aidial-sdk 0.40.0.dev4__tar.gz → 0.40.0.dev8__tar.gz

{aidial_sdk-0.40.0.dev4 → aidial_sdk-0.40.0.dev8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aidial-sdk
-Version: 0.40.0.dev4
+Version: 0.40.0.dev8
 Summary: Framework to create applications and model adapters for AI DIAL
 License-Expression: Apache-2.0
 License-File: LICENSE

{aidial_sdk-0.40.0.dev4 → aidial_sdk-0.40.0.dev8}/aidial_sdk/chat_completion/__init__.py

@@ -20,6 +20,8 @@ from aidial_sdk.chat_completion.request import (
     MessageContentRefusalPart,
     MessageContentTextPart,
     MessageCustomFields,
+    PromptCacheBreakpoint,
+    PromptCacheOptions,
     Request,
     ResponseFormat,
     ResponseFormatJsonObject,
@@ -77,6 +79,8 @@ __all__ = [
     "MessageContentRefusalPart",
     "MessageContentTextPart",
     "MessageCustomFields",
+    "PromptCacheBreakpoint",
+    "PromptCacheOptions",
     "Request",
     "RequestStage",
     "Response",

{aidial_sdk-0.40.0.dev4 → aidial_sdk-0.40.0.dev8}/aidial_sdk/chat_completion/request.py

@@ -81,14 +81,20 @@ class ImageURL(ExtraAllowModel):
     detail: Literal["auto", "low", "high"] | None = None
 
 
+class PromptCacheBreakpoint(ExtraAllowModel):
+    mode: Literal["explicit"]
+
+
 class MessageContentImagePart(ExtraAllowModel):
     type: Literal["image_url"]
     image_url: ImageURL
+    prompt_cache_breakpoint: PromptCacheBreakpoint | None = None
 
 
 class MessageContentTextPart(ExtraAllowModel):
     type: Literal["text"]
     text: StrictStr
+    prompt_cache_breakpoint: PromptCacheBreakpoint | None = None
 
 
 class InputFile(ExtraAllowModel):
@@ -100,6 +106,7 @@ class InputFile(ExtraAllowModel):
 class MessageContentFilePart(ExtraAllowModel):
     type: Literal["file"]
     file: InputFile
+    prompt_cache_breakpoint: PromptCacheBreakpoint | None = None
 
 
 class InputAudio(ExtraAllowModel):
@@ -110,6 +117,7 @@ class InputAudio(ExtraAllowModel):
 class MessageContentAudioPart(ExtraAllowModel):
     type: Literal["input_audio"]
     input_audio: InputAudio
+    prompt_cache_breakpoint: PromptCacheBreakpoint | None = None
 
 
 class MessageContentRefusalPart(ExtraAllowModel):
@@ -258,6 +266,11 @@ ResponseFormat = (
 )
 
 
+class PromptCacheOptions(ExtraAllowModel):
+    mode: Literal["implicit", "explicit"] | None = None
+    ttl: Literal["30m"] | StrictStr | None = None
+
+
 class ReasoningEffort(str, Enum):
     NONE = "none"
     MINIMAL = "minimal"
@@ -291,6 +304,8 @@ class AzureChatCompletionRequest(ExtraAllowModel):
     reasoning_effort: ReasoningEffort | None = None
     response_format: ResponseFormat | None = None
     parallel_tool_calls: StrictBool | None = None
+    prompt_cache_key: StrictStr | None = None
+    prompt_cache_options: PromptCacheOptions | None = None
 
 
 class ChatCompletionRequestCustomFields(ExtraAllowModel):

{aidial_sdk-0.40.0.dev4 → aidial_sdk-0.40.0.dev8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "aidial-sdk"
-version = "0.40.0.dev4"
+version = "0.40.0.dev8"
 description = "Framework to create applications and model adapters for AI DIAL"
 authors = [{ name = "EPAM RAIL", email = "SpecialEPM-DIALDevTeam@epam.com" }]
 license = "Apache-2.0"
@@ -42,7 +42,7 @@ nox = "^2023.4.22"
 pillow = "^12.3.0"
 httpx = "^0.25.0"
 respx = "^0.21.1"
-aiohttp = "^3.14.1"
+aiohttp = "^3.14.3"
 aiointercept = "^0.1.7"
 requests = "^2.33"
 responses = "^0.25.3"