openai/openai-python
Publicmirrored from https://github.com/openai/openai-pythonAvailable
README.md
715lines · modecode
| 1 | # OpenAI Python API library |
| 2 | |
| 3 | [](https://pypi.org/project/openai/) |
| 4 | |
| 5 | The OpenAI Python library provides convenient access to the OpenAI REST API from any Python 3.7+ |
| 6 | application. The library includes type definitions for all request params and response fields, |
| 7 | and offers both synchronous and asynchronous clients powered by [httpx](https://github.com/encode/httpx). |
| 8 | |
| 9 | It is generated from our [OpenAPI specification](https://github.com/openai/openai-openapi) with [Stainless](https://stainlessapi.com/). |
| 10 | |
| 11 | ## Documentation |
| 12 | |
| 13 | The REST API documentation can be found on [platform.openai.com](https://platform.openai.com/docs). The full API of this library can be found in [api.md](api.md). |
| 14 | |
| 15 | ## Installation |
| 16 | |
| 17 | > [!IMPORTANT] |
| 18 | > The SDK was rewritten in v1, which was released November 6th 2023. See the [v1 migration guide](https://github.com/openai/openai-python/discussions/742), which includes scripts to automatically update your code. |
| 19 | |
| 20 | ```sh |
| 21 | # install from PyPI |
| 22 | pip install openai |
| 23 | ``` |
| 24 | |
| 25 | ## Usage |
| 26 | |
| 27 | The full API of this library can be found in [api.md](api.md). |
| 28 | |
| 29 | ```python |
| 30 | import os |
| 31 | from openai import OpenAI |
| 32 | |
| 33 | client = OpenAI( |
| 34 | # This is the default and can be omitted |
| 35 | api_key=os.environ.get("OPENAI_API_KEY"), |
| 36 | ) |
| 37 | |
| 38 | chat_completion = client.chat.completions.create( |
| 39 | messages=[ |
| 40 | { |
| 41 | "role": "user", |
| 42 | "content": "Say this is a test", |
| 43 | } |
| 44 | ], |
| 45 | model="gpt-3.5-turbo", |
| 46 | ) |
| 47 | ``` |
| 48 | |
| 49 | While you can provide an `api_key` keyword argument, |
| 50 | we recommend using [python-dotenv](https://pypi.org/project/python-dotenv/) |
| 51 | to add `OPENAI_API_KEY="My API Key"` to your `.env` file |
| 52 | so that your API Key is not stored in source control. |
| 53 | |
| 54 | ### Vision |
| 55 | |
| 56 | With a hosted image: |
| 57 | |
| 58 | ```python |
| 59 | response = client.chat.completions.create( |
| 60 | model="gpt-4o-mini", |
| 61 | messages=[ |
| 62 | { |
| 63 | "role": "user", |
| 64 | "content": [ |
| 65 | {"type": "text", "text": prompt}, |
| 66 | { |
| 67 | "type": "image_url", |
| 68 | "image_url": {"url": f"{img_url}"}, |
| 69 | }, |
| 70 | ], |
| 71 | } |
| 72 | ], |
| 73 | ) |
| 74 | ``` |
| 75 | |
| 76 | With the image as a base64 encoded string: |
| 77 | |
| 78 | ```python |
| 79 | response = client.chat.completions.create( |
| 80 | model="gpt-4o-mini", |
| 81 | messages=[ |
| 82 | { |
| 83 | "role": "user", |
| 84 | "content": [ |
| 85 | {"type": "text", "text": prompt}, |
| 86 | { |
| 87 | "type": "image_url", |
| 88 | "image_url": {"url": f"data:{img_type};base64,{img_b64_str}"}, |
| 89 | }, |
| 90 | ], |
| 91 | } |
| 92 | ], |
| 93 | ) |
| 94 | ``` |
| 95 | |
| 96 | ### Polling Helpers |
| 97 | |
| 98 | When interacting with the API some actions such as starting a Run and adding files to vector stores are asynchronous and take time to complete. The SDK includes |
| 99 | helper functions which will poll the status until it reaches a terminal state and then return the resulting object. |
| 100 | If an API method results in an action that could benefit from polling there will be a corresponding version of the |
| 101 | method ending in '\_and_poll'. |
| 102 | |
| 103 | For instance to create a Run and poll until it reaches a terminal state you can run: |
| 104 | |
| 105 | ```python |
| 106 | run = client.beta.threads.runs.create_and_poll( |
| 107 | thread_id=thread.id, |
| 108 | assistant_id=assistant.id, |
| 109 | ) |
| 110 | ``` |
| 111 | |
| 112 | More information on the lifecycle of a Run can be found in the [Run Lifecycle Documentation](https://platform.openai.com/docs/assistants/how-it-works/run-lifecycle) |
| 113 | |
| 114 | ### Bulk Upload Helpers |
| 115 | |
| 116 | When creating and interacting with vector stores, you can use polling helpers to monitor the status of operations. |
| 117 | For convenience, we also provide a bulk upload helper to allow you to simultaneously upload several files at once. |
| 118 | |
| 119 | ```python |
| 120 | sample_files = [Path("sample-paper.pdf"), ...] |
| 121 | |
| 122 | batch = await client.vector_stores.file_batches.upload_and_poll( |
| 123 | store.id, |
| 124 | files=sample_files, |
| 125 | ) |
| 126 | ``` |
| 127 | |
| 128 | ### Streaming Helpers |
| 129 | |
| 130 | The SDK also includes helpers to process streams and handle incoming events. |
| 131 | |
| 132 | ```python |
| 133 | with client.beta.threads.runs.stream( |
| 134 | thread_id=thread.id, |
| 135 | assistant_id=assistant.id, |
| 136 | instructions="Please address the user as Jane Doe. The user has a premium account.", |
| 137 | ) as stream: |
| 138 | for event in stream: |
| 139 | # Print the text from text delta events |
| 140 | if event.type == "thread.message.delta" and event.data.delta.content: |
| 141 | print(event.data.delta.content[0].text) |
| 142 | ``` |
| 143 | |
| 144 | More information on streaming helpers can be found in the dedicated documentation: [helpers.md](helpers.md) |
| 145 | |
| 146 | ## Async usage |
| 147 | |
| 148 | Simply import `AsyncOpenAI` instead of `OpenAI` and use `await` with each API call: |
| 149 | |
| 150 | ```python |
| 151 | import os |
| 152 | import asyncio |
| 153 | from openai import AsyncOpenAI |
| 154 | |
| 155 | client = AsyncOpenAI( |
| 156 | # This is the default and can be omitted |
| 157 | api_key=os.environ.get("OPENAI_API_KEY"), |
| 158 | ) |
| 159 | |
| 160 | |
| 161 | async def main() -> None: |
| 162 | chat_completion = await client.chat.completions.create( |
| 163 | messages=[ |
| 164 | { |
| 165 | "role": "user", |
| 166 | "content": "Say this is a test", |
| 167 | } |
| 168 | ], |
| 169 | model="gpt-3.5-turbo", |
| 170 | ) |
| 171 | |
| 172 | |
| 173 | asyncio.run(main()) |
| 174 | ``` |
| 175 | |
| 176 | Functionality between the synchronous and asynchronous clients is otherwise identical. |
| 177 | |
| 178 | ## Streaming responses |
| 179 | |
| 180 | We provide support for streaming responses using Server Side Events (SSE). |
| 181 | |
| 182 | ```python |
| 183 | from openai import OpenAI |
| 184 | |
| 185 | client = OpenAI() |
| 186 | |
| 187 | stream = client.chat.completions.create( |
| 188 | model="gpt-4", |
| 189 | messages=[{"role": "user", "content": "Say this is a test"}], |
| 190 | stream=True, |
| 191 | ) |
| 192 | for chunk in stream: |
| 193 | print(chunk.choices[0].delta.content or "", end="") |
| 194 | ``` |
| 195 | |
| 196 | The async client uses the exact same interface. |
| 197 | |
| 198 | ```python |
| 199 | from openai import AsyncOpenAI |
| 200 | |
| 201 | client = AsyncOpenAI() |
| 202 | |
| 203 | |
| 204 | async def main(): |
| 205 | stream = await client.chat.completions.create( |
| 206 | model="gpt-4", |
| 207 | messages=[{"role": "user", "content": "Say this is a test"}], |
| 208 | stream=True, |
| 209 | ) |
| 210 | async for chunk in stream: |
| 211 | print(chunk.choices[0].delta.content or "", end="") |
| 212 | |
| 213 | |
| 214 | asyncio.run(main()) |
| 215 | ``` |
| 216 | |
| 217 | ## Module-level client |
| 218 | |
| 219 | > [!IMPORTANT] |
| 220 | > We highly recommend instantiating client instances instead of relying on the global client. |
| 221 | |
| 222 | We also expose a global client instance that is accessible in a similar fashion to versions prior to v1. |
| 223 | |
| 224 | ```py |
| 225 | import openai |
| 226 | |
| 227 | # optional; defaults to `os.environ['OPENAI_API_KEY']` |
| 228 | openai.api_key = '...' |
| 229 | |
| 230 | # all client options can be configured just like the `OpenAI` instantiation counterpart |
| 231 | openai.base_url = "https://..." |
| 232 | openai.default_headers = {"x-foo": "true"} |
| 233 | |
| 234 | completion = openai.chat.completions.create( |
| 235 | model="gpt-4", |
| 236 | messages=[ |
| 237 | { |
| 238 | "role": "user", |
| 239 | "content": "How do I output all files in a directory using Python?", |
| 240 | }, |
| 241 | ], |
| 242 | ) |
| 243 | print(completion.choices[0].message.content) |
| 244 | ``` |
| 245 | |
| 246 | The API is the exact same as the standard client instance-based API. |
| 247 | |
| 248 | This is intended to be used within REPLs or notebooks for faster iteration, **not** in application code. |
| 249 | |
| 250 | We recommend that you always instantiate a client (e.g., with `client = OpenAI()`) in application code because: |
| 251 | |
| 252 | - It can be difficult to reason about where client options are configured |
| 253 | - It's not possible to change certain client options without potentially causing race conditions |
| 254 | - It's harder to mock for testing purposes |
| 255 | - It's not possible to control cleanup of network connections |
| 256 | |
| 257 | ## Using types |
| 258 | |
| 259 | Nested request parameters are [TypedDicts](https://docs.python.org/3/library/typing.html#typing.TypedDict). Responses are [Pydantic models](https://docs.pydantic.dev) which also provide helper methods for things like: |
| 260 | |
| 261 | - Serializing back into JSON, `model.to_json()` |
| 262 | - Converting to a dictionary, `model.to_dict()` |
| 263 | |
| 264 | Typed requests and responses provide autocomplete and documentation within your editor. If you would like to see type errors in VS Code to help catch bugs earlier, set `python.analysis.typeCheckingMode` to `basic`. |
| 265 | |
| 266 | ## Pagination |
| 267 | |
| 268 | List methods in the OpenAI API are paginated. |
| 269 | |
| 270 | This library provides auto-paginating iterators with each list response, so you do not have to request successive pages manually: |
| 271 | |
| 272 | ```python |
| 273 | from openai import OpenAI |
| 274 | |
| 275 | client = OpenAI() |
| 276 | |
| 277 | all_jobs = [] |
| 278 | # Automatically fetches more pages as needed. |
| 279 | for job in client.fine_tuning.jobs.list( |
| 280 | limit=20, |
| 281 | ): |
| 282 | # Do something with job here |
| 283 | all_jobs.append(job) |
| 284 | print(all_jobs) |
| 285 | ``` |
| 286 | |
| 287 | Or, asynchronously: |
| 288 | |
| 289 | ```python |
| 290 | import asyncio |
| 291 | from openai import AsyncOpenAI |
| 292 | |
| 293 | client = AsyncOpenAI() |
| 294 | |
| 295 | |
| 296 | async def main() -> None: |
| 297 | all_jobs = [] |
| 298 | # Iterate through items across all pages, issuing requests as needed. |
| 299 | async for job in client.fine_tuning.jobs.list( |
| 300 | limit=20, |
| 301 | ): |
| 302 | all_jobs.append(job) |
| 303 | print(all_jobs) |
| 304 | |
| 305 | |
| 306 | asyncio.run(main()) |
| 307 | ``` |
| 308 | |
| 309 | Alternatively, you can use the `.has_next_page()`, `.next_page_info()`, or `.get_next_page()` methods for more granular control working with pages: |
| 310 | |
| 311 | ```python |
| 312 | first_page = await client.fine_tuning.jobs.list( |
| 313 | limit=20, |
| 314 | ) |
| 315 | if first_page.has_next_page(): |
| 316 | print(f"will fetch next page using these details: {first_page.next_page_info()}") |
| 317 | next_page = await first_page.get_next_page() |
| 318 | print(f"number of items we just fetched: {len(next_page.data)}") |
| 319 | |
| 320 | # Remove `await` for non-async usage. |
| 321 | ``` |
| 322 | |
| 323 | Or just work directly with the returned data: |
| 324 | |
| 325 | ```python |
| 326 | first_page = await client.fine_tuning.jobs.list( |
| 327 | limit=20, |
| 328 | ) |
| 329 | |
| 330 | print(f"next page cursor: {first_page.after}") # => "next page cursor: ..." |
| 331 | for job in first_page.data: |
| 332 | print(job.id) |
| 333 | |
| 334 | # Remove `await` for non-async usage. |
| 335 | ``` |
| 336 | |
| 337 | ## Nested params |
| 338 | |
| 339 | Nested parameters are dictionaries, typed using `TypedDict`, for example: |
| 340 | |
| 341 | ```python |
| 342 | from openai import OpenAI |
| 343 | |
| 344 | client = OpenAI() |
| 345 | |
| 346 | completion = client.chat.completions.create( |
| 347 | messages=[ |
| 348 | { |
| 349 | "role": "user", |
| 350 | "content": "Can you generate an example json object describing a fruit?", |
| 351 | } |
| 352 | ], |
| 353 | model="gpt-3.5-turbo-1106", |
| 354 | response_format={"type": "json_object"}, |
| 355 | ) |
| 356 | ``` |
| 357 | |
| 358 | ## File uploads |
| 359 | |
| 360 | Request parameters that correspond to file uploads can be passed as `bytes`, a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance or a tuple of `(filename, contents, media type)`. |
| 361 | |
| 362 | ```python |
| 363 | from pathlib import Path |
| 364 | from openai import OpenAI |
| 365 | |
| 366 | client = OpenAI() |
| 367 | |
| 368 | client.files.create( |
| 369 | file=Path("input.jsonl"), |
| 370 | purpose="fine-tune", |
| 371 | ) |
| 372 | ``` |
| 373 | |
| 374 | The async client uses the exact same interface. If you pass a [`PathLike`](https://docs.python.org/3/library/os.html#os.PathLike) instance, the file contents will be read asynchronously automatically. |
| 375 | |
| 376 | ## Handling errors |
| 377 | |
| 378 | When the library is unable to connect to the API (for example, due to network connection problems or a timeout), a subclass of `openai.APIConnectionError` is raised. |
| 379 | |
| 380 | When the API returns a non-success status code (that is, 4xx or 5xx |
| 381 | response), a subclass of `openai.APIStatusError` is raised, containing `status_code` and `response` properties. |
| 382 | |
| 383 | All errors inherit from `openai.APIError`. |
| 384 | |
| 385 | ```python |
| 386 | import openai |
| 387 | from openai import OpenAI |
| 388 | |
| 389 | client = OpenAI() |
| 390 | |
| 391 | try: |
| 392 | client.fine_tuning.jobs.create( |
| 393 | model="gpt-3.5-turbo", |
| 394 | training_file="file-abc123", |
| 395 | ) |
| 396 | except openai.APIConnectionError as e: |
| 397 | print("The server could not be reached") |
| 398 | print(e.__cause__) # an underlying Exception, likely raised within httpx. |
| 399 | except openai.RateLimitError as e: |
| 400 | print("A 429 status code was received; we should back off a bit.") |
| 401 | except openai.APIStatusError as e: |
| 402 | print("Another non-200-range status code was received") |
| 403 | print(e.status_code) |
| 404 | print(e.response) |
| 405 | ``` |
| 406 | |
| 407 | Error codes are as followed: |
| 408 | |
| 409 | | Status Code | Error Type | |
| 410 | | ----------- | -------------------------- | |
| 411 | | 400 | `BadRequestError` | |
| 412 | | 401 | `AuthenticationError` | |
| 413 | | 403 | `PermissionDeniedError` | |
| 414 | | 404 | `NotFoundError` | |
| 415 | | 422 | `UnprocessableEntityError` | |
| 416 | | 429 | `RateLimitError` | |
| 417 | | >=500 | `InternalServerError` | |
| 418 | | N/A | `APIConnectionError` | |
| 419 | |
| 420 | ## Request IDs |
| 421 | |
| 422 | > For more information on debugging requests, see [these docs](https://platform.openai.com/docs/api-reference/debugging-requests) |
| 423 | |
| 424 | All object responses in the SDK provide a `_request_id` property which is added from the `x-request-id` response header so that you can quickly log failing requests and report them back to OpenAI. |
| 425 | |
| 426 | ```python |
| 427 | completion = await client.chat.completions.create( |
| 428 | messages=[{"role": "user", "content": "Say this is a test"}], model="gpt-4" |
| 429 | ) |
| 430 | print(completion._request_id) # req_123 |
| 431 | ``` |
| 432 | |
| 433 | Note that unlike other properties that use an `_` prefix, the `_request_id` property |
| 434 | *is* public. Unless documented otherwise, *all* other `_` prefix properties, |
| 435 | methods and modules are *private*. |
| 436 | |
| 437 | |
| 438 | ### Retries |
| 439 | |
| 440 | Certain errors are automatically retried 2 times by default, with a short exponential backoff. |
| 441 | Connection errors (for example, due to a network connectivity problem), 408 Request Timeout, 409 Conflict, |
| 442 | 429 Rate Limit, and >=500 Internal errors are all retried by default. |
| 443 | |
| 444 | You can use the `max_retries` option to configure or disable retry settings: |
| 445 | |
| 446 | ```python |
| 447 | from openai import OpenAI |
| 448 | |
| 449 | # Configure the default for all requests: |
| 450 | client = OpenAI( |
| 451 | # default is 2 |
| 452 | max_retries=0, |
| 453 | ) |
| 454 | |
| 455 | # Or, configure per-request: |
| 456 | client.with_options(max_retries=5).chat.completions.create( |
| 457 | messages=[ |
| 458 | { |
| 459 | "role": "user", |
| 460 | "content": "How can I get the name of the current day in Node.js?", |
| 461 | } |
| 462 | ], |
| 463 | model="gpt-3.5-turbo", |
| 464 | ) |
| 465 | ``` |
| 466 | |
| 467 | ### Timeouts |
| 468 | |
| 469 | By default requests time out after 10 minutes. You can configure this with a `timeout` option, |
| 470 | which accepts a float or an [`httpx.Timeout`](https://www.python-httpx.org/advanced/#fine-tuning-the-configuration) object: |
| 471 | |
| 472 | ```python |
| 473 | from openai import OpenAI |
| 474 | |
| 475 | # Configure the default for all requests: |
| 476 | client = OpenAI( |
| 477 | # 20 seconds (default is 10 minutes) |
| 478 | timeout=20.0, |
| 479 | ) |
| 480 | |
| 481 | # More granular control: |
| 482 | client = OpenAI( |
| 483 | timeout=httpx.Timeout(60.0, read=5.0, write=10.0, connect=2.0), |
| 484 | ) |
| 485 | |
| 486 | # Override per-request: |
| 487 | client.with_options(timeout=5.0).chat.completions.create( |
| 488 | messages=[ |
| 489 | { |
| 490 | "role": "user", |
| 491 | "content": "How can I list all files in a directory using Python?", |
| 492 | } |
| 493 | ], |
| 494 | model="gpt-3.5-turbo", |
| 495 | ) |
| 496 | ``` |
| 497 | |
| 498 | On timeout, an `APITimeoutError` is thrown. |
| 499 | |
| 500 | Note that requests that time out are [retried twice by default](#retries). |
| 501 | |
| 502 | ## Advanced |
| 503 | |
| 504 | ### Logging |
| 505 | |
| 506 | We use the standard library [`logging`](https://docs.python.org/3/library/logging.html) module. |
| 507 | |
| 508 | You can enable logging by setting the environment variable `OPENAI_LOG` to `debug`. |
| 509 | |
| 510 | ```shell |
| 511 | $ export OPENAI_LOG=debug |
| 512 | ``` |
| 513 | |
| 514 | ### How to tell whether `None` means `null` or missing |
| 515 | |
| 516 | In an API response, a field may be explicitly `null`, or missing entirely; in either case, its value is `None` in this library. You can differentiate the two cases with `.model_fields_set`: |
| 517 | |
| 518 | ```py |
| 519 | if response.my_field is None: |
| 520 | if 'my_field' not in response.model_fields_set: |
| 521 | print('Got json like {}, without a "my_field" key present at all.') |
| 522 | else: |
| 523 | print('Got json like {"my_field": null}.') |
| 524 | ``` |
| 525 | |
| 526 | ### Accessing raw response data (e.g. headers) |
| 527 | |
| 528 | The "raw" Response object can be accessed by prefixing `.with_raw_response.` to any HTTP method call, e.g., |
| 529 | |
| 530 | ```py |
| 531 | from openai import OpenAI |
| 532 | |
| 533 | client = OpenAI() |
| 534 | response = client.chat.completions.with_raw_response.create( |
| 535 | messages=[{ |
| 536 | "role": "user", |
| 537 | "content": "Say this is a test", |
| 538 | }], |
| 539 | model="gpt-3.5-turbo", |
| 540 | ) |
| 541 | print(response.headers.get('X-My-Header')) |
| 542 | |
| 543 | completion = response.parse() # get the object that `chat.completions.create()` would have returned |
| 544 | print(completion) |
| 545 | ``` |
| 546 | |
| 547 | These methods return an [`LegacyAPIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_legacy_response.py) object. This is a legacy class as we're changing it slightly in the next major version. |
| 548 | |
| 549 | For the sync client this will mostly be the same with the exception |
| 550 | of `content` & `text` will be methods instead of properties. In the |
| 551 | async client, all methods will be async. |
| 552 | |
| 553 | A migration script will be provided & the migration in general should |
| 554 | be smooth. |
| 555 | |
| 556 | #### `.with_streaming_response` |
| 557 | |
| 558 | The above interface eagerly reads the full response body when you make the request, which may not always be what you want. |
| 559 | |
| 560 | To stream the response body, use `.with_streaming_response` instead, which requires a context manager and only reads the response body once you call `.read()`, `.text()`, `.json()`, `.iter_bytes()`, `.iter_text()`, `.iter_lines()` or `.parse()`. In the async client, these are async methods. |
| 561 | |
| 562 | As such, `.with_streaming_response` methods return a different [`APIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_response.py) object, and the async client returns an [`AsyncAPIResponse`](https://github.com/openai/openai-python/tree/main/src/openai/_response.py) object. |
| 563 | |
| 564 | ```python |
| 565 | with client.chat.completions.with_streaming_response.create( |
| 566 | messages=[ |
| 567 | { |
| 568 | "role": "user", |
| 569 | "content": "Say this is a test", |
| 570 | } |
| 571 | ], |
| 572 | model="gpt-3.5-turbo", |
| 573 | ) as response: |
| 574 | print(response.headers.get("X-My-Header")) |
| 575 | |
| 576 | for line in response.iter_lines(): |
| 577 | print(line) |
| 578 | ``` |
| 579 | |
| 580 | The context manager is required so that the response will reliably be closed. |
| 581 | |
| 582 | ### Making custom/undocumented requests |
| 583 | |
| 584 | This library is typed for convenient access to the documented API. |
| 585 | |
| 586 | If you need to access undocumented endpoints, params, or response properties, the library can still be used. |
| 587 | |
| 588 | #### Undocumented endpoints |
| 589 | |
| 590 | To make requests to undocumented endpoints, you can make requests using `client.get`, `client.post`, and other |
| 591 | http verbs. Options on the client will be respected (such as retries) will be respected when making this |
| 592 | request. |
| 593 | |
| 594 | ```py |
| 595 | import httpx |
| 596 | |
| 597 | response = client.post( |
| 598 | "/foo", |
| 599 | cast_to=httpx.Response, |
| 600 | body={"my_param": True}, |
| 601 | ) |
| 602 | |
| 603 | print(response.headers.get("x-foo")) |
| 604 | ``` |
| 605 | |
| 606 | #### Undocumented request params |
| 607 | |
| 608 | If you want to explicitly send an extra param, you can do so with the `extra_query`, `extra_body`, and `extra_headers` request |
| 609 | options. |
| 610 | |
| 611 | #### Undocumented response properties |
| 612 | |
| 613 | To access undocumented response properties, you can access the extra fields like `response.unknown_prop`. You |
| 614 | can also get all the extra fields on the Pydantic model as a dict with |
| 615 | [`response.model_extra`](https://docs.pydantic.dev/latest/api/base_model/#pydantic.BaseModel.model_extra). |
| 616 | |
| 617 | ### Configuring the HTTP client |
| 618 | |
| 619 | You can directly override the [httpx client](https://www.python-httpx.org/api/#client) to customize it for your use case, including: |
| 620 | |
| 621 | - Support for proxies |
| 622 | - Custom transports |
| 623 | - Additional [advanced](https://www.python-httpx.org/advanced/clients/) functionality |
| 624 | |
| 625 | ```python |
| 626 | from openai import OpenAI, DefaultHttpxClient |
| 627 | |
| 628 | client = OpenAI( |
| 629 | # Or use the `OPENAI_BASE_URL` env var |
| 630 | base_url="http://my.test.server.example.com:8083/v1", |
| 631 | http_client=DefaultHttpxClient( |
| 632 | proxies="http://my.test.proxy.example.com", |
| 633 | transport=httpx.HTTPTransport(local_address="0.0.0.0"), |
| 634 | ), |
| 635 | ) |
| 636 | ``` |
| 637 | |
| 638 | You can also customize the client on a per-request basis by using `with_options()`: |
| 639 | |
| 640 | ```python |
| 641 | client.with_options(http_client=DefaultHttpxClient(...)) |
| 642 | ``` |
| 643 | |
| 644 | ### Managing HTTP resources |
| 645 | |
| 646 | By default the library closes underlying HTTP connections whenever the client is [garbage collected](https://docs.python.org/3/reference/datamodel.html#object.__del__). You can manually close the client using the `.close()` method if desired, or with a context manager that closes when exiting. |
| 647 | |
| 648 | ## Microsoft Azure OpenAI |
| 649 | |
| 650 | To use this library with [Azure OpenAI](https://learn.microsoft.com/azure/ai-services/openai/overview), use the `AzureOpenAI` |
| 651 | class instead of the `OpenAI` class. |
| 652 | |
| 653 | > [!IMPORTANT] |
| 654 | > The Azure API shape differs from the core API shape which means that the static types for responses / params |
| 655 | > won't always be correct. |
| 656 | |
| 657 | ```py |
| 658 | from openai import AzureOpenAI |
| 659 | |
| 660 | # gets the API Key from environment variable AZURE_OPENAI_API_KEY |
| 661 | client = AzureOpenAI( |
| 662 | # https://learn.microsoft.com/azure/ai-services/openai/reference#rest-api-versioning |
| 663 | api_version="2023-07-01-preview", |
| 664 | # https://learn.microsoft.com/azure/cognitive-services/openai/how-to/create-resource?pivots=web-portal#create-a-resource |
| 665 | azure_endpoint="https://example-endpoint.openai.azure.com", |
| 666 | ) |
| 667 | |
| 668 | completion = client.chat.completions.create( |
| 669 | model="deployment-name", # e.g. gpt-35-instant |
| 670 | messages=[ |
| 671 | { |
| 672 | "role": "user", |
| 673 | "content": "How do I output all files in a directory using Python?", |
| 674 | }, |
| 675 | ], |
| 676 | ) |
| 677 | print(completion.to_json()) |
| 678 | ``` |
| 679 | |
| 680 | In addition to the options provided in the base `OpenAI` client, the following options are provided: |
| 681 | |
| 682 | - `azure_endpoint` (or the `AZURE_OPENAI_ENDPOINT` environment variable) |
| 683 | - `azure_deployment` |
| 684 | - `api_version` (or the `OPENAI_API_VERSION` environment variable) |
| 685 | - `azure_ad_token` (or the `AZURE_OPENAI_AD_TOKEN` environment variable) |
| 686 | - `azure_ad_token_provider` |
| 687 | |
| 688 | An example of using the client with Microsoft Entra ID (formerly known as Azure Active Directory) can be found [here](https://github.com/openai/openai-python/blob/main/examples/azure_ad.py). |
| 689 | |
| 690 | ## Versioning |
| 691 | |
| 692 | This package generally follows [SemVer](https://semver.org/spec/v2.0.0.html) conventions, though certain backwards-incompatible changes may be released as minor versions: |
| 693 | |
| 694 | 1. Changes that only affect static types, without breaking runtime behavior. |
| 695 | 2. Changes to library internals which are technically public but not intended or documented for external use. _(Please open a GitHub issue to let us know if you are relying on such internals)_. |
| 696 | 3. Changes that we do not expect to impact the vast majority of users in practice. |
| 697 | |
| 698 | We take backwards-compatibility seriously and work hard to ensure you can rely on a smooth upgrade experience. |
| 699 | |
| 700 | We are keen for your feedback; please open an [issue](https://www.github.com/openai/openai-python/issues) with questions, bugs, or suggestions. |
| 701 | |
| 702 | ### Determining the installed version |
| 703 | |
| 704 | If you've upgraded to the latest version but aren't seeing any new features you were expecting then your python environment is likely still using an older version. |
| 705 | |
| 706 | You can determine the version that is being used at runtime with: |
| 707 | |
| 708 | ```py |
| 709 | import openai |
| 710 | print(openai.__version__) |
| 711 | ``` |
| 712 | |
| 713 | ## Requirements |
| 714 | |
| 715 | Python 3.7 or higher. |
| 716 | |