ommlds 0.0.0.dev466__py3-none-any.whl → 0.0.0.dev467__py3-none-any.whl

ommlds/.omlish-manifests.json

@@ -229,11 +229,12 @@
     "module": ".minichain.backends.impls.mlx.chat",
     "attr": null,
     "file": "ommlds/minichain/backends/impls/mlx/chat.py",
-    "line": 28,
+    "line": 39,
     "value": {
       "!.minichain.backends.strings.manifests.BackendStringsManifest": {
         "service_cls_names": [
-          "ChatChoicesService"
+          "ChatChoicesService",
+          "ChatChoicesStreamService"
         ],
         "backend_name": "mlx",
         "model_names": null
@@ -244,7 +245,7 @@
     "module": ".minichain.backends.impls.mlx.chat",
     "attr": null,
     "file": "ommlds/minichain/backends/impls/mlx/chat.py",
-    "line": 37,
+    "line": 133,
     "value": {
       "!.minichain.registries.manifests.RegistryManifest": {
         "module": "ommlds.minichain.backends.impls.mlx.chat",
@@ -255,6 +256,21 @@
       }
     }
   },
+  {
+    "module": ".minichain.backends.impls.mlx.chat",
+    "attr": null,
+    "file": "ommlds/minichain/backends/impls/mlx/chat.py",
+    "line": 164,
+    "value": {
+      "!.minichain.registries.manifests.RegistryManifest": {
+        "module": "ommlds.minichain.backends.impls.mlx.chat",
+        "attr": "MlxChatChoicesStreamService",
+        "name": "mlx",
+        "aliases": null,
+        "type": "ChatChoicesStreamService"
+      }
+    }
+  },
   {
     "module": ".minichain.backends.impls.openai.chat",
     "attr": null,
@@ -356,7 +372,7 @@
     "module": ".minichain.backends.impls.openai.stream",
     "attr": null,
     "file": "ommlds/minichain/backends/impls/openai/stream.py",
-    "line": 37,
+    "line": 38,
     "value": {
       "!.minichain.registries.manifests.RegistryManifest": {
         "module": "ommlds.minichain.backends.impls.openai.stream",

ommlds/cli/sessions/chat/backends/catalog.py

@@ -32,7 +32,7 @@ class _CatalogBackendProvider(BackendProvider[ServiceT], lang.Abstract):
     @contextlib.asynccontextmanager
     async def _provide_backend(self, cls: type[ServiceT]) -> ta.AsyncIterator[ServiceT]:
         service: ServiceT
-        async with lang.async_maybe_managing(self._catalog.get_backend(
+        async with lang.async_or_sync_maybe_managing(self._catalog.get_backend(
                 cls,
                 self._name,
                 *(self._configs or []),

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

@@ -1,3 +1,4 @@
+import contextlib
 import typing as ta
 
 from omlish import check
@@ -5,6 +6,7 @@ from omlish import lang
 from omlish import typedvalues as tv
 
 from .....backends import mlx as mlxu
+from ....chat.choices.services import ChatChoicesOutputs
 from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.services import static_check_is_chat_choices_service
@@ -14,19 +16,28 @@ 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 ....chat.stream.types import AiChoicesDeltas
+from ....chat.stream.types import ContentAiChoiceDelta
 from ....configs import Config
 from ....llms.types import MaxTokens
 from ....models.configs import ModelPath
 from ....models.configs import ModelRepo
 from ....models.configs import ModelSpecifier
+from ....resources import UseResources
 from ....standard import DefaultOptions
+from ....stream.services import StreamResponseSink
+from ....stream.services import new_stream_response
 
 
 ##
 
 
 # @omlish-manifest $.minichain.backends.strings.manifests.BackendStringsManifest(
-#     ['ChatChoicesService'],
+#     ['ChatChoicesService', 'ChatChoicesStreamService'],
 #     'mlx',
 # )
 
@@ -34,12 +45,7 @@ from ....standard import DefaultOptions
 ##
 
 
-# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
-#     name='mlx',
-#     type='ChatChoicesService',
-# )
-@static_check_is_chat_choices_service
-class MlxChatChoicesService(lang.ExitStacked):
+class BaseMlxChatChoicesService(lang.ExitStacked):
     DEFAULT_MODEL: ta.ClassVar[ModelSpecifier] = (
         # 'mlx-community/DeepSeek-Coder-V2-Lite-Instruct-8bit'
         # 'mlx-community/Llama-3.3-70B-Instruct-4bit'
@@ -52,8 +58,8 @@ class MlxChatChoicesService(lang.ExitStacked):
         # 'mlx-community/Qwen2.5-0.5B-4bit'
         # 'mlx-community/Qwen2.5-32B-Instruct-8bit'
         # 'mlx-community/Qwen2.5-Coder-32B-Instruct-8bit'
-        # 'mlx-community/mamba-2.8b-hf-f16'
         # 'mlx-community/Qwen3-30B-A3B-6bit'
+        # 'mlx-community/mamba-2.8b-hf-f16'
     )
 
     def __init__(self, *configs: Config) -> None:
@@ -70,10 +76,7 @@ class MlxChatChoicesService(lang.ExitStacked):
     }
 
     def _get_msg_content(self, m: Message) -> str | None:
-        if isinstance(m, AiMessage):
-            return check.isinstance(m.c, str)
-
-        elif isinstance(m, (SystemMessage, UserMessage)):
+        if isinstance(m, (AiMessage, SystemMessage, UserMessage)):
             return check.isinstance(m.c, str)
 
         else:
@@ -96,10 +99,9 @@ class MlxChatChoicesService(lang.ExitStacked):
         max_tokens=MaxTokens,
     )
 
-    async def invoke(self, request: ChatChoicesRequest) -> ChatChoicesResponse:
-        loaded_model = self._load_model()
-
-        tokenizer = loaded_model.tokenization.tokenizer
+    @lang.cached_function(transient=True)
+    def _get_tokenizer(self) -> mlxu.tokenization.Tokenizer:
+        tokenizer = self._load_model().tokenization.tokenizer
 
         if not (
                 hasattr(tokenizer, 'apply_chat_template') and
@@ -107,26 +109,44 @@ class MlxChatChoicesService(lang.ExitStacked):
         ):
             raise RuntimeError(tokenizer)
 
-        prompt = tokenizer.apply_chat_template(
+        return tokenizer
+
+    def _build_prompt(self, messages: ta.Sequence[Message]) -> str:
+        return check.isinstance(self._get_tokenizer().apply_chat_template(
             [  # type: ignore[arg-type]
                 dict(
                     role=self.ROLES_MAP[type(m)],
                     content=self._get_msg_content(m),
                 )
-                for m in request.v
+                for m in messages
             ],
             tokenize=False,
             add_generation_prompt=True,
-        )
+        ), str)
 
-        kwargs = dict()
+    def _build_kwargs(self, oc: tv.TypedValuesConsumer) -> dict[str, ta.Any]:
+        kwargs: dict[str, ta.Any] = {}
+        kwargs.update(oc.pop_scalar_kwargs(**self._OPTION_KWARG_NAMES_MAP))
+        return kwargs
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='mlx',
+#     type='ChatChoicesService',
+# )
+@static_check_is_chat_choices_service
+class MlxChatChoicesService(BaseMlxChatChoicesService):
+    async def invoke(self, request: ChatChoicesRequest) -> ChatChoicesResponse:
+        loaded_model = self._load_model()
+
+        prompt = self._build_prompt(request.v)
 
         with tv.consume(
                 *self._default_options,
                 *request.options,
                 override=True,
         ) as oc:
-            kwargs.update(oc.pop_scalar_kwargs(**self._OPTION_KWARG_NAMES_MAP))
+            kwargs = self._build_kwargs(oc)
 
         response = mlxu.generate(
             loaded_model.model,
@@ -139,3 +159,57 @@ class MlxChatChoicesService(lang.ExitStacked):
         return ChatChoicesResponse([
             AiChoice([AiMessage(response)])  # noqa
         ])
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='mlx',
+#     type='ChatChoicesStreamService',
+# )
+@static_check_is_chat_choices_stream_service
+class MlxChatChoicesStreamService(BaseMlxChatChoicesService):
+    def __init__(self, *configs: Config) -> None:
+        super().__init__()
+
+        with tv.consume(*configs) as cc:
+            self._model = cc.pop(MlxChatChoicesService.DEFAULT_MODEL)
+            self._default_options: tv.TypedValues = DefaultOptions.pop(cc)
+
+    READ_CHUNK_SIZE = 64 * 1024
+
+    async def invoke(
+            self,
+            request: ChatChoicesStreamRequest,
+            *,
+            max_tokens: int = 4096,  # FIXME: ChatOption
+    ) -> ChatChoicesStreamResponse:
+        loaded_model = self._load_model()
+
+        prompt = self._build_prompt(request.v)
+
+        with tv.consume(
+                *self._default_options,
+                *request.options,
+                override=True,
+        ) as oc:
+            oc.pop(UseResources, None)
+            kwargs = self._build_kwargs(oc)
+
+        async with UseResources.or_new(request.options) as rs:
+            gen: ta.Iterator[mlxu.GenerationOutput] = rs.enter_context(contextlib.closing(mlxu.stream_generate(
+                loaded_model.model,
+                loaded_model.tokenization,
+                check.isinstance(prompt, str),
+                mlxu.GenerationParams(**kwargs),
+                # verbose=True,
+            )))
+
+            async def inner(sink: StreamResponseSink[AiChoicesDeltas]) -> ta.Sequence[ChatChoicesOutputs]:
+                for go in gen:
+                    if go.text:
+                        await sink.emit(AiChoicesDeltas([AiChoiceDeltas([
+                            ContentAiChoiceDelta(go.text),
+                        ])]))
+
+                return []
+
+            return await new_stream_response(rs, inner)

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

@@ -26,8 +26,8 @@ from ....chat.choices.services import static_check_is_chat_choices_service
 from ....models.configs import ModelName
 from ....standard import ApiKey
 from ....standard import DefaultOptions
-from .format2 import OpenaiChatRequestHandler
-from .format2 import build_mc_choices_response
+from .format import OpenaiChatRequestHandler
+from .format import build_mc_choices_response
 from .names import MODEL_NAMES
 
 

ommlds/minichain/backends/impls/openai/format.py

@@ -2,18 +2,17 @@ import typing as ta
 
 from omlish import cached
 from omlish import check
-from omlish import lang
 from omlish import typedvalues as tv
 from omlish.formats import json
 
+from .....backends.openai import protocol as pt
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.types import AiChoice
+from ....chat.choices.types import AiChoices
 from ....chat.choices.types import ChatChoicesOptions
-from ....chat.messages import AiChat
 from ....chat.messages import AiMessage
 from ....chat.messages import AnyAiMessage
 from ....chat.messages import Chat
-from ....chat.messages import Message
 from ....chat.messages import SystemMessage
 from ....chat.messages import ToolUseMessage
 from ....chat.messages import ToolUseResultMessage
@@ -28,7 +27,7 @@ from ....llms.types import MaxTokens
 from ....llms.types import Temperature
 from ....llms.types import TokenUsage
 from ....llms.types import TokenUsageOutput
-from ....tools.jsonschema import build_tool_spec_json_schema
+from ....tools.jsonschema import build_tool_spec_params_json_schema
 from ....tools.types import ToolSpec
 from ....tools.types import ToolUse
 from ....types import Option
@@ -37,61 +36,115 @@ from ....types import Option
 ##
 
 
-def build_request_messages(chat: Chat) -> ta.Sequence[ta.Mapping[str, ta.Any]]:
-    out: list[dict[str, ta.Any]] = []
+def build_oai_request_msgs(mc_chat: Chat) -> ta.Sequence[pt.ChatCompletionMessage]:
+    oai_msgs: list[pt.ChatCompletionMessage] = []
 
-    for m in chat:
-        if isinstance(m, SystemMessage):
-            out.append(dict(
-                role='system',
-                content=m.c,
+    for mc_msg in mc_chat:
+        if isinstance(mc_msg, SystemMessage):
+            oai_msgs.append(pt.SystemChatCompletionMessage(
+                content=check.isinstance(mc_msg.c, str),
             ))
 
-        elif isinstance(m, AiMessage):
-            out.append(dict(
-                role='assistant',
-                content=check.isinstance(m.c, (str, None)),
+        elif isinstance(mc_msg, AiMessage):
+            oai_msgs.append(pt.AssistantChatCompletionMessage(
+                content=check.isinstance(mc_msg.c, (str, None)),
             ))
 
-        elif isinstance(m, ToolUseMessage):
-            out.append(dict(
-                role='assistant',
-                tool_calls=[
-                    dict(
-                        id=m.tu.id,
-                        function=dict(
-                            arguments=check.not_none(m.tu.raw_args),
-                            name=m.tu.name,
-                        ),
-                        type='function',
+        elif isinstance(mc_msg, ToolUseMessage):
+            oai_msgs.append(pt.AssistantChatCompletionMessage(
+                tool_calls=[pt.AssistantChatCompletionMessage.ToolCall(
+                    id=check.not_none(mc_msg.tu.id),
+                    function=pt.AssistantChatCompletionMessage.ToolCall.Function(
+                        arguments=check.not_none(mc_msg.tu.raw_args),
+                        name=mc_msg.tu.name,
                     ),
-                ],
+                )],
             ))
 
-        elif isinstance(m, UserMessage):
-            out.append(dict(
-                role='user',
-                content=prepare_content_str(m.c),
+        elif isinstance(mc_msg, UserMessage):
+            oai_msgs.append(pt.UserChatCompletionMessage(
+                content=prepare_content_str(mc_msg.c),
             ))
 
-        elif isinstance(m, ToolUseResultMessage):
+        elif isinstance(mc_msg, ToolUseResultMessage):
             tc: str
-            if isinstance(m.tur.c, str):
-                tc = m.tur.c
-            elif isinstance(m.tur.c, JsonContent):
-                tc = json.dumps_compact(m.tur.c)
+            if isinstance(mc_msg.tur.c, str):
+                tc = mc_msg.tur.c
+            elif isinstance(mc_msg.tur.c, JsonContent):
+                tc = json.dumps_compact(mc_msg.tur.c)
             else:
-                raise TypeError(m.tur.c)
-            out.append(dict(
-                role='tool',
-                tool_call_id=m.tur.id,
+                raise TypeError(mc_msg.tur.c)
+            oai_msgs.append(pt.ToolChatCompletionMessage(
+                tool_call_id=check.not_none(mc_msg.tur.id),
                 content=tc,
             ))
 
         else:
-            raise TypeError(m)
+            raise TypeError(mc_msg)
 
-    return out
+    return oai_msgs
+
+
+#
+
+
+def build_mc_ai_choice(oai_choice: pt.ChatCompletionResponseChoice) -> AiChoice:
+    cur: list[AnyAiMessage] = []
+
+    oai_msg = oai_choice.message
+
+    if (oai_c := oai_msg.content) is not None:
+        cur.append(AiMessage(check.isinstance(oai_c, str)))
+
+    for oai_tc in oai_msg.tool_calls or []:
+        cur.append(ToolUseMessage(ToolUse(
+            id=oai_tc.id,
+            name=oai_tc.function.name,
+            args=json.loads(oai_tc.function.arguments or '{}'),
+            raw_args=oai_tc.function.arguments,
+        )))
+
+    return AiChoice(cur)
+
+
+def build_mc_ai_choices(oai_resp: pt.ChatCompletionResponse) -> AiChoices:
+    return [
+        build_mc_ai_choice(oai_choice)
+        for oai_choice in oai_resp.choices
+    ]
+
+
+def build_mc_choices_response(oai_resp: pt.ChatCompletionResponse) -> ChatChoicesResponse:
+    return ChatChoicesResponse(
+        build_mc_ai_choices(oai_resp),
+
+        tv.TypedValues(
+            *([TokenUsageOutput(TokenUsage(
+                input=tu.prompt_tokens,
+                output=tu.completion_tokens,
+                total=tu.total_tokens,
+            ))] if (tu := oai_resp.usage) is not None else []),
+        ),
+    )
+
+
+def build_mc_ai_choice_delta(delta: pt.ChatCompletionChunkChoiceDelta) -> AiChoiceDelta:
+    if delta.content is not None:
+        check.state(not delta.tool_calls)
+        return ContentAiChoiceDelta(delta.content)
+
+    elif delta.tool_calls is not None:
+        check.state(delta.content is None)
+        tc = check.single(delta.tool_calls)
+        tc_fn = check.not_none(tc.function)
+        return PartialToolUseAiChoiceDelta(
+            id=tc.id,
+            name=tc_fn.name,
+            raw_args=tc_fn.arguments,
+        )
+
+    else:
+        raise ValueError(delta)
 
 
 ##
@@ -112,14 +165,6 @@ class OpenaiChatRequestHandler:
         self._model = model
         self._mandatory_kwargs = mandatory_kwargs
 
-    ROLES_MAP: ta.ClassVar[ta.Mapping[type[Message], str]] = {
-        SystemMessage: 'system',
-        UserMessage: 'user',
-        AiMessage: 'assistant',
-        ToolUseMessage: 'assistant',
-        ToolUseResultMessage: 'tool',
-    }
-
     DEFAULT_OPTIONS: ta.ClassVar[tv.TypedValues[Option]] = tv.TypedValues[Option](
         Temperature(0.),
         MaxTokens(1024),
@@ -162,72 +207,26 @@ class OpenaiChatRequestHandler:
         )
 
     @cached.function
-    def raw_request(self) -> ta.Mapping[str, ta.Any]:
+    def oai_request(self) -> pt.ChatCompletionRequest:
         po = self._process_options()
 
-        tools = [
-            dict(
-                type='function',
-                function=build_tool_spec_json_schema(ts),
+        tools: list[pt.ChatCompletionRequestTool] = [
+            pt.ChatCompletionRequestTool(
+                function=pt.ChatCompletionRequestTool.Function(
+                    name=check.not_none(ts.name),
+                    description=prepare_content_str(ts.desc),
+                    parameters=build_tool_spec_params_json_schema(ts),
+                ),
             )
             for ts in po.tools_by_name.values()
         ]
 
-        return dict(
+        return pt.ChatCompletionRequest(
             model=self._model,
-            messages=build_request_messages(self._chat),
+            messages=build_oai_request_msgs(self._chat),
             top_p=1,
-            **lang.opt_kw(tools=tools),
+            tools=tools or None,
             frequency_penalty=0.0,
             presence_penalty=0.0,
             **po.kwargs,
         )
-
-    def build_ai_chat(self, message: ta.Mapping[str, ta.Any]) -> AiChat:
-        out: list[AnyAiMessage] = []
-        if (c := message.get('content')) is not None:
-            out.append(AiMessage(c))
-        for tc in message.get('tool_calls', []):
-            out.append(ToolUseMessage(
-                ToolUse(
-                    id=tc['id'],
-                    name=tc['function']['name'],
-                    args=json.loads(tc['function']['arguments'] or '{}'),
-                    raw_args=tc['function']['arguments'],
-                ),
-            ))
-        return out
-
-    def build_response(self, raw_response: ta.Mapping[str, ta.Any]) -> ChatChoicesResponse:
-        return ChatChoicesResponse(
-            [
-                AiChoice(self.build_ai_chat(choice['message']))
-                for choice in raw_response['choices']
-            ],
-
-            tv.TypedValues(
-                *([TokenUsageOutput(TokenUsage(
-                    input=tu['prompt_tokens'],
-                    output=tu['completion_tokens'],
-                    total=tu['total_tokens'],
-                ))] if (tu := raw_response.get('usage')) is not None else []),
-            ),
-        )
-
-    def build_ai_choice_delta(self, delta: ta.Mapping[str, ta.Any]) -> AiChoiceDelta:
-        if (c := delta.get('content')) is not None:
-            check.state(not delta.get('tool_calls'))
-            return ContentAiChoiceDelta(c)
-
-        elif (tcs := delta.get('tool_calls')) is not None:  # noqa
-            check.state(delta.get('content') is None)
-            tc = check.single(tcs)
-            tc_fn = tc['function']
-            return PartialToolUseAiChoiceDelta(
-                id=tc.get('id'),
-                name=tc_fn.get('name'),
-                raw_args=tc_fn.get('arguments'),
-            )
-
-        else:
-            raise ValueError(delta)

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

@@ -11,7 +11,7 @@ from omlish.http import all as http
 from omlish.http import sse
 from omlish.io.buffers import DelimitingBuffer
 
-from .....backends.openai.protocol.chatcompletion.chunk import ChatCompletionChunk
+from .....backends.openai import protocol as pt
 from ....chat.choices.services import ChatChoicesOutputs
 from ....chat.stream.services import ChatChoicesStreamRequest
 from ....chat.stream.services import ChatChoicesStreamResponse
@@ -28,6 +28,7 @@ from ....stream.services import StreamResponseSink
 from ....stream.services import new_stream_response
 from .chat import OpenaiChatChoicesService
 from .format import OpenaiChatRequestHandler
+from .format import build_mc_ai_choice_delta
 from .names import MODEL_NAMES
 
 
@@ -62,13 +63,13 @@ class OpenaiChatChoicesStreamService:
             model=MODEL_NAMES.resolve(self._model_name.v),
             mandatory_kwargs=dict(
                 stream=True,
-                stream_options=dict(
+                stream_options=pt.ChatCompletionRequest.StreamOptions(
                     include_usage=True,
                 ),
             ),
         )
 
-        raw_request = rh.raw_request()
+        raw_request = msh.marshal(rh.oai_request())
 
         http_request = http.HttpRequest(
             'https://api.openai.com/v1/chat/completions',
@@ -105,20 +106,20 @@ class OpenaiChatChoicesStreamService:
 
                                 check.state(sj['object'] == 'chat.completion.chunk')
 
-                                ccc = msh.unmarshal(sj, ChatCompletionChunk)  # noqa
-                                # print(ccc)
+                                ccc = msh.unmarshal(sj, pt.ChatCompletionChunk)
 
                                 # FIXME: stop reason
-                                if not sj['choices']:
+                                if not ccc.choices:
                                     continue
 
-                                if any(choice['delta'] for choice in sj['choices']):
-                                    await sink.emit(AiChoicesDeltas([
-                                        AiChoiceDeltas(
-                                            [rh.build_ai_choice_delta(choice['delta'])] if choice['delta'] else [],
-                                        )
-                                        for choice in sj['choices']
-                                    ]))
+                                if any(choice.finish_reason for choice in ccc.choices):
+                                    check.state(all(choice.finish_reason for choice in ccc.choices))
+                                    break
+
+                                await sink.emit(AiChoicesDeltas([
+                                    AiChoiceDeltas([build_mc_ai_choice_delta(choice.delta)])
+                                    for choice in ccc.choices
+                                ]))
 
                     if not b:
                         return []

ommlds/minichain/chat/stream/types.py

@@ -42,6 +42,9 @@ class AiChoiceDelta(lang.Sealed, lang.Abstract):
     pass
 
 
+#
+
+
 @dc.dataclass(frozen=True)
 class ContentAiChoiceDelta(AiChoiceDelta, lang.Final):
     c: Content

{ommlds-0.0.0.dev466.dist-info → ommlds-0.0.0.dev467.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ommlds
-Version: 0.0.0.dev466
+Version: 0.0.0.dev467
 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.dev466
-Requires-Dist: omlish==0.0.0.dev466
+Requires-Dist: omdev==0.0.0.dev467
+Requires-Dist: omlish==0.0.0.dev467
 Provides-Extra: all
 Requires-Dist: llama-cpp-python~=0.3; extra == "all"
 Requires-Dist: mlx~=0.29; extra == "all"

{ommlds-0.0.0.dev466.dist-info → ommlds-0.0.0.dev467.dist-info}/RECORD

@@ -1,4 +1,4 @@
-ommlds/.omlish-manifests.json,sha256=ZrDlaAwG8hoshkjW-up0pk0dMvDI3g5dW1M92uaf5KI,17930
+ommlds/.omlish-manifests.json,sha256=MyJQsh5T1CMMXcGdxwqI9abQ8-j-ZlRGRluiikbeKRY,18414
 ommlds/__about__.py,sha256=uAJgr2I_m_oZPlV5P8XLFeYpBlEM-DdzeyF6O5OK_qs,1759
 ommlds/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/huggingface.py,sha256=JfEyfKOxU3-SY_ojtXBJFNeD-NIuKjvMe3GL3e93wNA,1175
@@ -91,7 +91,7 @@ ommlds/cli/sessions/chat/driver.py,sha256=ddnCYTKqWiPxV8U4UbFwb7E3yi81ItjZ9j3AJd
 ommlds/cli/sessions/chat/inject.py,sha256=7Yg6wUs2Oej4UjNZCAWCJCEsDJZWvT4G8XvkvVUMC7U,1928
 ommlds/cli/sessions/chat/session.py,sha256=eqwelLE74JFC-fBpk_hdwMD2nP4pLv3ZPwUn99200B8,521
 ommlds/cli/sessions/chat/backends/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/cli/sessions/chat/backends/catalog.py,sha256=GroogxBQf_zlrhEEwTSq18v13HWsuXOP276VcuphPic,1756
+ommlds/cli/sessions/chat/backends/catalog.py,sha256=gc03dqXEUUSi4WCWJ30HrkHPWlAkZHXkj1pOq7KVakU,1764
 ommlds/cli/sessions/chat/backends/inject.py,sha256=VbZ-Fb679kTItRpAhIYCqSM8vXUFeRDQWssUfrFgGi8,882
 ommlds/cli/sessions/chat/backends/injection.py,sha256=GCn5OvNIEowgB70kQVuU84z3i8lLA4vOVkTZlQG8s0o,327
 ommlds/cli/sessions/chat/backends/types.py,sha256=5eImYHXLKqbC5MDrN443eMGamP9snCmV1n7LtAsqgPk,696
@@ -190,15 +190,14 @@ ommlds/minichain/backends/impls/llamacpp/completion.py,sha256=oJ2I6wUoIPXYLm9Vc7
 ommlds/minichain/backends/impls/llamacpp/format.py,sha256=fcLMwk7r7FbNrYCH39G3fDRInKvlPIqcoxyLj95CooA,778
 ommlds/minichain/backends/impls/llamacpp/stream.py,sha256=uGog3xPNqCjGgyZjXEjhlxKbIbakWbapjANAEsmW-U4,3378
 ommlds/minichain/backends/impls/mlx/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/minichain/backends/impls/mlx/chat.py,sha256=kNIDvkvNpoB80LfA1y7UpSzEEm2Z4K2w56HOuMwT9zE,4558
+ommlds/minichain/backends/impls/mlx/chat.py,sha256=sMlhgiFZrxAC-kKkLSJ6c-2uJn0IHZXH4EiPET_-CKI,7458
 ommlds/minichain/backends/impls/openai/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/minichain/backends/impls/openai/chat.py,sha256=3hiKX2WqRo1cwF4AhcnzcCb2fsmwNLW7YPGYpask41A,2660
+ommlds/minichain/backends/impls/openai/chat.py,sha256=eMRjxPNrzrRjaw83LJuYzP9DGvwGyY2ObJSZub4Z9bY,2658
 ommlds/minichain/backends/impls/openai/completion.py,sha256=0XTC08mZzbW23Y2DNW2xfRR0eDX4nTyejF8CR1BdHZs,1756
 ommlds/minichain/backends/impls/openai/embedding.py,sha256=kkDJ3_0EqwQ_E0eXsSH1TuWXQmRqaijK8zG90fnlf3s,1582
-ommlds/minichain/backends/impls/openai/format.py,sha256=M1AYWDhz1QApazFeae4xTO9ng_59sx4uYs0FKt0GIKM,7275
-ommlds/minichain/backends/impls/openai/format2.py,sha256=OQ3N8VR4uL3PvHxjOQSdgg1bQ4_WiDz_sOV4WhVEXpQ,6611
+ommlds/minichain/backends/impls/openai/format.py,sha256=teGX8mNU3sXNWP4YWGD8d59M4X9_r75ImSzfTJgtNCM,7351
 ommlds/minichain/backends/impls/openai/names.py,sha256=b74t8FwSbGEveVtVz4SqM5tiRDyTKNlUKlseV6AX3Yo,1211
-ommlds/minichain/backends/impls/openai/stream.py,sha256=X45qIXgwAk7IVe4LL6gzL3uJivdaB-hUGutltHeswTc,5280
+ommlds/minichain/backends/impls/openai/stream.py,sha256=M7II7kZFsy33j8NQwdM1CCeKet3lw-XLOQdDzrzn-Yo,5297
 ommlds/minichain/backends/impls/sentencepiece/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/impls/sentencepiece/tokens.py,sha256=tUEBKyBgkTowssS_AdcAuPkyFzfyDfE935x4JG8PXM0,1602
 ommlds/minichain/backends/impls/tinygrad/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -231,7 +230,7 @@ ommlds/minichain/chat/stream/_marshal.py,sha256=r6NYBUviV7jIssaFprzv2rVEj8cFEuBl
 ommlds/minichain/chat/stream/adapters.py,sha256=3hKo3-MLtVIB-Nhdlxt17LP9vZESr-2fBZQ3Yr6l_Ps,1077
 ommlds/minichain/chat/stream/joining.py,sha256=oPxLT4qEYWCaxclnZvt54ztQP5md4V6u6Uwn4qd2e9M,2936
 ommlds/minichain/chat/stream/services.py,sha256=TxNEOm85QEFYtKb59q_uP6eSNh75v1fF-IpsJjhY4to,1252
-ommlds/minichain/chat/stream/types.py,sha256=t1udlFSMlSlEyQHRnBEQYI_f-FuE6twRBFGzR66blWQ,1585
+ommlds/minichain/chat/stream/types.py,sha256=kpHsWLNHk7hmaNPDSCqLH-ECSAiz83lRfr00LhSWb5U,1589
 ommlds/minichain/chat/tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/chat/tools/execution.py,sha256=tCPsz1kCt5RcoRX7dwfaJRvObniJJv_D2hCwz1Slo_A,573
 ommlds/minichain/chat/tools/ids.py,sha256=DFBKrpeDTCnMcU-P38VbPWX0YBDaz_HzMgx3yXWjFWQ,759
@@ -368,9 +367,9 @@ 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.dev466.dist-info/licenses/LICENSE,sha256=B_hVtavaA8zCYDW99DYdcpDLKz1n3BBRjZrcbv8uG8c,1451
-ommlds-0.0.0.dev466.dist-info/METADATA,sha256=cTdtmfR8ON19GS5ay_ImJD5oZ5uXwylxukZIaX7NNUM,3224
-ommlds-0.0.0.dev466.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-ommlds-0.0.0.dev466.dist-info/entry_points.txt,sha256=Z5YWtX7ClfiCKdW-dd_CSVvM0h4yQpJPi-2G3q6gNFo,35
-ommlds-0.0.0.dev466.dist-info/top_level.txt,sha256=Rbnk5d5wi58vnAXx13WFZqdQ4VX8hBCS2hEL3WeXOhY,7
-ommlds-0.0.0.dev466.dist-info/RECORD,,
+ommlds-0.0.0.dev467.dist-info/licenses/LICENSE,sha256=B_hVtavaA8zCYDW99DYdcpDLKz1n3BBRjZrcbv8uG8c,1451
+ommlds-0.0.0.dev467.dist-info/METADATA,sha256=NvYqf0PtfEdrj2en5RFFpTHB_cGjm41groVHgy54WZQ,3224
+ommlds-0.0.0.dev467.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+ommlds-0.0.0.dev467.dist-info/entry_points.txt,sha256=Z5YWtX7ClfiCKdW-dd_CSVvM0h4yQpJPi-2G3q6gNFo,35
+ommlds-0.0.0.dev467.dist-info/top_level.txt,sha256=Rbnk5d5wi58vnAXx13WFZqdQ4VX8hBCS2hEL3WeXOhY,7
+ommlds-0.0.0.dev467.dist-info/RECORD,,

ommlds/minichain/backends/impls/openai/format2.py

@@ -1,210 +0,0 @@
-import typing as ta
-
-from omlish import cached
-from omlish import check
-from omlish import typedvalues as tv
-from omlish.formats import json
-
-from .....backends.openai import protocol as pt
-from ....chat.choices.services import ChatChoicesResponse
-from ....chat.choices.types import AiChoice
-from ....chat.choices.types import AiChoices
-from ....chat.choices.types import ChatChoicesOptions
-from ....chat.messages import AiMessage
-from ....chat.messages import AnyAiMessage
-from ....chat.messages import Chat
-from ....chat.messages import SystemMessage
-from ....chat.messages import ToolUseMessage
-from ....chat.messages import ToolUseResultMessage
-from ....chat.messages import UserMessage
-from ....chat.tools.types import Tool
-from ....content.json import JsonContent
-from ....content.prepare import prepare_content_str
-from ....llms.types import MaxTokens
-from ....llms.types import Temperature
-from ....llms.types import TokenUsage
-from ....llms.types import TokenUsageOutput
-from ....tools.jsonschema import build_tool_spec_params_json_schema
-from ....tools.types import ToolSpec
-from ....tools.types import ToolUse
-from ....types import Option
-
-
-##
-
-
-def build_oai_request_msgs(mc_chat: Chat) -> ta.Sequence[pt.ChatCompletionMessage]:
-    oai_msgs: list[pt.ChatCompletionMessage] = []
-
-    for mc_msg in mc_chat:
-        if isinstance(mc_msg, SystemMessage):
-            oai_msgs.append(pt.SystemChatCompletionMessage(
-                content=check.isinstance(mc_msg.c, str),
-            ))
-
-        elif isinstance(mc_msg, AiMessage):
-            oai_msgs.append(pt.AssistantChatCompletionMessage(
-                content=check.isinstance(mc_msg.c, (str, None)),
-            ))
-
-        elif isinstance(mc_msg, ToolUseMessage):
-            oai_msgs.append(pt.AssistantChatCompletionMessage(
-                tool_calls=[pt.AssistantChatCompletionMessage.ToolCall(
-                    id=check.not_none(mc_msg.tu.id),
-                    function=pt.AssistantChatCompletionMessage.ToolCall.Function(
-                        arguments=check.not_none(mc_msg.tu.raw_args),
-                        name=mc_msg.tu.name,
-                    ),
-                )],
-            ))
-
-        elif isinstance(mc_msg, UserMessage):
-            oai_msgs.append(pt.UserChatCompletionMessage(
-                content=prepare_content_str(mc_msg.c),
-            ))
-
-        elif isinstance(mc_msg, ToolUseResultMessage):
-            tc: str
-            if isinstance(mc_msg.tur.c, str):
-                tc = mc_msg.tur.c
-            elif isinstance(mc_msg.tur.c, JsonContent):
-                tc = json.dumps_compact(mc_msg.tur.c)
-            else:
-                raise TypeError(mc_msg.tur.c)
-            oai_msgs.append(pt.ToolChatCompletionMessage(
-                tool_call_id=check.not_none(mc_msg.tur.id),
-                content=tc,
-            ))
-
-        else:
-            raise TypeError(mc_msg)
-
-    return oai_msgs
-
-
-#
-
-
-def build_mc_ai_choice(oai_choice: pt.ChatCompletionResponseChoice) -> AiChoice:
-    cur: list[AnyAiMessage] = []
-
-    oai_msg = oai_choice.message
-
-    if (oai_c := oai_msg.content) is not None:
-        cur.append(AiMessage(check.isinstance(oai_c, str)))
-
-    for oai_tc in oai_msg.tool_calls or []:
-        cur.append(ToolUseMessage(ToolUse(
-            id=oai_tc.id,
-            name=oai_tc.function.name,
-            args=json.loads(oai_tc.function.arguments or '{}'),
-            raw_args=oai_tc.function.arguments,
-        )))
-
-    return AiChoice(cur)
-
-
-def build_mc_ai_choices(oai_resp: pt.ChatCompletionResponse) -> AiChoices:
-    return [
-        build_mc_ai_choice(oai_choice)
-        for oai_choice in oai_resp.choices
-    ]
-
-
-def build_mc_choices_response(oai_resp: pt.ChatCompletionResponse) -> ChatChoicesResponse:
-    return ChatChoicesResponse(
-        build_mc_ai_choices(oai_resp),
-
-        tv.TypedValues(
-            *([TokenUsageOutput(TokenUsage(
-                input=tu.prompt_tokens,
-                output=tu.completion_tokens,
-                total=tu.total_tokens,
-            ))] if (tu := oai_resp.usage) is not None else []),
-        ),
-    )
-
-
-##
-
-
-class OpenaiChatRequestHandler:
-    def __init__(
-            self,
-            chat: Chat,
-            *options: ChatChoicesOptions,
-            model: str,
-            mandatory_kwargs: ta.Mapping[str, ta.Any] | None = None,
-    ) -> None:
-        super().__init__()
-
-        self._chat = chat
-        self._options = options
-        self._model = model
-        self._mandatory_kwargs = mandatory_kwargs
-
-    DEFAULT_OPTIONS: ta.ClassVar[tv.TypedValues[Option]] = tv.TypedValues[Option](
-        Temperature(0.),
-        MaxTokens(1024),
-    )
-
-    _OPTION_KWARG_NAMES_MAP: ta.ClassVar[ta.Mapping[str, type[ChatChoicesOptions]]] = dict(
-        temperature=Temperature,
-        max_tokens=MaxTokens,
-    )
-
-    class _ProcessedOptions(ta.NamedTuple):
-        kwargs: dict[str, ta.Any]
-        tools_by_name: dict[str, ToolSpec]
-
-    @cached.function
-    def _process_options(self) -> _ProcessedOptions:
-        kwargs: dict = dict(
-            temperature=0,
-            max_tokens=1024,
-        )
-
-        tools_by_name: dict[str, ToolSpec] = {}
-
-        with tv.TypedValues(*self._options).consume() as oc:
-            kwargs.update(oc.pop_scalar_kwargs(**self._OPTION_KWARG_NAMES_MAP))
-
-            for t in oc.pop(Tool, []):
-                if t.spec.name in tools_by_name:
-                    raise NameError(t.spec.name)
-                tools_by_name[check.non_empty_str(t.spec.name)] = t.spec
-
-        if (mk := self._mandatory_kwargs):
-            for k, v in mk.items():
-                check.not_in(k, kwargs)
-                kwargs[k] = v
-
-        return self._ProcessedOptions(
-            kwargs=kwargs,
-            tools_by_name=tools_by_name,
-        )
-
-    @cached.function
-    def oai_request(self) -> pt.ChatCompletionRequest:
-        po = self._process_options()
-
-        tools: list[pt.ChatCompletionRequestTool] = [
-            pt.ChatCompletionRequestTool(
-                function=pt.ChatCompletionRequestTool.Function(
-                    name=check.not_none(ts.name),
-                    description=prepare_content_str(ts.desc),
-                    parameters=build_tool_spec_params_json_schema(ts),
-                ),
-            )
-            for ts in po.tools_by_name.values()
-        ]
-
-        return pt.ChatCompletionRequest(
-            model=self._model,
-            messages=build_oai_request_msgs(self._chat),
-            top_p=1,
-            tools=tools or None,
-            frequency_penalty=0.0,
-            presence_penalty=0.0,
-            **po.kwargs,
-        )