ommlds 0.0.0.dev445__py3-none-any.whl → 0.0.0.dev447__py3-none-any.whl

ommlds/.omlish-manifests.json

@@ -37,7 +37,8 @@
     "value": {
       "!.minichain.backends.strings.manifests.BackendStringsManifest": {
         "service_cls_names": [
-          "ChatChoicesService"
+          "ChatChoicesService",
+          "ChatChoicesStreamService"
         ],
         "backend_name": "anthropic",
         "model_names": {
@@ -63,7 +64,7 @@
     "module": ".minichain.backends.impls.anthropic.stream",
     "attr": null,
     "file": "ommlds/minichain/backends/impls/anthropic/stream.py",
-    "line": 35,
+    "line": 33,
     "value": {
       "!.minichain.registries.manifests.RegistryManifest": {
         "module": "ommlds.minichain.backends.impls.anthropic.stream",
@@ -95,7 +96,7 @@
     "module": ".minichain.backends.impls.google.chat",
     "attr": null,
     "file": "ommlds/minichain/backends/impls/google/chat.py",
-    "line": 30,
+    "line": 29,
     "value": {
       "!.minichain.registries.manifests.RegistryManifest": {
         "module": "ommlds.minichain.backends.impls.google.chat",
@@ -133,6 +134,21 @@
       }
     }
   },
+  {
+    "module": ".minichain.backends.impls.google.stream",
+    "attr": null,
+    "file": "ommlds/minichain/backends/impls/google/stream.py",
+    "line": 34,
+    "value": {
+      "!.minichain.registries.manifests.RegistryManifest": {
+        "module": "ommlds.minichain.backends.impls.google.stream",
+        "attr": "GoogleChatChoicesStreamService",
+        "name": "google",
+        "aliases": null,
+        "type": "ChatChoicesStreamService"
+      }
+    }
+  },
   {
     "module": ".minichain.backends.impls.huggingface.repos",
     "attr": null,

ommlds/backends/google/protocol/types.py

@@ -9,67 +9,387 @@ from omlish import marshal as msh
 
 
 @dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
 @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-class GenerateContentRequest:
-    """https://ai.google.dev/api/generate-content#request-body"""
+class Blob(lang.Final):
+    mine_type: str
+    data: bytes
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class FunctionCall(lang.Final):
+    id: str
+    name: str
+    args: ta.Mapping[str, ta.Any] | None = None
+
+
+Scheduling: ta.TypeAlias = ta.Literal[
+    # This value is unused.
+    'SCHEDULING_UNSPECIFIED',
+
+    # Only add the result to the conversation context, do not interrupt or trigger generation.
+    'SILENT',
+
+    # Add the result to the conversation context, and prompt to generate output without interrupting ongoing
+    # generation.
+    'WHEN_IDLE',
+
+    # Add the result to the conversation context, interrupt ongoing generation and prompt to generate output.
+    'INTERRUPT',
+]
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class FunctionResponse(lang.Final):
+    id: str
+    name: str
+    response: ta.Mapping[str, ta.Any] | None = None
+    will_continue: bool | None = None
+    scheduling: Scheduling | None = None
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class FileData(lang.Final):
+    mime_type: str
+    file_uri: str
+
+
+Language: ta.TypeAlias = ta.Literal[
+    # Unspecified language. This value should not be used.
+    'LANGUAGE_UNSPECIFIED',
+
+    # Python >= 3.10, with numpy and simpy available.
+    'PYTHON',
+]
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class ExecutableCode(lang.Final):
+    language: Language
+    code: str
+
+
+Outcome: ta.TypeAlias = ta.Literal[
+    # Unspecified status. This value should not be used.
+    'OUTCOME_UNSPECIFIED',
+
+    # Code execution completed successfully.
+    'OUTCOME_OK',
+
+    # Code execution finished but with a failure. stderr should contain the reason.
+    'OUTCOME_FAILED',
+
+    # Code execution ran for too long, and was cancelled. There may or may not be a partial output present.
+    'OUTCOME_DEADLINE_EXCEEDED',
+]
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class CodeExecutionResult(lang.Final):
+    outcome: Outcome
+    output: str
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class VideoMetadata(lang.Final):
+    start_offset: str  # Duration
+    end_offset: str  # Duration
+    fps: float
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Part(lang.Final):
+    # TODO: data: msh.oneof ...
+    text: str | None = None
+    inline_data: Blob | None = None
+    function_call: FunctionCall | None = None
+    function_response: FunctionResponse | None = None
+    file_data: FileData | None = None
+    executable_code: ExecutableCode | None = None
+    code_execution_result: CodeExecutionResult | None = None
+
+    thought: bool | None = None
+    thought_signature: bytes | None = None
+
+    # TODO: metadata: msh.oneof ...
+    video_metadata: VideoMetadata | None = None
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Content(lang.Final):
+    parts: ta.Sequence[Part] | None = None
+    role: ta.Literal['user', 'model'] | None = None
+
+
+##
+
+
+Type: ta.TypeAlias = ta.Literal[
+    # Not specified, should not be used.
+    'TYPE_UNSPECIFIED',
+
+    # String type.
+    'STRING',
+
+    # Number type.
+    'NUMBER',
+
+    # Integer type.
+    'INTEGER',
+
+    # Boolean type.
+    'BOOLEAN',
+
+    # Array type.
+    'ARRAY',
+
+    # Object type.
+    'OBJECT',
+
+    # Null type.
+    'NULL',
+]
+
+
+Struct: ta.TypeAlias = ta.Mapping[str, 'Value']
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Value(lang.Abstract, lang.Sealed):
+    pass
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class NullValue(Value, lang.Final):
+    null_value: None = None
 
-    @dc.dataclass(frozen=True, kw_only=True)
-    @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-    class Content(lang.Final):
-        @dc.dataclass(frozen=True, kw_only=True)
-        @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-        class Part(lang.Final):
-            text: str
 
-        parts: ta.Sequence[Part]
-        role: ta.Literal['user', 'model']
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class NumberValue(Value, lang.Final):
+    number_value: float
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class StringValue(Value, lang.Final):
+    string_value: str
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class BoolValue(Value, lang.Final):
+    bool_value: bool
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class StructValue(Value, lang.Final):
+    struct_value: Struct
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class ListValue(Value, lang.Final):
+    list_value: ta.Sequence[Value]
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Schema(lang.Final):
+    type: Type | None = None
+    format: str | None = None
+    title: str | None = None
+    description: str | None = None
+    nullable: bool | None = None
+    enum: ta.Sequence[str] | None = None
+    max_items: str | None = None  # int64
+    min_items: str | None = None  # int64
+    properties: ta.Mapping[str, 'Schema'] | None = None
+    required: ta.Sequence[str] | None = None
+    min_properties: str | None = None  # int64
+    max_properties: str | None = None  # int64
+    min_length: str | None = None  # int64
+    max_length: str | None = None  # int64
+    pattern: str | None = None
+    example: Value | None = None
+    any_of: ta.Sequence['Schema'] | None = None
+    property_ordering: ta.Sequence[str] | None = None
+    default: Value | None = None
+    items: ta.Optional['Schema'] = None
+    minimum: float | None = None
+    maximum: float | None = None
+
+
+FunctionBehavior: ta.TypeAlias = ta.Literal[
+    #This value is unused.
+    'UNSPECIFIED',
+
+    # If set, the system will wait to receive the function response before continuing the conversation.
+    'BLOCKING',
+
+    # If set, the system will not wait to receive the function response. Instead, it will attempt to handle function
+    # responses as they become available while maintaining the conversation between the user and the model.
+    'NON_BLOCKING',
+]
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class FunctionDeclaration(lang.Final):
+    name: str
+    description: str
+
+    behavior: FunctionBehavior
+
+    parameters: Schema
+    parameters_json_schema: Value
+
+    response: Schema
+    response_json_schema: Value
+
+
+DynamicRetrievalMode: ta.TypeAlias = ta.Literal[
+    # Always trigger retrieval.
+    'MODE_UNSPECIFIED',
+
+    # Run retrieval only when system decides it is necessary.
+    'MODE_DYNAMIC',
+]
+
 
-    contents: ta.Sequence[Content]
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class DynamicRetrievalConfig(lang.Final):
+    mode: DynamicRetrievalMode | None = None
+
+    dynamic_threshold: int | float | None = None
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class GoogleSearchRetrieval(lang.Final):
+    dynamic_retrieval_config: DynamicRetrievalConfig
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class CodeExecution(lang.Final):
+    pass
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Interval(lang.Final):
+    start_time: str  # Timestamp
+    end_time: str  # Timestamp
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class GoogleSearch(lang.Final):
+    time_range_filter: Interval | None = None
 
 
 @dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
 @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-class GenerateContentResponse:
+class UrlContext(lang.Final):
+    pass
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class Tool(lang.Final):
+    function_declarations: ta.Sequence[FunctionDeclaration] | None = None
+    google_search_retrieval: GoogleSearchRetrieval | None = None
+    code_execution: CodeExecution | None = None
+    google_search: GoogleSearch | None = None
+    url_context: UrlContext | None = None
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class GenerateContentRequest(lang.Final):
+    """https://ai.google.dev/api/generate-content#request-body"""
+
+    contents: ta.Sequence[Content] | None = None
+
+
+@dc.dataclass(frozen=True, kw_only=True)
+@msh.update_fields_metadata(omit_if=lang.is_none)
+@msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
+class GenerateContentResponse(lang.Final):
     """https://ai.google.dev/api/generate-content#v1beta.GenerateContentResponse"""
 
     @dc.dataclass(frozen=True, kw_only=True)
+    @msh.update_fields_metadata(omit_if=lang.is_none)
     @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
     class Candidate(lang.Final):
-        @dc.dataclass(frozen=True, kw_only=True)
-        @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-        class Content(lang.Final):
-            @dc.dataclass(frozen=True, kw_only=True)
-            @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
-            class Part(lang.Final):
-                text: str
-
-            parts: ta.Sequence[Part]
-            role: ta.Literal['user', 'model']
-
-        content: Content
-        finish_reason: ta.Literal['STOP'] | None
-        index: int
+        content: Content | None = None
+        finish_reason: ta.Literal['STOP'] | None = None
+        index: int | None = None
 
-    candidates: ta.Sequence[Candidate]
+    candidates: ta.Sequence[Candidate] | None = None
 
     @dc.dataclass(frozen=True, kw_only=True)
+    @msh.update_fields_metadata(omit_if=lang.is_none)
     @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
     class UsageMetadata(lang.Final):
-        prompt_token_count: int
-        candidates_token_count: int
-        total_token_count: int
-        thoughts_token_count: int
+        prompt_token_count: int | None = None
+        cached_content_token_count: int | None = None
+        candidates_token_count: int | None = None
+        total_token_count: int | None = None
+        thoughts_token_count: int | None = None
 
         @dc.dataclass(frozen=True, kw_only=True)
+        @msh.update_fields_metadata(omit_if=lang.is_none)
         @msh.update_object_metadata(field_naming=msh.Naming.LOW_CAMEL)
         class ModalityTokenCount:
-            modality: str
-            token_count: int
+            modality: str | None = None
+            token_count: int | None = None
 
-        prompt_tokens_details: ta.Sequence[ModalityTokenCount]
+        prompt_tokens_details: ta.Sequence[ModalityTokenCount] | None = None
+        cache_tokens_details: ta.Sequence[ModalityTokenCount] | None = None
+        candidates_tokens_details: ta.Sequence[ModalityTokenCount] | None = None
+        tool_use_prompt_tokens_details: ta.Sequence[ModalityTokenCount] | None = None
 
-    usage_metadata: UsageMetadata
+    usage_metadata: UsageMetadata | None = None
 
-    model_version: str
+    model_version: str | None = None
 
-    response_id: str
+    response_id: str | None = None

ommlds/minichain/backends/impls/anthropic/names.py

@@ -34,7 +34,10 @@ MODEL_NAMES = ModelNameCollection(
 
 # @omlish-manifest
 _BACKEND_STRINGS_MANIFEST = BackendStringsManifest(
-    ['ChatChoicesService'],
+    [
+        'ChatChoicesService',
+        'ChatChoicesStreamService',
+    ],
     'anthropic',
     model_names=MODEL_NAMES,
 )

ommlds/minichain/backends/impls/anthropic/stream.py

@@ -9,8 +9,6 @@ from omlish.http import all as http
 from omlish.http import sse
 from omlish.io.buffers import DelimitingBuffer
 
-from .....backends.anthropic.protocol import types as pt
-from .....backends.anthropic.protocol.sse.assemble import AnthropicSseMessageAssembler
 from .....backends.anthropic.protocol.sse.events import AnthropicSseDecoderEvents
 from ....chat.choices.services import ChatChoicesOutputs
 from ....chat.messages import SystemMessage
@@ -89,9 +87,12 @@ class AnthropicChatChoicesStreamService:
             http_response = rs.enter_context(http_client.stream_request(http_request))
 
             async def inner(sink: StreamResponseSink[AiChoiceDeltas]) -> ta.Sequence[ChatChoicesOutputs] | None:
+                msg_start: AnthropicSseDecoderEvents.MessageStart | None = None
+                cbk_start: AnthropicSseDecoderEvents.ContentBlockStart | None = None
+                msg_stop: AnthropicSseDecoderEvents.MessageStop | None = None
+
                 db = DelimitingBuffer([b'\r', b'\n', b'\r\n'])
                 sd = sse.SseDecoder()
-                ass = AnthropicSseMessageAssembler()
                 while True:
                     # FIXME: read1 not on response stream protocol
                     b = http_response.stream.read1(self.READ_CHUNK_SIZE)  # type: ignore[attr-defined]
@@ -110,14 +111,56 @@ class AnthropicChatChoicesStreamService:
                                 dct = json.loads(ss)
                                 check.equal(dct['type'], so.type.decode('utf-8'))
                                 ae = msh.unmarshal(dct, AnthropicSseDecoderEvents.Event)
-                                for am in ass(ae):
-                                    if isinstance(am, pt.Message):
-                                        mt = check.isinstance(check.single(check.not_none(am.content)), pt.Text)
-                                        await sink.emit([
-                                            AiChoiceDelta(AiMessageDelta(mt.text)),
-                                        ])
+
+                                match ae:
+                                    case AnthropicSseDecoderEvents.MessageStart():
+                                        check.none(msg_start)
+                                        msg_start = ae
+                                        if msg_start.message.content:
+                                            raise NotImplementedError
+
+                                    case AnthropicSseDecoderEvents.ContentBlockStart():
+                                        check.not_none(msg_start)
+                                        check.none(cbk_start)
+                                        cbk_start = ae
+                                        if isinstance(ae.content_block, AnthropicSseDecoderEvents.ContentBlockStart.Text):  # noqa
+                                            await sink.emit([AiChoiceDelta(AiMessageDelta(
+                                                ae.content_block.text,
+                                            ))])
+                                        else:
+                                            raise TypeError(ae.content_block)
+
+                                    case AnthropicSseDecoderEvents.ContentBlockDelta():
+                                        check.not_none(cbk_start)
+                                        if isinstance(ae.delta, AnthropicSseDecoderEvents.ContentBlockDelta.TextDelta):
+                                            await sink.emit([AiChoiceDelta(AiMessageDelta(
+                                                ae.delta.text,
+                                            ))])
+                                        else:
+                                            raise TypeError(ae.delta)
+
+                                    case AnthropicSseDecoderEvents.ContentBlockStop():
+                                        check.not_none(cbk_start)
+                                        cbk_start = None
+
+                                    case AnthropicSseDecoderEvents.MessageDelta():
+                                        check.not_none(msg_start)
+                                        check.none(cbk_start)
+
+                                    case AnthropicSseDecoderEvents.MessageStop():
+                                        check.not_none(msg_start)
+                                        check.none(msg_stop)
+                                        msg_stop = ae
+
+                                    case AnthropicSseDecoderEvents.Ping():
+                                        pass
+
+                                    case _:
+                                        raise TypeError(ae)
 
                     if not b:
+                        check.not_none(msg_stop)
+                        check.none(cbk_start)
                         return []
 
             # raw_response = json.loads(check.not_none(http_response.data).decode('utf-8'))

ommlds/minichain/backends/impls/google/chat.py

@@ -9,8 +9,7 @@ from omlish import typedvalues as tv
 from omlish.formats import json
 from omlish.http import all as http
 
-from .....backends.google.protocol.types import GenerateContentRequest
-from .....backends.google.protocol.types import GenerateContentResponse
+from .....backends.google.protocol import types as pt
 from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.services import static_check_is_chat_choices_service
@@ -66,10 +65,10 @@ class GoogleChatChoicesService:
     ) -> ChatChoicesResponse:
         key = check.not_none(self._api_key).reveal()
 
-        g_req = GenerateContentRequest(
+        g_req = pt.GenerateContentRequest(
             contents=[
-                GenerateContentRequest.Content(
-                    parts=[GenerateContentRequest.Content.Part(
+                pt.Content(
+                    parts=[pt.Part(
                         text=check.not_none(self._get_msg_content(m)),
                     )],
                     role=self.ROLES_MAP[type(m)],  # type: ignore[arg-type]
@@ -91,9 +90,9 @@ class GoogleChatChoicesService:
 
         resp_dct = json.loads(check.not_none(resp.data).decode('utf-8'))
 
-        g_resp = msh.unmarshal(resp_dct, GenerateContentResponse)
+        g_resp = msh.unmarshal(resp_dct, pt.GenerateContentResponse)
 
         return ChatChoicesResponse([
-            AiChoice(AiMessage(c.content.parts[0].text))
-            for c in g_resp.candidates
+            AiChoice(AiMessage(check.not_none(check.not_none(check.not_none(c.content).parts)[0].text)))
+            for c in g_resp.candidates or []
         ])

ommlds/minichain/backends/impls/google/stream.py

@@ -0,0 +1,116 @@
+"""
+https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models
+"""
+import typing as ta
+
+from omlish import check
+from omlish import marshal as msh
+from omlish import typedvalues as tv
+from omlish.formats import json
+from omlish.http import all as http
+from omlish.io.buffers import DelimitingBuffer
+
+from .....backends.google.protocol import types as pt
+from ....chat.choices.types import ChatChoicesOutputs
+from ....chat.messages import AiMessage
+from ....chat.messages import Message
+from ....chat.messages import SystemMessage
+from ....chat.messages import UserMessage
+from ....chat.stream.services import ChatChoicesStreamRequest
+from ....chat.stream.services import ChatChoicesStreamResponse
+from ....chat.stream.services import static_check_is_chat_choices_stream_service
+from ....chat.stream.types import AiChoiceDeltas
+from ....models.configs import ModelName
+from ....resources import UseResources
+from ....standard import ApiKey
+from ....stream.services import StreamResponseSink
+from ....stream.services import new_stream_response
+from .names import MODEL_NAMES
+
+
+##
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='google',
+#     type='ChatChoicesStreamService',
+# )
+@static_check_is_chat_choices_stream_service
+class GoogleChatChoicesStreamService:
+    DEFAULT_MODEL_NAME: ta.ClassVar[ModelName] = ModelName(check.not_none(MODEL_NAMES.default))
+
+    def __init__(self, *configs: ApiKey | ModelName) -> None:
+        super().__init__()
+
+        with tv.consume(*configs) as cc:
+            self._model_name = cc.pop(self.DEFAULT_MODEL_NAME)
+            self._api_key = ApiKey.pop_secret(cc, env='GEMINI_API_KEY')
+
+    def _get_msg_content(self, m: Message) -> str | None:
+        if isinstance(m, AiMessage):
+            return check.isinstance(m.c, str)
+
+        elif isinstance(m, (SystemMessage, UserMessage)):
+            return check.isinstance(m.c, str)
+
+        else:
+            raise TypeError(m)
+
+    BASE_URL: ta.ClassVar[str] = 'https://generativelanguage.googleapis.com/v1beta/models'
+
+    ROLES_MAP: ta.ClassVar[ta.Mapping[type[Message], str]] = {
+        SystemMessage: 'system',
+        UserMessage: 'user',
+        AiMessage: 'assistant',
+    }
+
+    async def invoke(
+            self,
+            request: ChatChoicesStreamRequest,
+    ) -> ChatChoicesStreamResponse:
+        key = check.not_none(self._api_key).reveal()
+
+        g_req = pt.GenerateContentRequest(
+            contents=[
+                pt.Content(
+                    parts=[pt.Part(
+                        text=check.not_none(self._get_msg_content(m)),
+                    )],
+                    role=self.ROLES_MAP[type(m)],  # type: ignore[arg-type]
+                )
+                for m in request.v
+            ],
+        )
+
+        req_dct = msh.marshal(g_req)
+
+        model_name = MODEL_NAMES.resolve(self._model_name.v)
+
+        http_request = http.HttpRequest(
+            f'{self.BASE_URL.rstrip("/")}/{model_name}:generateContent?key={key}',
+            headers={'Content-Type': 'application/json'},
+            data=json.dumps_compact(req_dct).encode('utf-8'),
+            method='POST',
+        )
+
+        async with UseResources.or_new(request.options) as rs:
+            http_client = rs.enter_context(http.client())
+            http_response = rs.enter_context(http_client.stream_request(http_request))
+
+            async def inner(sink: StreamResponseSink[AiChoiceDeltas]) -> ta.Sequence[ChatChoicesOutputs] | None:
+                db = DelimitingBuffer([b'\r', b'\n', b'\r\n'])
+                while True:
+                    # FIXME: read1 not on response stream protocol
+                    b = http_response.stream.read1(self.READ_CHUNK_SIZE)  # type: ignore[attr-defined]
+                    for bl in db.feed(b):
+                        if isinstance(bl, DelimitingBuffer.Incomplete):
+                            # FIXME: handle
+                            return []
+                        l = bl.decode('utf-8')
+                        if not l:
+                            continue
+                        if l.startswith('data: '):
+                            gcr = msh.unmarshal(json.loads(l[6:]), pt.GenerateContentResponse)  # noqa
+                            await sink.emit([])
+
+            return await new_stream_response(rs, inner)

ommlds/minichain/stream/services.py

@@ -1,4 +1,5 @@
 import abc
+import itertools
 import typing as ta
 
 from omlish import check
@@ -124,13 +125,19 @@ class _StreamServiceResponse(StreamResponseIterator[V, OutputT]):
         self._state = 'closed'
         if old_state != 'running':
             return
-        if self._cr.cr_running:
+        if self._cr.cr_running or self._cr.cr_suspended:
             cex = StreamServiceCancelledError()
-            try:
-                self._g.throw(cex)
-            except StreamServiceCancelledError as cex2:
-                if cex2 is not cex:
+            for i in itertools.count():
+                try:
+                    if not i:
+                        x = self._g.throw(cex)
+                    else:
+                        x = self._g.send(None)
+                except StreamServiceCancelledError as cex2:
+                    if cex2 is cex:
+                        break
                     raise
+                await x
         if self._cr.cr_running:
             raise RuntimeError(f'Coroutine {self._cr!r} not terminated')
         if self._g is not self._a:

{ommlds-0.0.0.dev445.dist-info → ommlds-0.0.0.dev447.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ommlds
-Version: 0.0.0.dev445
+Version: 0.0.0.dev447
 Summary: ommlds
 Author: wrmsr
 License-Expression: BSD-3-Clause
@@ -14,8 +14,8 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.13
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: omdev==0.0.0.dev445
-Requires-Dist: omlish==0.0.0.dev445
+Requires-Dist: omdev==0.0.0.dev447
+Requires-Dist: omlish==0.0.0.dev447
 Provides-Extra: all
 Requires-Dist: llama-cpp-python~=0.3; extra == "all"
 Requires-Dist: mlx~=0.29; extra == "all"

{ommlds-0.0.0.dev445.dist-info → ommlds-0.0.0.dev447.dist-info}/RECORD

@@ -1,4 +1,4 @@
-ommlds/.omlish-manifests.json,sha256=F_mMa6P17FqHpbqQh1qWiqAgXIBnJrNfDML615bnPKk,17481
+ommlds/.omlish-manifests.json,sha256=0nHO21Xk5kpNbvdLvitrGGWPsn0baEhGD9I835QE_Es,17984
 ommlds/__about__.py,sha256=Z9VIVQnuNBbIYEtIm9XZU4T2QGRqMNjtmQX2OOTaUc0,1759
 ommlds/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/huggingface.py,sha256=JfEyfKOxU3-SY_ojtXBJFNeD-NIuKjvMe3GL3e93wNA,1175
@@ -15,7 +15,7 @@ ommlds/backends/anthropic/protocol/sse/assemble.py,sha256=LGA_zc04AP4xon9qYmUpDe
 ommlds/backends/anthropic/protocol/sse/events.py,sha256=7k5Vl44dMbyK5TSzEfh9eHdsVwNN9syVOhTCfTka7FE,2182
 ommlds/backends/google/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/backends/google/protocol/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/backends/google/protocol/types.py,sha256=tvHOx74JlVYSubQ3MZOcq-JhFoqZc4KCewNPmMueH_c,2321
+ommlds/backends/google/protocol/types.py,sha256=SmtuHAHfT1WReISUiGIe3ysma21DttihCsBNfoWeJ40,12007
 ommlds/backends/llamacpp/__init__.py,sha256=zXFpLXE4a2vEl0jcPDyKlPHHfZ3Z8Dz0twhEIyZ8-vg,59
 ommlds/backends/llamacpp/buildwheel.py,sha256=q9ghCLVbm8Jm6syrZlBP-x1qNDd0wSl15B2OXBtDBQ8,3813
 ommlds/backends/llamacpp/logging.py,sha256=nCEC6ASEuTpJqx47DMLhnbr5KelDlbxhM0nKQt4bc3w,773
@@ -125,14 +125,15 @@ ommlds/minichain/backends/impls/mistral.py,sha256=CqQIyH7UF7lNMZjzmrzTdBVf3LnpHe
 ommlds/minichain/backends/impls/sqlite.py,sha256=NOFm_fgr-OZ8mo7etj0zwvxsDnidRwKzhdDom58e6ks,2157
 ommlds/minichain/backends/impls/anthropic/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/impls/anthropic/chat.py,sha256=eyh7hWADTcQiMAHjYgjwoZZcpVxHLUhqSJhuraw27K8,3712
-ommlds/minichain/backends/impls/anthropic/names.py,sha256=CNZzE0EQwdIPGV0JNr5YbpuM1lf0l5AN1Su6XpwJ-LM,1098
-ommlds/minichain/backends/impls/anthropic/stream.py,sha256=I95OVhJ6y1AiZDwxRzK9W8pyNvzFIFmfyqhT4KmAQV0,5410
+ommlds/minichain/backends/impls/anthropic/names.py,sha256=vA1PpOSuH3cYfrTbc9GzWWZKYLHClnZY5vB5gW5Ag-M,1149
+ommlds/minichain/backends/impls/anthropic/stream.py,sha256=ubCe-28npGfhxJDSEm5zjroAjMyeZbXBG9E944ArRak,7764
 ommlds/minichain/backends/impls/duckduckgo/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/impls/duckduckgo/search.py,sha256=igzeU9P9b1MMiu4KAJVS9H6KLIoPm68wXi4Kx3_DHyQ,940
 ommlds/minichain/backends/impls/google/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/minichain/backends/impls/google/chat.py,sha256=dwdDtbJD1ghJvilZw9VLE3re3un7LM4n2Y1jbalwbxk,3269
+ommlds/minichain/backends/impls/google/chat.py,sha256=Dq_PrkRt07dN5A3UPj0qwB3IMIS9caAQff6076q6kUQ,3192
 ommlds/minichain/backends/impls/google/names.py,sha256=HxHJ31HeKZg6aW1C_Anqp-gamCXpq9pOdKj8_yVgE8Y,871
 ommlds/minichain/backends/impls/google/search.py,sha256=5-2nAZ1QmbqHSQcwWnqqcgCM-Duy2ryctJEIv2tcpZg,3260
+ommlds/minichain/backends/impls/google/stream.py,sha256=3XWZwYFQrA8U_QIexjJkjaWfLo2lfLbF_x7DsqcNRF4,4313
 ommlds/minichain/backends/impls/huggingface/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/impls/huggingface/configs.py,sha256=6jsBtPNXOP57PcpxNTVLGWLc-18Iwn_lDbGouwCJTIQ,258
 ommlds/minichain/backends/impls/huggingface/repos.py,sha256=8BDxJmra9elSQL2vzp2nr2p4Hpq56A3zTk7hTTnfJU4,861
@@ -241,7 +242,7 @@ ommlds/minichain/services/requests.py,sha256=VAfKbYu4T0CZTWVQmZ2LUmYU7DNm6IerYMN
 ommlds/minichain/services/responses.py,sha256=4W6Z4Fx4_GFqKgle27OeLr0zzjVTA0pkZrlsZiFQNdo,1534
 ommlds/minichain/services/services.py,sha256=WjkQNYIp87SflLSReOHMkG2qIVAOem6vsrs_2NxWN_M,325
 ommlds/minichain/stream/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/minichain/stream/services.py,sha256=oOdJ-GpkeXpb-kqEY7FZi_J0NzbVJk5VpPUk7ZIRFt4,5020
+ommlds/minichain/stream/services.py,sha256=OQ-CjCM8iKYReTg8aj0VK3XzZLdRude86kg6FNr-wew,5279
 ommlds/minichain/stream/wrap.py,sha256=nQC0aCi49I18nF0Yx8qiiLkhIAECV6s6o4pvOy5Kx98,2041
 ommlds/minichain/text/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/text/applypatch.py,sha256=YIN5JChJ0FXyK1I6OiAHQmE7BT-exHfaAMM9ay7ylyc,17705
@@ -299,9 +300,9 @@ ommlds/wiki/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU
 ommlds/wiki/utils/io.py,sha256=UKgDJGtmpnWvIqVd2mJc2QNPOqlToEY1GEveNp6_pMo,7088
 ommlds/wiki/utils/progress.py,sha256=EhvKcMFYtsarCQhIahlO6f0SboyAKP3UwUyrnVnP-Vk,3222
 ommlds/wiki/utils/xml.py,sha256=vVV8Ctn13aaRM9eYfs9Wd6rHn5WOCEUzQ44fIhOvJdg,3754
-ommlds-0.0.0.dev445.dist-info/licenses/LICENSE,sha256=B_hVtavaA8zCYDW99DYdcpDLKz1n3BBRjZrcbv8uG8c,1451
-ommlds-0.0.0.dev445.dist-info/METADATA,sha256=3gBk3VJaTAXENiec6FZ8zAClGnIb8AqH8mlqZKc6Miw,3224
-ommlds-0.0.0.dev445.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-ommlds-0.0.0.dev445.dist-info/entry_points.txt,sha256=Z5YWtX7ClfiCKdW-dd_CSVvM0h4yQpJPi-2G3q6gNFo,35
-ommlds-0.0.0.dev445.dist-info/top_level.txt,sha256=Rbnk5d5wi58vnAXx13WFZqdQ4VX8hBCS2hEL3WeXOhY,7
-ommlds-0.0.0.dev445.dist-info/RECORD,,
+ommlds-0.0.0.dev447.dist-info/licenses/LICENSE,sha256=B_hVtavaA8zCYDW99DYdcpDLKz1n3BBRjZrcbv8uG8c,1451
+ommlds-0.0.0.dev447.dist-info/METADATA,sha256=a2muFyUNAsNLhxd5yCTZTUo1ZwX7RpjX8HI03hB57Lc,3224
+ommlds-0.0.0.dev447.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+ommlds-0.0.0.dev447.dist-info/entry_points.txt,sha256=Z5YWtX7ClfiCKdW-dd_CSVvM0h4yQpJPi-2G3q6gNFo,35
+ommlds-0.0.0.dev447.dist-info/top_level.txt,sha256=Rbnk5d5wi58vnAXx13WFZqdQ4VX8hBCS2hEL3WeXOhY,7
+ommlds-0.0.0.dev447.dist-info/RECORD,,