aidial-sdk 0.46.0.dev8__tar.gz → 0.47.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- aidial_sdk-0.46.0.dev8/README.md → aidial_sdk-0.47.0/PKG-INFO +110 -0
- aidial_sdk-0.46.0.dev8/PKG-INFO → aidial_sdk-0.47.0/README.md +67 -40
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/application.py +2 -2
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/request.py +3 -3
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/response.py +3 -3
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/from_request_mixin.py +12 -9
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/telemetry/init.py +19 -9
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/__init__.py +28 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/app.py +134 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/cache_breakpoints.py +48 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/dial_extensions.py +66 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/from_chat_completions.py +237 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/reasoning.py +48 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/streaming.py +245 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/system_turns.py +26 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/chat_completions/to_chat_completions.py +695 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/common.py +61 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/core_client.py +64 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/errors.py +153 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/settings.py +21 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/sse_stream.py +209 -0
- aidial_sdk-0.47.0/aidial_sdk/translators/messages_to_chat_completions/translation_state.py +39 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/streaming.py +1 -2
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/pyproject.toml +21 -14
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/LICENSE +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/_errors.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/_pydantic/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/_pydantic/_compat.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/_pydantic/_model_config.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/_cache.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/_types.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/base.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/choice.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/choice_base.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/chunks.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/enums.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/form.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/function_call.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/function_tool_call.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/chat_completion/stage.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/_headers.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/configuration.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/rate.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/tokenize.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/deployment/truncate_prompt.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/embeddings/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/embeddings/base.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/embeddings/request.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/embeddings/response.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/exceptions.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/header_propagator.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/py.typed +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/pydantic/v2.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/pydantic_v1/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/response_headers.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/telemetry/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/telemetry/_context.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/telemetry/types.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/__init__.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_attachment.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_cancel_scope.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_content_stream.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_disconnect_middleware.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_indexed_list.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_json_log_formatter.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_logging.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/_reflection.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/env.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/errors.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/json.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/log_config.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/logging.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/merge_chunks.py +0 -0
- {aidial_sdk-0.46.0.dev8 → aidial_sdk-0.47.0}/aidial_sdk/utils/pydantic.py +0 -0
|
@@ -1,3 +1,45 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: aidial-sdk
|
|
3
|
+
Version: 0.47.0
|
|
4
|
+
Summary: Framework to create applications and model adapters for AI DIAL
|
|
5
|
+
License-Expression: Apache-2.0
|
|
6
|
+
License-File: LICENSE
|
|
7
|
+
Author: EPAM RAIL
|
|
8
|
+
Author-email: SpecialEPM-DIALDevTeam@epam.com
|
|
9
|
+
Requires-Python: >=3.11,<4.0
|
|
10
|
+
Classifier: Programming Language :: Python :: 3
|
|
11
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
12
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
13
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
14
|
+
Classifier: Programming Language :: Python :: 3.14
|
|
15
|
+
Provides-Extra: httpx
|
|
16
|
+
Provides-Extra: telemetry
|
|
17
|
+
Provides-Extra: translators
|
|
18
|
+
Requires-Dist: anthropic (>=0.105.0,<1.0) ; extra == "translators"
|
|
19
|
+
Requires-Dist: fastapi (>=0.51,<1.0)
|
|
20
|
+
Requires-Dist: httpx (>=0.25.0,<1.0) ; extra == "httpx"
|
|
21
|
+
Requires-Dist: httpx (>=0.25.0,<1.0) ; extra == "translators"
|
|
22
|
+
Requires-Dist: openai (>=2.46.0,<3.0) ; extra == "translators"
|
|
23
|
+
Requires-Dist: opentelemetry-api (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
24
|
+
Requires-Dist: opentelemetry-exporter-otlp-proto-grpc (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
25
|
+
Requires-Dist: opentelemetry-exporter-prometheus (>=0.64b0) ; extra == "telemetry"
|
|
26
|
+
Requires-Dist: opentelemetry-instrumentation-aiohttp-client (>=0.64b0) ; extra == "telemetry"
|
|
27
|
+
Requires-Dist: opentelemetry-instrumentation-fastapi (>=0.64b0) ; extra == "telemetry"
|
|
28
|
+
Requires-Dist: opentelemetry-instrumentation-httpx (>=0.64b0) ; extra == "telemetry"
|
|
29
|
+
Requires-Dist: opentelemetry-instrumentation-logging (>=0.64b0) ; extra == "telemetry"
|
|
30
|
+
Requires-Dist: opentelemetry-instrumentation-requests (>=0.64b0) ; extra == "telemetry"
|
|
31
|
+
Requires-Dist: opentelemetry-instrumentation-system-metrics (>=0.64b0) ; extra == "telemetry"
|
|
32
|
+
Requires-Dist: opentelemetry-instrumentation-urllib (>=0.64b0) ; extra == "telemetry"
|
|
33
|
+
Requires-Dist: opentelemetry-sdk (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
34
|
+
Requires-Dist: prometheus-client (>=0.17.1,<1.0) ; extra == "telemetry"
|
|
35
|
+
Requires-Dist: pydantic (>=1.10.17,<3)
|
|
36
|
+
Requires-Dist: uvicorn (>=0.19,<1.0)
|
|
37
|
+
Requires-Dist: wrapt (>=1.10,<3)
|
|
38
|
+
Project-URL: Documentation, https://dialx.ai/dial_api
|
|
39
|
+
Project-URL: Homepage, https://dialx.ai
|
|
40
|
+
Project-URL: Repository, https://github.com/epam/ai-dial-sdk
|
|
41
|
+
Description-Content-Type: text/markdown
|
|
42
|
+
|
|
1
43
|
<h1 align="center">
|
|
2
44
|
AI DIAL Python SDK
|
|
3
45
|
</h1>
|
|
@@ -23,6 +65,7 @@
|
|
|
23
65
|
- [Run](#run)
|
|
24
66
|
- [Check](#check)
|
|
25
67
|
- [Endpoints](#endpoints)
|
|
68
|
+
- [Anthropic Messages to Chat Completions translator](#anthropic-messages-to-chat-completions-translator)
|
|
26
69
|
- [Development](#development)
|
|
27
70
|
- [Development Environment](#development-environment)
|
|
28
71
|
- [Setup](#setup)
|
|
@@ -167,6 +210,72 @@ An `/openai/v1` request with none of the headers set fails with `500 Internal Se
|
|
|
167
210
|
|
|
168
211
|
---
|
|
169
212
|
|
|
213
|
+
### Anthropic Messages to Chat Completions translator
|
|
214
|
+
|
|
215
|
+
The translator lets an Anthropic-native client (e.g. Claude Code) talk to a model that only speaks the [Chat Completions API](https://platform.openai.com/docs/api-reference/chat). It converts an Anthropic Messages request into a Chat Completions request to DIAL Core at `DIAL_URL`, and converts the response, streaming or not, back into the Anthropic Messages shape.
|
|
216
|
+
|
|
217
|
+
The translator requires the `translators` extra and Pydantic V2 models (`PYDANTIC_V2=True`):
|
|
218
|
+
|
|
219
|
+
```sh
|
|
220
|
+
pip install "aidial-sdk[translators]"
|
|
221
|
+
```
|
|
222
|
+
|
|
223
|
+
Mount it on your application:
|
|
224
|
+
|
|
225
|
+
```python
|
|
226
|
+
import httpx
|
|
227
|
+
|
|
228
|
+
from aidial_sdk import DIALApp
|
|
229
|
+
from aidial_sdk.translators.messages_to_chat_completions import mount_translator
|
|
230
|
+
|
|
231
|
+
app = DIALApp()
|
|
232
|
+
mount_translator(app, httpx.AsyncClient())
|
|
233
|
+
```
|
|
234
|
+
|
|
235
|
+
The second argument is the HTTP client the translator calls DIAL Core with: either an `httpx.AsyncClient` or an async function that takes the incoming `Request` and returns one. The `path` (`/openaiChatCompletions` by default) and `name` keyword arguments configure the mount.
|
|
236
|
+
|
|
237
|
+
|Endpoint|Calls on DIAL Core|
|
|
238
|
+
|---|---|
|
|
239
|
+
|`POST /openaiChatCompletions/anthropic/v1/messages`|`POST {DIAL_URL}/openai/v1/chat/completions`|
|
|
240
|
+
|`POST /openaiChatCompletions/legacy/anthropic/v1/messages`|`POST {DIAL_URL}/openai/deployments/{deployment}/chat/completions`|
|
|
241
|
+
|
|
242
|
+
`DIAL_URL` must be set, otherwise the endpoints answer `500`.
|
|
243
|
+
|
|
244
|
+
To expose a Chat Completions model under `/anthropic/v1/messages`, add an `anthropicMessages` interface to its DIAL Core config, pointing `base_url` at the mount path of the application that runs the translator (or at `/openaiChatCompletions/legacy` to call DIAL Core's legacy per-deployment endpoint):
|
|
245
|
+
|
|
246
|
+
<details>
|
|
247
|
+
<summary>DIAL Core config example</summary>
|
|
248
|
+
|
|
249
|
+
```json
|
|
250
|
+
{
|
|
251
|
+
"models": {
|
|
252
|
+
"gpt-5.5": {
|
|
253
|
+
"type": "chat",
|
|
254
|
+
"interfaces": {
|
|
255
|
+
"openaiChatCompletions": {
|
|
256
|
+
"base_url": "${OPENAI_ADAPTER_ORIGIN}"
|
|
257
|
+
},
|
|
258
|
+
"anthropicMessages": {
|
|
259
|
+
"base_url": "${TRANSLATOR_ORIGIN}/openaiChatCompletions"
|
|
260
|
+
}
|
|
261
|
+
},
|
|
262
|
+
"upstreams": [
|
|
263
|
+
{
|
|
264
|
+
"endpoint": "https://api.openai.com/v1/chat/completions",
|
|
265
|
+
"key": "..."
|
|
266
|
+
}
|
|
267
|
+
]
|
|
268
|
+
}
|
|
269
|
+
}
|
|
270
|
+
}
|
|
271
|
+
```
|
|
272
|
+
|
|
273
|
+
</details>
|
|
274
|
+
|
|
275
|
+
See the [translation guide](docs/messages-to-chat-translation.md) for how request parameters, content blocks, tools, caching, usage, and errors are translated, and for the translator's limitations.
|
|
276
|
+
|
|
277
|
+
---
|
|
278
|
+
|
|
170
279
|
## Development
|
|
171
280
|
|
|
172
281
|
### Development Environment
|
|
@@ -269,3 +378,4 @@ make publish
|
|
|
269
378
|
```
|
|
270
379
|
|
|
271
380
|
---
|
|
381
|
+
|
|
@@ -1,42 +1,3 @@
|
|
|
1
|
-
Metadata-Version: 2.4
|
|
2
|
-
Name: aidial-sdk
|
|
3
|
-
Version: 0.46.0.dev8
|
|
4
|
-
Summary: Framework to create applications and model adapters for AI DIAL
|
|
5
|
-
License-Expression: Apache-2.0
|
|
6
|
-
License-File: LICENSE
|
|
7
|
-
Author: EPAM RAIL
|
|
8
|
-
Author-email: SpecialEPM-DIALDevTeam@epam.com
|
|
9
|
-
Requires-Python: >=3.10,<4.0
|
|
10
|
-
Classifier: Programming Language :: Python :: 3
|
|
11
|
-
Classifier: Programming Language :: Python :: 3.10
|
|
12
|
-
Classifier: Programming Language :: Python :: 3.11
|
|
13
|
-
Classifier: Programming Language :: Python :: 3.12
|
|
14
|
-
Classifier: Programming Language :: Python :: 3.13
|
|
15
|
-
Classifier: Programming Language :: Python :: 3.14
|
|
16
|
-
Provides-Extra: httpx
|
|
17
|
-
Provides-Extra: telemetry
|
|
18
|
-
Requires-Dist: fastapi (>=0.51,<1.0)
|
|
19
|
-
Requires-Dist: httpx (>=0.25.0,<1.0) ; extra == "httpx"
|
|
20
|
-
Requires-Dist: opentelemetry-api (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
21
|
-
Requires-Dist: opentelemetry-exporter-otlp-proto-grpc (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
22
|
-
Requires-Dist: opentelemetry-exporter-prometheus (>=0.43b0) ; extra == "telemetry"
|
|
23
|
-
Requires-Dist: opentelemetry-instrumentation-aiohttp-client (>=0.43b0) ; extra == "telemetry"
|
|
24
|
-
Requires-Dist: opentelemetry-instrumentation-fastapi (>=0.43b0) ; extra == "telemetry"
|
|
25
|
-
Requires-Dist: opentelemetry-instrumentation-httpx (>=0.43b0) ; extra == "telemetry"
|
|
26
|
-
Requires-Dist: opentelemetry-instrumentation-logging (>=0.43b0) ; extra == "telemetry"
|
|
27
|
-
Requires-Dist: opentelemetry-instrumentation-requests (>=0.43b0) ; extra == "telemetry"
|
|
28
|
-
Requires-Dist: opentelemetry-instrumentation-system-metrics (>=0.43b0) ; extra == "telemetry"
|
|
29
|
-
Requires-Dist: opentelemetry-instrumentation-urllib (>=0.43b0) ; extra == "telemetry"
|
|
30
|
-
Requires-Dist: opentelemetry-sdk (>=1.22.0,<2.0) ; extra == "telemetry"
|
|
31
|
-
Requires-Dist: prometheus-client (>=0.17.1,<1.0) ; extra == "telemetry"
|
|
32
|
-
Requires-Dist: pydantic (>=1.10.17,<3)
|
|
33
|
-
Requires-Dist: uvicorn (>=0.19,<1.0)
|
|
34
|
-
Requires-Dist: wrapt (>=1.10,<3)
|
|
35
|
-
Project-URL: Documentation, https://dialx.ai/dial_api
|
|
36
|
-
Project-URL: Homepage, https://dialx.ai
|
|
37
|
-
Project-URL: Repository, https://github.com/epam/ai-dial-sdk
|
|
38
|
-
Description-Content-Type: text/markdown
|
|
39
|
-
|
|
40
1
|
<h1 align="center">
|
|
41
2
|
AI DIAL Python SDK
|
|
42
3
|
</h1>
|
|
@@ -62,6 +23,7 @@ Description-Content-Type: text/markdown
|
|
|
62
23
|
- [Run](#run)
|
|
63
24
|
- [Check](#check)
|
|
64
25
|
- [Endpoints](#endpoints)
|
|
26
|
+
- [Anthropic Messages to Chat Completions translator](#anthropic-messages-to-chat-completions-translator)
|
|
65
27
|
- [Development](#development)
|
|
66
28
|
- [Development Environment](#development-environment)
|
|
67
29
|
- [Setup](#setup)
|
|
@@ -206,6 +168,72 @@ An `/openai/v1` request with none of the headers set fails with `500 Internal Se
|
|
|
206
168
|
|
|
207
169
|
---
|
|
208
170
|
|
|
171
|
+
### Anthropic Messages to Chat Completions translator
|
|
172
|
+
|
|
173
|
+
The translator lets an Anthropic-native client (e.g. Claude Code) talk to a model that only speaks the [Chat Completions API](https://platform.openai.com/docs/api-reference/chat). It converts an Anthropic Messages request into a Chat Completions request to DIAL Core at `DIAL_URL`, and converts the response, streaming or not, back into the Anthropic Messages shape.
|
|
174
|
+
|
|
175
|
+
The translator requires the `translators` extra and Pydantic V2 models (`PYDANTIC_V2=True`):
|
|
176
|
+
|
|
177
|
+
```sh
|
|
178
|
+
pip install "aidial-sdk[translators]"
|
|
179
|
+
```
|
|
180
|
+
|
|
181
|
+
Mount it on your application:
|
|
182
|
+
|
|
183
|
+
```python
|
|
184
|
+
import httpx
|
|
185
|
+
|
|
186
|
+
from aidial_sdk import DIALApp
|
|
187
|
+
from aidial_sdk.translators.messages_to_chat_completions import mount_translator
|
|
188
|
+
|
|
189
|
+
app = DIALApp()
|
|
190
|
+
mount_translator(app, httpx.AsyncClient())
|
|
191
|
+
```
|
|
192
|
+
|
|
193
|
+
The second argument is the HTTP client the translator calls DIAL Core with: either an `httpx.AsyncClient` or an async function that takes the incoming `Request` and returns one. The `path` (`/openaiChatCompletions` by default) and `name` keyword arguments configure the mount.
|
|
194
|
+
|
|
195
|
+
|Endpoint|Calls on DIAL Core|
|
|
196
|
+
|---|---|
|
|
197
|
+
|`POST /openaiChatCompletions/anthropic/v1/messages`|`POST {DIAL_URL}/openai/v1/chat/completions`|
|
|
198
|
+
|`POST /openaiChatCompletions/legacy/anthropic/v1/messages`|`POST {DIAL_URL}/openai/deployments/{deployment}/chat/completions`|
|
|
199
|
+
|
|
200
|
+
`DIAL_URL` must be set, otherwise the endpoints answer `500`.
|
|
201
|
+
|
|
202
|
+
To expose a Chat Completions model under `/anthropic/v1/messages`, add an `anthropicMessages` interface to its DIAL Core config, pointing `base_url` at the mount path of the application that runs the translator (or at `/openaiChatCompletions/legacy` to call DIAL Core's legacy per-deployment endpoint):
|
|
203
|
+
|
|
204
|
+
<details>
|
|
205
|
+
<summary>DIAL Core config example</summary>
|
|
206
|
+
|
|
207
|
+
```json
|
|
208
|
+
{
|
|
209
|
+
"models": {
|
|
210
|
+
"gpt-5.5": {
|
|
211
|
+
"type": "chat",
|
|
212
|
+
"interfaces": {
|
|
213
|
+
"openaiChatCompletions": {
|
|
214
|
+
"base_url": "${OPENAI_ADAPTER_ORIGIN}"
|
|
215
|
+
},
|
|
216
|
+
"anthropicMessages": {
|
|
217
|
+
"base_url": "${TRANSLATOR_ORIGIN}/openaiChatCompletions"
|
|
218
|
+
}
|
|
219
|
+
},
|
|
220
|
+
"upstreams": [
|
|
221
|
+
{
|
|
222
|
+
"endpoint": "https://api.openai.com/v1/chat/completions",
|
|
223
|
+
"key": "..."
|
|
224
|
+
}
|
|
225
|
+
]
|
|
226
|
+
}
|
|
227
|
+
}
|
|
228
|
+
}
|
|
229
|
+
```
|
|
230
|
+
|
|
231
|
+
</details>
|
|
232
|
+
|
|
233
|
+
See the [translation guide](docs/messages-to-chat-translation.md) for how request parameters, content blocks, tools, caching, usage, and errors are translated, and for the translator's limitations.
|
|
234
|
+
|
|
235
|
+
---
|
|
236
|
+
|
|
209
237
|
## Development
|
|
210
238
|
|
|
211
239
|
### Development Environment
|
|
@@ -308,4 +336,3 @@ make publish
|
|
|
308
336
|
```
|
|
309
337
|
|
|
310
338
|
---
|
|
311
|
-
|
|
@@ -24,7 +24,7 @@ from aidial_sdk.chat_completion.response import (
|
|
|
24
24
|
from aidial_sdk.deployment.configuration import ConfigurationRequest
|
|
25
25
|
from aidial_sdk.deployment.from_request_mixin import (
|
|
26
26
|
FromRequestMixin,
|
|
27
|
-
|
|
27
|
+
get_upstream_deployment_id,
|
|
28
28
|
)
|
|
29
29
|
from aidial_sdk.deployment.rate import RateRequest
|
|
30
30
|
from aidial_sdk.deployment.tokenize import TokenizeRequest
|
|
@@ -288,7 +288,7 @@ class DIALApp(FastAPI):
|
|
|
288
288
|
|
|
289
289
|
def _v1_dispatcher(self, endpoint: str) -> Handler:
|
|
290
290
|
async def _handler(original_request: Request) -> Response:
|
|
291
|
-
deployment_id =
|
|
291
|
+
deployment_id = get_upstream_deployment_id(
|
|
292
292
|
original_request.headers, None
|
|
293
293
|
)
|
|
294
294
|
set_log_deployment(deployment_id)
|
|
@@ -1,8 +1,6 @@
|
|
|
1
1
|
from collections.abc import Mapping
|
|
2
2
|
from enum import Enum
|
|
3
|
-
from typing import Annotated, Any, Literal
|
|
4
|
-
|
|
5
|
-
from typing_extensions import assert_never
|
|
3
|
+
from typing import Annotated, Any, Literal, assert_never
|
|
6
4
|
|
|
7
5
|
from aidial_sdk._pydantic import (
|
|
8
6
|
PYDANTIC_V2,
|
|
@@ -279,6 +277,8 @@ class ReasoningEffort(str, Enum):
|
|
|
279
277
|
LOW = "low"
|
|
280
278
|
MEDIUM = "medium"
|
|
281
279
|
HIGH = "high"
|
|
280
|
+
XHIGH = "xhigh"
|
|
281
|
+
MAX = "max"
|
|
282
282
|
|
|
283
283
|
|
|
284
284
|
class AzureChatCompletionRequest(ExtraAllowModel):
|
|
@@ -1,11 +1,9 @@
|
|
|
1
1
|
import asyncio
|
|
2
2
|
from collections.abc import Callable, Coroutine
|
|
3
3
|
from time import time
|
|
4
|
-
from typing import Any
|
|
4
|
+
from typing import Any, assert_never
|
|
5
5
|
from uuid import uuid4
|
|
6
6
|
|
|
7
|
-
from typing_extensions import assert_never
|
|
8
|
-
|
|
9
7
|
from aidial_sdk.chat_completion._cache import CacheBreakpointPath
|
|
10
8
|
from aidial_sdk.chat_completion._types import ChunkQueue
|
|
11
9
|
from aidial_sdk.chat_completion.choice import Choice
|
|
@@ -29,6 +27,7 @@ from aidial_sdk.deployment._headers import (
|
|
|
29
27
|
DIAL_CACHE_EXPIRE_AT,
|
|
30
28
|
DIAL_CACHE_EXTRA_METADATA,
|
|
31
29
|
)
|
|
30
|
+
from aidial_sdk.deployment.from_request_mixin import get_client_deployment_id
|
|
32
31
|
from aidial_sdk.exceptions import HTTPException as DIALException
|
|
33
32
|
from aidial_sdk.exceptions import RequestValidationError, RuntimeServerError
|
|
34
33
|
from aidial_sdk.utils._cancel_scope import CancelScope
|
|
@@ -71,6 +70,7 @@ class Response:
|
|
|
71
70
|
if self.request.stream
|
|
72
71
|
else "chat.completion"
|
|
73
72
|
),
|
|
73
|
+
model=get_client_deployment_id(request.headers),
|
|
74
74
|
)
|
|
75
75
|
|
|
76
76
|
self._headers = []
|
|
@@ -36,24 +36,27 @@ from aidial_sdk.utils.pydantic import ExtraAllowModel
|
|
|
36
36
|
T = TypeVar("T", bound="FromRequestMixin")
|
|
37
37
|
|
|
38
38
|
|
|
39
|
-
def
|
|
39
|
+
def get_upstream_deployment_id(
|
|
40
40
|
headers: Mapping[str, str], deployment_id: str | None
|
|
41
41
|
) -> str:
|
|
42
|
-
"""The deployment id
|
|
43
|
-
sent to or, for the deployment-agnostic paths, from its headers."""
|
|
42
|
+
"""The upstream deployment id that a DIAL application is expected to call."""
|
|
44
43
|
|
|
45
|
-
|
|
44
|
+
return (
|
|
46
45
|
headers.get(DIAL_OVERRIDE_NAME)
|
|
47
46
|
or deployment_id
|
|
48
|
-
or headers
|
|
47
|
+
or get_client_deployment_id(headers)
|
|
49
48
|
)
|
|
50
49
|
|
|
51
|
-
|
|
50
|
+
|
|
51
|
+
def get_client_deployment_id(headers: Mapping[str, str]) -> str:
|
|
52
|
+
"""The client-facing deployment id that a DIAL application is expected
|
|
53
|
+
to return as a model parameter in the response."""
|
|
54
|
+
|
|
55
|
+
if (deployment_id := headers.get(DIAL_DEPLOYMENT_ID)) is None:
|
|
52
56
|
raise InternalServerError(
|
|
53
57
|
f"The request headers are missing {DIAL_DEPLOYMENT_ID} header."
|
|
54
58
|
)
|
|
55
|
-
|
|
56
|
-
return effective_deployment_id
|
|
59
|
+
return deployment_id
|
|
57
60
|
|
|
58
61
|
|
|
59
62
|
class FromRequestMixin(ABC, ExtraAllowModel):
|
|
@@ -235,7 +238,7 @@ class FromRequestDeploymentMixin(FromRequestMixin):
|
|
|
235
238
|
bearer_token_secret=(
|
|
236
239
|
SecretStr(bearer_token) if bearer_token else None
|
|
237
240
|
),
|
|
238
|
-
deployment_id=
|
|
241
|
+
deployment_id=get_upstream_deployment_id(headers, deployment_id),
|
|
239
242
|
api_version=request.query_params.get("api-version"),
|
|
240
243
|
headers=headers,
|
|
241
244
|
original_request=request,
|
|
@@ -18,7 +18,7 @@ from opentelemetry.instrumentation.system_metrics import (
|
|
|
18
18
|
)
|
|
19
19
|
from opentelemetry.instrumentation.urllib import URLLibInstrumentor
|
|
20
20
|
from opentelemetry.metrics import set_meter_provider
|
|
21
|
-
from opentelemetry.sdk._logs import LoggerProvider
|
|
21
|
+
from opentelemetry.sdk._logs import LoggerProvider
|
|
22
22
|
from opentelemetry.sdk._logs.export import (
|
|
23
23
|
BatchLogRecordProcessor,
|
|
24
24
|
ConsoleLogRecordExporter,
|
|
@@ -86,13 +86,6 @@ def init_telemetry(app: FastAPI | None, config: TelemetryConfig):
|
|
|
86
86
|
except ImportError:
|
|
87
87
|
pass
|
|
88
88
|
|
|
89
|
-
set_logging_format = config.tracing.logging
|
|
90
|
-
if set_logging_format:
|
|
91
|
-
# Remove any competing handlers to avoid duplicate logging
|
|
92
|
-
remove_stream_handlers(logging.getLogger(), sys.stderr)
|
|
93
|
-
|
|
94
|
-
LoggingInstrumentor().instrument(set_logging_format=set_logging_format)
|
|
95
|
-
|
|
96
89
|
if config.logs is not None:
|
|
97
90
|
# Adding a handler to the root logger which exports the logs to OTLP or to the console as JSON.
|
|
98
91
|
provider = LoggerProvider(resource=resource)
|
|
@@ -120,7 +113,24 @@ def init_telemetry(app: FastAPI | None, config: TelemetryConfig):
|
|
|
120
113
|
)
|
|
121
114
|
|
|
122
115
|
set_logger_provider(provider)
|
|
123
|
-
|
|
116
|
+
|
|
117
|
+
if config.tracing is not None or config.logs is not None:
|
|
118
|
+
# OTel's text format writes to stderr, which `console_export` takes over.
|
|
119
|
+
set_logging_format = (
|
|
120
|
+
config.tracing is not None
|
|
121
|
+
and config.tracing.logging
|
|
122
|
+
and not (config.logs is not None and config.logs.console_export)
|
|
123
|
+
)
|
|
124
|
+
|
|
125
|
+
if set_logging_format:
|
|
126
|
+
# Remove any competing handlers to avoid duplicate logging
|
|
127
|
+
remove_stream_handlers(logging.getLogger(), sys.stderr)
|
|
128
|
+
|
|
129
|
+
LoggingInstrumentor().instrument(
|
|
130
|
+
set_logging_format=set_logging_format,
|
|
131
|
+
inject_trace_context=config.tracing is not None,
|
|
132
|
+
enable_log_auto_instrumentation=config.logs is not None,
|
|
133
|
+
)
|
|
124
134
|
|
|
125
135
|
if config.metrics is not None:
|
|
126
136
|
metric_readers = []
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
import httpx
|
|
2
|
+
from fastapi import FastAPI
|
|
3
|
+
|
|
4
|
+
from aidial_sdk._pydantic import PYDANTIC_V2
|
|
5
|
+
|
|
6
|
+
if not PYDANTIC_V2:
|
|
7
|
+
raise ImportError(
|
|
8
|
+
"The Anthropic Messages API translator requires Pydantic v2: "
|
|
9
|
+
"install pydantic>=2 and set PYDANTIC_V2=True"
|
|
10
|
+
)
|
|
11
|
+
|
|
12
|
+
from aidial_sdk.translators.messages_to_chat_completions.chat_completions.app import ( # noqa: E402
|
|
13
|
+
ClientFactory,
|
|
14
|
+
create_anthropic_api_app,
|
|
15
|
+
)
|
|
16
|
+
|
|
17
|
+
__all__ = ["ClientFactory", "create_anthropic_api_app", "mount_translator"]
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
def mount_translator(
|
|
21
|
+
app: FastAPI,
|
|
22
|
+
get_client: ClientFactory[httpx.AsyncClient],
|
|
23
|
+
*,
|
|
24
|
+
path: str = "/openaiChatCompletions",
|
|
25
|
+
name: str = "Anthropic Messages API to Chat Completions API translator",
|
|
26
|
+
) -> None:
|
|
27
|
+
translator = create_anthropic_api_app(get_client)
|
|
28
|
+
app.mount(path=path, app=translator, name=name)
|
|
@@ -0,0 +1,134 @@
|
|
|
1
|
+
import json
|
|
2
|
+
import logging
|
|
3
|
+
from collections.abc import Awaitable, Callable
|
|
4
|
+
from typing import TypeVar
|
|
5
|
+
|
|
6
|
+
import httpx
|
|
7
|
+
from anthropic.types.beta.message_create_params import MessageCreateParams
|
|
8
|
+
from fastapi import FastAPI, Request
|
|
9
|
+
from fastapi.responses import JSONResponse, Response, StreamingResponse
|
|
10
|
+
from openai import AsyncStream
|
|
11
|
+
from openai.types.chat import ChatCompletion, ChatCompletionChunk
|
|
12
|
+
|
|
13
|
+
from aidial_sdk.translators.messages_to_chat_completions.chat_completions.from_chat_completions import (
|
|
14
|
+
to_anthropic_response,
|
|
15
|
+
)
|
|
16
|
+
from aidial_sdk.translators.messages_to_chat_completions.chat_completions.streaming import (
|
|
17
|
+
translate_stream,
|
|
18
|
+
)
|
|
19
|
+
from aidial_sdk.translators.messages_to_chat_completions.chat_completions.to_chat_completions import (
|
|
20
|
+
to_chat_completions_request,
|
|
21
|
+
)
|
|
22
|
+
from aidial_sdk.translators.messages_to_chat_completions.common import (
|
|
23
|
+
not_found,
|
|
24
|
+
with_debug_logging,
|
|
25
|
+
)
|
|
26
|
+
from aidial_sdk.translators.messages_to_chat_completions.core_client import (
|
|
27
|
+
core_headers,
|
|
28
|
+
core_legacy_client,
|
|
29
|
+
core_v1_client,
|
|
30
|
+
)
|
|
31
|
+
from aidial_sdk.translators.messages_to_chat_completions.errors import (
|
|
32
|
+
AnthropicErrorType,
|
|
33
|
+
AnthropicHTTPError,
|
|
34
|
+
translator_error_handler,
|
|
35
|
+
)
|
|
36
|
+
from aidial_sdk.translators.messages_to_chat_completions.settings import (
|
|
37
|
+
get_dial_url,
|
|
38
|
+
)
|
|
39
|
+
from aidial_sdk.translators.messages_to_chat_completions.translation_state import (
|
|
40
|
+
TranslationState,
|
|
41
|
+
)
|
|
42
|
+
from aidial_sdk.utils.logging import logger as log
|
|
43
|
+
|
|
44
|
+
ClientT = TypeVar("ClientT")
|
|
45
|
+
ClientFactory = Callable[[Request], Awaitable[ClientT]] | ClientT
|
|
46
|
+
|
|
47
|
+
|
|
48
|
+
async def _handle_messages(
|
|
49
|
+
request: Request,
|
|
50
|
+
get_client: ClientFactory[httpx.AsyncClient],
|
|
51
|
+
legacy: bool,
|
|
52
|
+
) -> Response:
|
|
53
|
+
dial_url = get_dial_url()
|
|
54
|
+
translation_state = TranslationState()
|
|
55
|
+
|
|
56
|
+
try:
|
|
57
|
+
req: MessageCreateParams = await request.json()
|
|
58
|
+
except ValueError as e:
|
|
59
|
+
raise AnthropicHTTPError(
|
|
60
|
+
AnthropicErrorType.INVALID_REQUEST,
|
|
61
|
+
f"Request body is not valid JSON: {e}",
|
|
62
|
+
) from e
|
|
63
|
+
|
|
64
|
+
deployment = request.headers.get("x-dial-deployment-id") or req["model"]
|
|
65
|
+
chat_request = to_chat_completions_request(
|
|
66
|
+
req, deployment, translation_state
|
|
67
|
+
)
|
|
68
|
+
|
|
69
|
+
body = chat_request.model_dump(mode="json", exclude_none=True)
|
|
70
|
+
if log.isEnabledFor(logging.DEBUG):
|
|
71
|
+
log.debug(f"Translated request to Core: {json.dumps(body)}")
|
|
72
|
+
|
|
73
|
+
http_client = (
|
|
74
|
+
get_client
|
|
75
|
+
if isinstance(get_client, httpx.AsyncClient)
|
|
76
|
+
else await get_client(request)
|
|
77
|
+
)
|
|
78
|
+
client = (
|
|
79
|
+
core_legacy_client(dial_url, deployment, http_client)
|
|
80
|
+
if legacy
|
|
81
|
+
else core_v1_client(dial_url, http_client)
|
|
82
|
+
)
|
|
83
|
+
headers = core_headers(request)
|
|
84
|
+
|
|
85
|
+
if chat_request.stream:
|
|
86
|
+
events = await client.post(
|
|
87
|
+
"/chat/completions",
|
|
88
|
+
cast_to=ChatCompletion,
|
|
89
|
+
body=body,
|
|
90
|
+
options={"headers": headers},
|
|
91
|
+
stream=True,
|
|
92
|
+
stream_cls=AsyncStream[ChatCompletionChunk],
|
|
93
|
+
)
|
|
94
|
+
return StreamingResponse(
|
|
95
|
+
translate_stream(events, deployment, translation_state),
|
|
96
|
+
media_type="text/event-stream",
|
|
97
|
+
)
|
|
98
|
+
|
|
99
|
+
response = await client.post(
|
|
100
|
+
"/chat/completions",
|
|
101
|
+
cast_to=ChatCompletion,
|
|
102
|
+
body=body,
|
|
103
|
+
options={"headers": headers},
|
|
104
|
+
)
|
|
105
|
+
message = to_anthropic_response(response, deployment, translation_state)
|
|
106
|
+
return JSONResponse(
|
|
107
|
+
content=message.model_dump(mode="json", exclude_unset=True)
|
|
108
|
+
)
|
|
109
|
+
|
|
110
|
+
|
|
111
|
+
def create_anthropic_api_app(
|
|
112
|
+
get_client: ClientFactory[httpx.AsyncClient],
|
|
113
|
+
) -> FastAPI:
|
|
114
|
+
app = FastAPI()
|
|
115
|
+
|
|
116
|
+
@app.post("/anthropic/v1/messages")
|
|
117
|
+
@with_debug_logging
|
|
118
|
+
@translator_error_handler
|
|
119
|
+
async def _handle_v1_messages(request: Request) -> Response:
|
|
120
|
+
return await _handle_messages(request, get_client, legacy=False)
|
|
121
|
+
|
|
122
|
+
@app.post("/legacy/anthropic/v1/messages")
|
|
123
|
+
@with_debug_logging
|
|
124
|
+
@translator_error_handler
|
|
125
|
+
async def _handle_legacy_messages(request: Request) -> Response:
|
|
126
|
+
return await _handle_messages(request, get_client, legacy=True)
|
|
127
|
+
|
|
128
|
+
app.add_api_route(
|
|
129
|
+
"/{full_path:path}",
|
|
130
|
+
not_found,
|
|
131
|
+
methods=["GET", "POST", "PUT", "DELETE", "PATCH", "OPTIONS"],
|
|
132
|
+
)
|
|
133
|
+
|
|
134
|
+
return app
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
import re
|
|
2
|
+
from datetime import UTC, datetime, timedelta
|
|
3
|
+
|
|
4
|
+
from anthropic.types.beta import BetaCacheControlEphemeralParam as CacheControl
|
|
5
|
+
|
|
6
|
+
from aidial_sdk.chat_completion.request import CacheBreakpoint
|
|
7
|
+
from aidial_sdk.utils.logging import logger as log
|
|
8
|
+
|
|
9
|
+
_TTL = re.compile(r"^([1-9]\d*)([mh])$")
|
|
10
|
+
|
|
11
|
+
_UNITS = {
|
|
12
|
+
"m": timedelta(minutes=1),
|
|
13
|
+
"h": timedelta(hours=1),
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
|
|
17
|
+
def to_dial_cache_breakpoint(
|
|
18
|
+
controls: list[CacheControl],
|
|
19
|
+
) -> CacheBreakpoint | None:
|
|
20
|
+
if not controls:
|
|
21
|
+
return None
|
|
22
|
+
expires_at: str | None = None
|
|
23
|
+
for control in controls:
|
|
24
|
+
try:
|
|
25
|
+
if (ttl := _ttl(control)) is not None:
|
|
26
|
+
candidate = _expire_at(ttl)
|
|
27
|
+
expires_at = (
|
|
28
|
+
max(expires_at, candidate) if expires_at else candidate
|
|
29
|
+
)
|
|
30
|
+
except (OverflowError, ValueError):
|
|
31
|
+
log.warning(
|
|
32
|
+
f"Ignoring unreadable cache_control.ttl: {control.get('ttl')}"
|
|
33
|
+
)
|
|
34
|
+
return CacheBreakpoint(expire_at=expires_at)
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
def _ttl(control: CacheControl) -> timedelta | None:
|
|
38
|
+
raw = control.get("ttl")
|
|
39
|
+
if raw is None:
|
|
40
|
+
return None
|
|
41
|
+
if isinstance(raw, str) and (match := _TTL.match(raw)):
|
|
42
|
+
return int(match.group(1)) * _UNITS[match.group(2)]
|
|
43
|
+
raise ValueError("Invalid ttl")
|
|
44
|
+
|
|
45
|
+
|
|
46
|
+
def _expire_at(ttl: timedelta) -> str:
|
|
47
|
+
expires_at = datetime.now(UTC) + ttl
|
|
48
|
+
return expires_at.replace(microsecond=0).isoformat().replace("+00:00", "Z")
|