        self._api_key = TrismikUtils.required_option(
-            api_key, "api_key", environment_settings["trismik_api_key"]
-        )
-
-        # Set default headers with API key
-        default_headers = {"x-api-key": self._api_key}
-
-        self._http_client = http_client or httpx.AsyncClient(
-            base_url=self._service_url, headers=default_headers, timeout=30.0
-        )
-
-    def _handle_http_error(self, e: httpx.HTTPStatusError) -> Exception:
-        """
-        Handle HTTP errors and return appropriate Trismik exceptions.
-
-        Args:
-            e (httpx.HTTPStatusError): The HTTP status error to handle.
-
-        Returns:
-            Exception: The appropriate Trismik exception to raise.
-        """
-        if e.response.status_code == 413:
-            # Handle payload too large error specifically
-            try:
-                backend_message = e.response.json().get(
-                    "detail", "Payload too large."
-                )
-            except Exception:
-                backend_message = "Payload too large."
-            return TrismikPayloadTooLargeError(backend_message)
-        elif e.response.status_code == 422:
-            # Handle validation error specifically
-            try:
-                backend_message = e.response.json().get(
-                    "detail", "Validation failed."
-                )
-            except Exception:
-                backend_message = "Validation failed."
-            return TrismikValidationError(backend_message)
-        else:
-            return TrismikApiError(TrismikUtils.get_error_message(e.response))
-
-    async def list_datasets(self) -> List[TrismikDataset]:
-        """
-        Get a list of available datasets.
-
-        Returns:
-            List[TrismikDataset]: List of available datasets.
-
-        Raises:
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "/datasets"
-            response = await self._http_client.get(url)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_datasets(json)
-        except httpx.HTTPStatusError as e:
-            raise TrismikApiError(
-                TrismikUtils.get_error_message(e.response)
-            ) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def start_run(
-        self,
-        dataset_id: str,
-        project_id: str,
-        experiment: str,
-        metadata: Optional[TrismikRunMetadata] = None,
-    ) -> TrismikRunResponse:
-        """
-        Start a new run for a dataset and get the first item.
-
-        Args:
-            dataset_id (str): ID of the dataset.
-            project_id (str): ID of the project.
-            experiment (str): Name of the experiment.
-            metadata (Optional[TrismikRunMetadata]): Run metadata.
-
-        Returns:
-            TrismikRunResponse: Run response.
-
-        Raises:
-            TrismikPayloadTooLargeError: If the request payload exceeds the
-            server's size limit.
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "/runs/start"
-            body = {
-                "datasetId": dataset_id,
-                "projectId": project_id,
-                "experiment": experiment,
-                "metadata": metadata.toDict() if metadata else {},
-            }
-            response = await self._http_client.post(url, json=body)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_run_response(json)
-        except httpx.HTTPStatusError as e:
-            raise self._handle_http_error(e) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def continue_run(
-        self, run_id: str, item_choice_id: str
-    ) -> TrismikRunResponse:
-        """
-        Continue a run: respond to the current item and get the next one.
-
-        Args:
-            run_id (str): ID of the run.
-            item_choice_id (str): ID of the chosen item response.
-
-        Returns:
-            TrismikRunResponse: Run response.
-
-        Raises:
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "/runs/continue"
-            body = {"itemChoiceId": item_choice_id, "runId": run_id}
-            response = await self._http_client.post(url, json=body)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_run_response(json)
-        except httpx.HTTPStatusError as e:
-            raise TrismikApiError(
-                TrismikUtils.get_error_message(e.response)
-            ) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def run_summary(self, run_id: str) -> TrismikRunSummary:
-        """
-        Get run summary including responses, dataset, and state.
-
-        Args:
-            run_id (str): ID of the run.
-
-        Returns:
-            TrismikRunSummary: Complete run summary with responses,
-                dataset, state, and metadata.
-
-        Raises:
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = f"/runs/adaptive/{run_id}"
-            response = await self._http_client.get(url)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_run_summary(json)
-        except httpx.HTTPStatusError as e:
-            raise TrismikApiError(
-                TrismikUtils.get_error_message(e.response)
-            ) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def submit_replay(
-        self,
-        run_id: str,
-        replay_request: TrismikReplayRequest,
-        metadata: Optional[TrismikRunMetadata] = None,
-    ) -> TrismikReplayResponse:
-        """
-        Submit a replay of a run with specific responses.
-
-        Args:
-            run_id (str): ID of the run to replay.
-            replay_request (TrismikReplayRequest): Request containing responses
-                to submit.
-            metadata (Optional[TrismikRunMetadata]): Run metadata.
-
-        Returns:
-            TrismikReplayResponse: Response from the replay endpoint.
-
-        Raises:
-            TrismikPayloadTooLargeError: If the request payload exceeds the
-                server's size limit.
-            TrismikValidationError: If the request fails validation (e.g.,
-                duplicate item IDs, unknown item IDs).
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = f"runs/{run_id}/replay"
-
-            # Convert TrismikReplayRequestItem objects to dictionaries
-            responses_dict = [
-                {"itemId": item.itemId, "itemChoiceId": item.itemChoiceId}
-                for item in replay_request.responses
-            ]
-
-            body = {
-                "responses": responses_dict,
-                "metadata": metadata.toDict() if metadata else {},
-            }
-            response = await self._http_client.post(url, json=body)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_replay_response(json)
-        except httpx.HTTPStatusError as e:
-            raise self._handle_http_error(e) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def me(self) -> TrismikMeResponse:
-        """
-        Get current user information.
-
-        Returns:
-            TrismikMeResponse: User information including validity and payload.
-
-        Raises:
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "../admin/api-keys/me"
-            response = await self._http_client.get(url)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_me_response(json)
-        except httpx.HTTPStatusError as e:
-            raise TrismikApiError(
-                TrismikUtils.get_error_message(e.response)
-            ) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def submit_classic_eval(
-        self, classic_eval_request: TrismikClassicEvalRequest
-    ) -> TrismikClassicEvalResponse:
-        """
-        Submit a classic evaluation run with pre-computed results.
-
-        Args:
-            classic_eval_request (TrismikClassicEvalRequest): Request containing
-                project info, dataset, model outputs, and metrics.
-
-        Returns:
-            TrismikClassicEvalResponse: Response from the classic evaluation
-                endpoint.
-
-        Raises:
-            TrismikPayloadTooLargeError: If the request payload exceeds the
-                server's size limit.
-            TrismikValidationError: If the request fails validation.
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "/runs/classic"
-
-            # Convert request object to dictionary
-            items_dict = [
-                {
-                    "datasetItemId": item.datasetItemId,
-                    "modelInput": item.modelInput,
-                    "modelOutput": item.modelOutput,
-                    "goldOutput": item.goldOutput,
-                    "metrics": item.metrics,
-                }
-                for item in classic_eval_request.items
-            ]
-
-            metrics_dict = [
-                {
-                    "metricId": metric.metricId,
-                    "valueType": TrismikUtils.metric_value_to_type(
-                        metric.value
-                    ),
-                    "value": metric.value,
-                }
-                for metric in classic_eval_request.metrics
-            ]
-
-            body = {
-                "projectId": classic_eval_request.projectId,
-                "experimentName": classic_eval_request.experimentName,
-                "datasetId": classic_eval_request.datasetId,
-                "modelName": classic_eval_request.modelName,
-                "hyperparameters": classic_eval_request.hyperparameters,
-                "items": items_dict,
-                "metrics": metrics_dict,
-            }
-
-            response = await self._http_client.post(url, json=body)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_classic_eval_response(json)
-        except httpx.HTTPStatusError as e:
-            raise self._handle_http_error(e) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e
-
-    async def create_project(
-        self,
-        name: str,
-        team_id: Optional[str] = None,
-        description: Optional[str] = None,
-    ) -> TrismikProject:
-        """
-        Create a new project.
-
-        Args:
-            name (str): Name of the project.
-            team_id (Optional[str]): ID of the team to create the
-                project in.
-            description (Optional[str]): Optional description of the project.
-
-        Returns:
-            TrismikProject: Created project information.
-
-        Raises:
-            TrismikValidationError: If the request fails validation.
-            TrismikApiError: If API request fails.
-        """
-        try:
-            url = "../admin/public/projects"
-
-            body = {
-                "name": name,
-            }
-            if team_id is not None:
-                body["teamId"] = team_id
-            if description is not None:
-                body["description"] = description
-
-            response = await self._http_client.post(url, json=body)
-            response.raise_for_status()
-            json = response.json()
-            return TrismikResponseMapper.to_project(json)
-        except httpx.HTTPStatusError as e:
-            raise self._handle_http_error(e) from e
-        except httpx.HTTPError as e:
-            raise TrismikApiError(str(e)) from e

trismik-0.9.12.dist-info/METADATA

@@ -1,177 +0,0 @@
-Metadata-Version: 2.4
-Name: trismik
-Version: 0.9.12
-Summary: 
-License-File: LICENSE
-Author: Bartosz Kielczewski
-Author-email: bk352@cam.ac.uk
-Requires-Python: >=3.9
-Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.9
-Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3.11
-Classifier: Programming Language :: Python :: 3.12
-Classifier: Programming Language :: Python :: 3.13
-Classifier: Programming Language :: Python :: 3.14
-Provides-Extra: examples
-Requires-Dist: accelerate (>=1.7.0,<2.0.0) ; extra == "examples"
-Requires-Dist: httpx (>=0.27.2,<1.0.0)
-Requires-Dist: nest-asyncio (>=1.6.0,<2.0.0)
-Requires-Dist: notebook (>=7.4.4,<8.0.0) ; extra == "examples"
-Requires-Dist: openai (>=1.81.0,<2.0.0) ; extra == "examples"
-Requires-Dist: torch (>=2.7.0,<3.0.0) ; extra == "examples"
-Requires-Dist: torchaudio (>=2.7.0,<3.0.0) ; extra == "examples"
-Requires-Dist: torchvision (>=0.22.0,<1.0.0) ; extra == "examples"
-Requires-Dist: tqdm (>=4.67.1,<5.0.0)
-Requires-Dist: transformers (>=4.51.3,<5.0.0) ; extra == "examples"
-Description-Content-Type: text/markdown
-
-<h1 align="center"> Trismik SDK</h1>
-
-<p align="center">
-  <img alt="PyPI - Version" src="https://img.shields.io/pypi/v/trismik">
-  <img alt="Python Version" src="https://img.shields.io/badge/python-3.9%2B-blue">
-  <img alt="License" src="https://img.shields.io/badge/license-MIT-green">
-</p>
-
-## Table of Contents
-- [Overview](#overview)
-- [Quick Start](#quick-start)
-  - [Installation](#installation)
-  - [API Key Setup](#api-key-setup)
-  - [Basic Usage](#basic-usage)
-- [Interpreting Results](#interpreting-results)
-  - [Theta (θ)](#theta-θ)
-  - [Other Metrics](#other-metrics)
-- [Contributing](#contributing)
-- [License](#license)
-
-## Overview
-
-[**Trismik**](https://trismik.com) is a Cambridge, UK based startup offering adversarial testing for LLMs. The APIs we provide through this library allow you to call our adaptive test engine and evaluate LLMs up to 95% faster (and cheaper!) than traditional evaluation techniques.
-
-Our **adaptive testing** algorithm allows to estimate the precision of the model by looking only at a small portion of a dataset. Through this library, we provide access to a number of open source datasets over several dimensions (reasoning, toxicity, tool use...) to speed up model testing in several scenarios, like foundation model training, supervised fine tuning, prompt engineering, and so on.
-
-## Quick Start
-
-### Installation
-
-To use our API, you need to get an API key first. Please register on [dashboard.trismik.com](https://dashboard.trismik.com) and obtain an API key.
-
-Trismik is available via [pypi](https://pypi.org/project/trismik/). To install Trismik, run the following in your terminal (in a virtualenv, if you use one):
-
-```bash
-pip install trismik
-```
-
-### API Key Setup
-
-You can provide your API key in one of the following ways:
-
-1. **Environment Variable**:
-   ```bash
-   export TRISMIK_API_KEY="your-api-key"
-   ```
-
-2. **`.env` File**:
-   ```bash
-   # .env
-   TRISMIK_API_KEY=your-api-key
-   ```
-   Then load it with `python-dotenv`:
-   ```python
-   from dotenv import load_dotenv
-   load_dotenv()
-   ```
-
-3. **Direct Initialization**:
-   ```python
-   client = TrismikAsyncClient(api_key="YOUR_API_KEY")
-   ```
-
-### Basic Usage
-
-Running a test is straightforward:
-
-1. Implement a method that wraps model inference over a dataset item
-2. Create an `AdaptiveTest` instance
-3. Run the test!
-
-Here's a basic example:
-
-```python
-def model_inference(item: TrismikItem) -> Any:
-    model_output = ...  # call your model here
-    return model_output
-
-
-# Initialize the test runner
-runner = AdaptiveTest(model_inference)
-
-# Run the test
-results = await runner.run_async(
-    "MMLUPro2025",  # or any dataset we support
-    with_responses=True,
-    run_metadata=sample_metadata,
-)
-
-# Print the test output
-for result in results:
-    print(f"{result.trait} ({result.name}): {result.value}")
-```
-
-### Examples
-
-You can find more examples in the `examples` folder:
-- [`example_transformers.py`](examples/example_transformers.py) - Example using Hugging Face Transformers models
-- [`example_openai.py`](examples/example_openai.py) - Example using OpenAI models
-- [`example_adaptive_test.py`](examples/example_adaptive_test.py) - Example of adaptive testing configuration
-
-To run the examples, you will need to clone this repo, navigate to the
-source folder, and then run:
-
-```bash
-poetry install --with examples
-poetry run python examples/example_adaptive_test.py # or any other example
-```
-
-## Interpreting Results
-
-### Theta (θ)
-
-Our adversarial test returns several values; however, you will be interested mainly in `theta`. Theta ($\theta$) is our metric; it measures the ability of the model on a certain dataset, and it can be used as a proxy to approximate the original metric used on that dataset. For example, on an accuracy-based dataset, a high theta correlates with a high accuracy, and low theta correlates with low accuracy.
-
-To interpret a theta score, consider that $\theta=0$ corresponds to a 50% chance for a model to get an answer right - in other words, to an accuracy of 50%.
-A negative theta means that the model will give more bad answers then good ones, while a positive theta means that the model will give more good answers then bad answers.
-While theta is unbounded in our implementation (i.e. $-\infty < \theta < \infty$), in practice we have that for most cases $\theta$ will take values between -3 and 3.
-
-Compared to classical benchmark testing, the estimated accuracy from adaptive testing uses fewer but more informative items while avoiding noise from overly easy or difficult questions. This makes it a more efficient and stable measure, especially on very large datasets.
-
-### Other Metrics
-
-- **Standard Deviation (`std`)**:
-  - A measure of the uncertainty or error in the theta estimate
-  - A smaller `std` indicates a more precise estimate
-  - You should see a `std` around or below 0.25
-
-- **Correct Responses (`responsesCorrect`)**:
-  - The number of correct answers delivered by the model
-
-  - **Important note**: A higher number of correct answers does not necessarily
-  correlate with a high theta. Our algorithm navigates the dataset to find a
-   balance of “hard” and “easy” items for your model, so by the end of the test,
-  it encounters a representative mix of inputs it can and cannot handle. In
-   practice, expect responsesCorrect to be roughly half of responsesTotal.
-
-- **Total Responses (`responsesTotal`)**:
-  - The number of items processed before reaching a stable theta.
-  - Expected range: 60 ≤ responses_total ≤ 80
-
-## Contributing
-
-See `CONTRIBUTING.md`.
-
-## License
-
-This library is licensed under the MIT license. See `LICENSE` file.
-

trismik-0.9.12.dist-info/RECORD

@@ -1,12 +0,0 @@
-trismik/__init__.py,sha256=20SwXrda9YsgykaoPohwz6foj2FkraniPA-GTQS9m00,197
-trismik/_mapper.py,sha256=BFryJLwYYKDlX_vhEPKarDbdX8FkSYfHFtVQOnAK67o,10994
-trismik/_utils.py,sha256=4dVRTWapyOQn8suGhzxD0-5Vn5m8_Uuc7gB434n0SdM,3848
-trismik/adaptive_test.py,sha256=vG8OfL02IeQfuPpj3SdotxZRWKuMZGZiuPBhOU-PBpw,21594
-trismik/client_async.py,sha256=VXMxdIFHmIUP0zcHO1Kq7-uHWXYSqWbUqASPm8MSQjU,13909
-trismik/exceptions.py,sha256=2wb4_K7GdDf00s3xUaiSfw6718ZV3Eaa4M2lYbiEZl4,1945
-trismik/settings.py,sha256=ErXj3f9Uw1AgewytjxmIY7TmM5jB7RE_LJMp3T7szBY,419
-trismik/types.py,sha256=zo6QcXpo7ZfdEntvJDzftBPq_8_g0fS-XDfNTtKVg8k,6555
-trismik-0.9.12.dist-info/METADATA,sha256=sHnT177cyW8nQQ_Av9oXjrnMWFHR3KetBdUJ3i9cdmM,6815
-trismik-0.9.12.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
-trismik-0.9.12.dist-info/licenses/LICENSE,sha256=tgetRhapGLh7ZxfknW6Mm-WobfziPd64nAK52X5XKaw,1077
-trismik-0.9.12.dist-info/RECORD,,