ommlds 0.0.0.dev479__py3-none-any.whl → 0.0.0.dev481__py3-none-any.whl

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

@@ -17,6 +17,7 @@ from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.services import static_check_is_chat_choices_service
 from ....chat.choices.types import AiChoice
+from ....chat.choices.types import ChatChoicesOptions
 from ....chat.messages import AiMessage
 from ....chat.messages import AnyAiMessage
 from ....chat.messages import Message
@@ -24,9 +25,12 @@ from ....chat.messages import SystemMessage
 from ....chat.messages import ToolUseMessage
 from ....chat.messages import UserMessage
 from ....chat.tools.types import Tool
+from ....llms.types import MaxTokens
+from ....llms.types import Temperature
 from ....models.configs import ModelName
 from ....standard import ApiKey
 from ....tools.types import ToolUse
+from ....types import Option
 from .names import MODEL_NAMES
 from .protocol import build_protocol_chat_messages
 from .protocol import build_protocol_tool
@@ -67,16 +71,31 @@ class AnthropicChatChoicesService:
         else:
             raise TypeError(m)
 
+    DEFAULT_OPTIONS: ta.ClassVar[tv.TypedValues[Option]] = tv.TypedValues[Option](
+        MaxTokens(4096),
+    )
+
+    _OPTION_KWARG_NAMES_MAP: ta.ClassVar[ta.Mapping[str, type[ChatChoicesOptions]]] = dict(
+        temperature=Temperature,
+        max_tokens=MaxTokens,
+    )
+
     async def invoke(
             self,
             request: ChatChoicesRequest,
-            *,
-            max_tokens: int = 4096,  # FIXME: ChatOption
     ) -> ChatChoicesResponse:
         messages, system = build_protocol_chat_messages(request.v)
 
+        kwargs: dict = dict()
+
         tools: list[pt.ToolSpec] = []
-        with tv.TypedValues(*request.options).consume() as oc:
+        with tv.TypedValues(
+            *self.DEFAULT_OPTIONS,
+                *request.options,
+                override=True,
+        ).consume() as oc:
+            kwargs.update(oc.pop_scalar_kwargs(**self._OPTION_KWARG_NAMES_MAP))
+
             t: Tool
             for t in oc.pop(Tool, []):
                 tools.append(build_protocol_tool(t))
@@ -86,7 +105,7 @@ class AnthropicChatChoicesService:
             system=system,
             messages=messages,
             tools=tools or None,
-            max_tokens=max_tokens,
+            **kwargs,
         )
 
         raw_request = msh.marshal(a_req)

ommlds/minichain/backends/impls/duckduckgo/search.py

@@ -1,4 +1,4 @@
-import ddgs
+from omlish import lang
 
 from ....search import SearchHit
 from ....search import SearchHits
@@ -7,6 +7,10 @@ from ....search import SearchResponse
 from ....search import static_check_is_search_service
 
 
+with lang.auto_proxy_import(globals()):
+    import ddgs
+
+
 ##
 
 

ommlds/minichain/backends/impls/huggingface/repos.py

@@ -3,8 +3,6 @@ TODO:
  - local-only check first
   - cat ~/.cache/.../models/.../refs/main -> c5bfd839cd4cda0e5a39a97e00218d9c56e468af
 """
-import typing as ta
-
 from omlish import lang
 
 from ....models.configs import ModelRepo
@@ -12,10 +10,8 @@ from ....models.repos.resolving import ModelRepoResolver
 from ....models.repos.resolving import ResolvedModelRepo
 
 
-if ta.TYPE_CHECKING:
+with lang.auto_proxy_import(globals()):
     import huggingface_hub as hf
-else:
-    hf = lang.proxy_import('huggingface_hub')
 
 
 ##

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

@@ -2,13 +2,10 @@ import contextlib
 import os.path
 import typing as ta
 
-import llama_cpp as lcc
-
 from omlish import check
 from omlish import lang
 from omlish import typedvalues as tv
 
-from .....backends import llamacpp as lcu
 from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.services import static_check_is_chat_choices_service
@@ -27,6 +24,12 @@ from .format import ROLES_MAP
 from .format import get_msg_content
 
 
+with lang.auto_proxy_import(globals()):
+    import llama_cpp as lcc
+
+    from .....backends import llamacpp as lcu
+
+
 ##
 
 

ommlds/minichain/backends/impls/llamacpp/completion.py

@@ -2,11 +2,9 @@ import contextlib
 import os.path
 import typing as ta
 
-import llama_cpp as lcc
-
+from omlish import lang
 from omlish import typedvalues as tv
 
-from .....backends import llamacpp as lcu
 from ....completion import CompletionOption
 from ....completion import CompletionRequest
 from ....completion import CompletionResponse
@@ -18,6 +16,12 @@ from ....llms.types import Temperature
 from ....models.configs import ModelPath
 
 
+with lang.auto_proxy_import(globals()):
+    import llama_cpp as lcc
+
+    from .....backends import llamacpp as lcu
+
+
 ##
 
 

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

@@ -2,13 +2,10 @@ import contextlib
 import threading
 import typing as ta  # noqa
 
-import llama_cpp as lcc
-
 from omlish import check
 from omlish import lang
 from omlish import typedvalues as tv
 
-from .....backends import llamacpp as lcu
 from ....chat.choices.services import ChatChoicesOutputs
 from ....chat.choices.stream.services import ChatChoicesStreamRequest
 from ....chat.choices.stream.services import ChatChoicesStreamResponse
@@ -26,6 +23,12 @@ from .format import ROLES_MAP
 from .format import get_msg_content
 
 
+with lang.auto_proxy_import(globals()):
+    import llama_cpp as lcc
+
+    from .....backends import llamacpp as lcu
+
+
 ##
 
 

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

@@ -5,7 +5,6 @@ from omlish import check
 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
@@ -33,6 +32,10 @@ from ....stream.services import StreamResponseSink
 from ....stream.services import new_stream_response
 
 
+with lang.auto_proxy_import(globals()):
+    from .....backends import mlx as mlxu
+
+
 ##
 
 
@@ -83,7 +86,7 @@ class BaseMlxChatChoicesService(lang.ExitStacked):
             raise TypeError(m)
 
     @lang.cached_function(transient=True)
-    def _load_model(self) -> mlxu.LoadedModel:
+    def _load_model(self) -> 'mlxu.LoadedModel':
         # FIXME: walk state, find all mx.arrays, dealloc/set to empty
         check.not_none(self._exit_stack)
 
@@ -100,7 +103,7 @@ class BaseMlxChatChoicesService(lang.ExitStacked):
     )
 
     @lang.cached_function(transient=True)
-    def _get_tokenizer(self) -> mlxu.tokenization.Tokenizer:
+    def _get_tokenizer(self) -> 'mlxu.tokenization.Tokenizer':
         tokenizer = self._load_model().tokenization.tokenizer
 
         if not (

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

@@ -23,6 +23,7 @@ from ....chat.stream.types import PartialToolUseAiDelta
 from ....chat.tools.types import Tool
 from ....content.json import JsonContent
 from ....content.prepare import prepare_content_str
+from ....llms.types import MaxCompletionTokens
 from ....llms.types import MaxTokens
 from ....llms.types import Temperature
 from ....llms.types import TokenUsage
@@ -173,6 +174,7 @@ class OpenaiChatRequestHandler:
     _OPTION_KWARG_NAMES_MAP: ta.ClassVar[ta.Mapping[str, type[ChatChoicesOptions]]] = dict(
         temperature=Temperature,
         max_tokens=MaxTokens,
+        max_completion_tokens=MaxCompletionTokens,
     )
 
     class _ProcessedOptions(ta.NamedTuple):

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

@@ -30,6 +30,8 @@ _GPT_MODEL_NAMES = [
     'gpt-5-chat-latest',
     'gpt-5-mini',
     'gpt-5-nano',
+
+    'gpt-5.1',
 ]
 
 
@@ -46,7 +48,7 @@ CHAT_MODEL_NAMES = ModelNameCollection(
             for n in _GPT_MODEL_NAMES
         },
 
-        'gpt': 'gpt-5',
+        'gpt': 'gpt-5.1',
         'gpt-mini': 'gpt-5-mini',
 
         #

ommlds/minichain/backends/impls/sentencepiece/tokens.py

@@ -1,23 +1,26 @@
 import typing as ta
 
-import sentencepiece as spm
-
 from omlish import check
+from omlish import lang
 
 from .... import tokens as tks
 
 
+with lang.auto_proxy_import(globals()):
+    import sentencepiece as spm
+
+
 ##
 
 
-def build_vocab(spm_tokenizer: spm.SentencePieceProcessor) -> tks.Vocab:
+def build_vocab(spm_tokenizer: 'spm.SentencePieceProcessor') -> tks.Vocab:
     return tks.Vocab([
         (ta.cast(tks.Token, i), tks.TokenStr(spm_tokenizer.id_to_piece(i)))  # noqa
         for i in range(spm_tokenizer.get_piece_size())  # noqa
     ])
 
 
-def build_specials(spm_tokenizer: spm.SentencePieceProcessor) -> tks.SpecialTokens:
+def build_specials(spm_tokenizer: 'spm.SentencePieceProcessor') -> tks.SpecialTokens:
     # FIXME
     return tks.SpecialTokens([])
 
@@ -28,7 +31,7 @@ def build_specials(spm_tokenizer: spm.SentencePieceProcessor) -> tks.SpecialToke
 class SentencepieceTokenizer(tks.BaseTokenizer):
     def __init__(
             self,
-            spm_tokenizer: spm.SentencePieceProcessor,
+            spm_tokenizer: 'spm.SentencePieceProcessor',
     ) -> None:
         self._spm_tokenizer = check.isinstance(spm_tokenizer, spm.SentencePieceProcessor)
 
@@ -38,7 +41,7 @@ class SentencepieceTokenizer(tks.BaseTokenizer):
         )
 
     @property
-    def spm_tokenizer(self) -> spm.SentencePieceProcessor:
+    def spm_tokenizer(self) -> 'spm.SentencePieceProcessor':
         return self._spm_tokenizer
 
     #

ommlds/minichain/backends/impls/tavily.py

@@ -0,0 +1,66 @@
+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 ....backends.tavily import protocol as pt
+from ...search import SearchHit
+from ...search import SearchHits
+from ...search import SearchRequest
+from ...search import SearchResponse
+from ...search import static_check_is_search_service
+from ...standard import ApiKey
+
+
+##
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='tavily',
+#     type='SearchService',
+# )
+@static_check_is_search_service
+class TavilySearchService:
+    def __init__(
+            self,
+            *configs: ApiKey,
+            http_client: http.AsyncHttpClient | None = None,
+    ) -> None:
+        super().__init__()
+
+        self._http_client = http_client
+
+        with tv.consume(*configs) as cc:
+            self._api_key = ApiKey.pop_secret(cc, env='TAVILY_API_KEY')
+
+    async def invoke(self, request: SearchRequest) -> SearchResponse:
+        pt_request = pt.SearchRequest(
+            query=request.v,
+        )
+
+        raw_request = msh.marshal(pt_request)
+
+        http_response = await http.async_request(
+            'https://api.tavily.com/search',
+            headers={
+                http.consts.HEADER_CONTENT_TYPE: http.consts.CONTENT_TYPE_JSON,
+                http.consts.HEADER_AUTH: http.consts.format_bearer_auth_header(check.not_none(self._api_key).reveal()),
+            },
+            data=json.dumps(raw_request).encode('utf-8'),
+            client=self._http_client,
+        )
+
+        raw_response = json.loads(check.not_none(http_response.data).decode('utf-8'))
+
+        pt_response = msh.unmarshal(raw_response, pt.SearchResponse)
+
+        return SearchResponse(SearchHits(
+            l=[
+                SearchHit(
+                    title=r.title,
+                    url=r.url,
+                )
+                for r in pt_response.results or []
+            ],
+        ))

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

@@ -4,7 +4,6 @@ import typing as ta
 from omlish import check
 from omlish import lang
 
-from .....backends.tinygrad.models import llama3 as tgl3
 from ....chat.choices.services import ChatChoicesOptions
 from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
@@ -28,6 +27,10 @@ from ....stream.services import StreamResponseSink
 from ....stream.services import new_stream_response
 
 
+with lang.auto_proxy_import(globals()):
+    from .....backends.tinygrad.models import llama3 as tgl3
+
+
 ##
 
 
@@ -39,7 +42,7 @@ def _load_model(
         *,
         size: str | None = None,
         temperature: float | None = None,
-) -> tgl3.Llama3Llm:
+) -> 'tgl3.Llama3Llm':
     if size is None:
         size = DEFAULT_SIZE
     if temperature is None:
@@ -58,7 +61,7 @@ def _load_model(
 
 
 def _prepare_toks(
-        llm: tgl3.Llama3Llm,
+        llm: 'tgl3.Llama3Llm',
         chat: Chat,
         options: ta.Sequence[ChatChoicesOptions],
 ) -> list[int]:
@@ -100,7 +103,7 @@ class BaseTinygradLlama3ChatService(lang.ExitStacked, lang.Abstract):
         self._temperature = temperature
 
     @lang.cached_function(transient=True)
-    def _load_model(self) -> tgl3.Llama3Llm:
+    def _load_model(self) -> 'tgl3.Llama3Llm':
         check.not_none(self._exit_stack)
 
         return _load_model(

ommlds/minichain/backends/impls/tokenizers/tokens.py

@@ -1,23 +1,26 @@
 import typing as ta
 
-import tokenizers as tos
-
 from omlish import check
+from omlish import lang
 
 from .... import tokens as tks
 
 
+with lang.auto_proxy_import(globals()):
+    import tokenizers as tos
+
+
 ##
 
 
-def build_vocab(tos_tokenizer: tos.Tokenizer) -> tks.Vocab:
+def build_vocab(tos_tokenizer: 'tos.Tokenizer') -> tks.Vocab:
     return tks.Vocab([
         (ta.cast(tks.Token, i), tks.TokenStr(s))
         for s, i in tos_tokenizer.get_vocab().items()
     ])
 
 
-def build_specials(tos_tokenizer: tos.Tokenizer) -> tks.SpecialTokens:
+def build_specials(tos_tokenizer: 'tos.Tokenizer') -> tks.SpecialTokens:
     # FIXME
     return tks.SpecialTokens([])
 
@@ -28,7 +31,7 @@ def build_specials(tos_tokenizer: tos.Tokenizer) -> tks.SpecialTokens:
 class TokenizersTokenizer(tks.BaseTokenizer):
     def __init__(
             self,
-            tos_tokenizer: tos.Tokenizer,
+            tos_tokenizer: 'tos.Tokenizer',
     ) -> None:
         self._tos_tokenizer = check.isinstance(tos_tokenizer, tos.Tokenizer)
 
@@ -38,7 +41,7 @@ class TokenizersTokenizer(tks.BaseTokenizer):
         )
 
     @property
-    def tos_tokenizer(self) -> tos.Tokenizer:
+    def tos_tokenizer(self) -> 'tos.Tokenizer':
         return self._tos_tokenizer
 
     #

ommlds/minichain/backends/impls/transformers/sentence.py

@@ -1,7 +1,6 @@
 import typing as ta
 
-import sentence_transformers as stfm
-
+from omlish import lang
 from omlish import typedvalues as tv
 
 from ....configs import Config
@@ -13,6 +12,10 @@ from ....vectors.embeddings import static_check_is_embedding_service
 from ....vectors.types import Vector
 
 
+with lang.auto_proxy_import(globals()):
+    import sentence_transformers as stfm
+
+
 ##
 
 

ommlds/minichain/backends/impls/transformers/tokens.py

@@ -1,17 +1,20 @@
 import typing as ta
 
-import transformers as tfm
-
 from omlish import check
 from omlish import collections as col
+from omlish import lang
 
 from .... import tokens as tks
 
 
+with lang.auto_proxy_import(globals()):
+    import transformers as tfm
+
+
 ##
 
 
-def build_vocab(tfm_tokenizer: tfm.PreTrainedTokenizerBase) -> tks.Vocab:
+def build_vocab(tfm_tokenizer: 'tfm.PreTrainedTokenizerBase') -> tks.Vocab:
     return tks.Vocab([
         (ta.cast(tks.Token, i), tks.TokenStr(s))
         for s, i in tfm_tokenizer.get_vocab().items()
@@ -32,7 +35,7 @@ SPECIAL_TOKEN_ATTR_MAP: col.BiMap[type[tks.SpecialToken], str] = col.make_bi_map
 })
 
 
-def build_specials(tfm_tokenizer: tfm.PreTrainedTokenizerBase) -> tks.SpecialTokens:
+def build_specials(tfm_tokenizer: 'tfm.PreTrainedTokenizerBase') -> tks.SpecialTokens:
     return tks.SpecialTokens.from_dict({
         st: getattr(tfm_tokenizer, a)
         for st, a in SPECIAL_TOKEN_ATTR_MAP.items()
@@ -45,7 +48,7 @@ def build_specials(tfm_tokenizer: tfm.PreTrainedTokenizerBase) -> tks.SpecialTok
 class TransformersTokenizer(tks.BaseTokenizer):
     def __init__(
             self,
-            tfm_tokenizer: tfm.PreTrainedTokenizerBase,
+            tfm_tokenizer: 'tfm.PreTrainedTokenizerBase',
     ) -> None:
         self._tfm_tokenizer = check.isinstance(tfm_tokenizer, tfm.PreTrainedTokenizerBase)
 
@@ -55,7 +58,7 @@ class TransformersTokenizer(tks.BaseTokenizer):
         )
 
     @property
-    def tfm_tokenizer(self) -> tfm.PreTrainedTokenizerBase:
+    def tfm_tokenizer(self) -> 'tfm.PreTrainedTokenizerBase':
         return self._tfm_tokenizer
 
     #

ommlds/minichain/backends/impls/transformers/transformers.py

@@ -7,15 +7,11 @@ import sys
 import threading
 import typing as ta
 
-import transformers as tfm
-
 from omlish import check
 from omlish import lang
 from omlish import typedvalues as tv
 from omlish.asyncs.asyncio.sync import AsyncioBufferRelay
 
-from .....backends.transformers.filecache import file_cache_patch_context
-from .....backends.transformers.streamers import CancellableTextStreamer
 from ....chat.choices.services import ChatChoicesRequest
 from ....chat.choices.services import ChatChoicesResponse
 from ....chat.choices.services import static_check_is_chat_choices_service
@@ -44,6 +40,12 @@ from ....stream.services import new_stream_response
 from ...impls.huggingface.configs import HuggingfaceHubToken
 
 
+with lang.auto_proxy_import(globals()):
+    import transformers as tfm
+
+    from .....backends import transformers as tfm_u
+
+
 ##
 
 
@@ -168,7 +170,7 @@ class BaseTransformersChatChoicesService(lang.ExitStacked):
             self._huggingface_hub_token = HuggingfaceHubToken.pop_secret(cc, env='HUGGINGFACE_HUB_TOKEN')
 
     @lang.cached_function(transient=True)
-    def _load_pipeline(self) -> tfm.Pipeline:
+    def _load_pipeline(self) -> 'tfm.Pipeline':
         # FIXME: unload
         check.not_none(self._exit_stack)
 
@@ -181,7 +183,7 @@ class BaseTransformersChatChoicesService(lang.ExitStacked):
         for pkw_cfg in self._pipeline_kwargs:
             pkw.update(pkw_cfg.v)
 
-        with file_cache_patch_context(
+        with tfm_u.file_cache_patch_context(
                 local_first=True,
                 local_config_present_is_authoritative=True,
         ):
@@ -246,7 +248,7 @@ class TransformersChatChoicesStreamService(BaseTransformersChatChoicesService):
             if text or stream_end:
                 relay.push(text, *([None] if stream_end else []))
 
-        streamer = CancellableTextStreamer(
+        streamer = tfm_u.CancellableTextStreamer(
             check.not_none(pipeline.tokenizer),  # type: ignore[arg-type]
             streamer_callback,  # noqa
             skip_prompt=True,
@@ -255,7 +257,7 @@ class TransformersChatChoicesStreamService(BaseTransformersChatChoicesService):
 
         async with UseResources.or_new(request.options) as rs:
             thread = threading.Thread(
-                target=CancellableTextStreamer.ignoring_cancelled(pipeline),
+                target=tfm_u.CancellableTextStreamer.ignoring_cancelled(pipeline),
                 args=(
                     inputs,
                 ),

ommlds/minichain/llms/types.py

@@ -31,6 +31,10 @@ class MaxTokens(LlmOption, tv.UniqueScalarTypedValue[int]):
     pass
 
 
+class MaxCompletionTokens(LlmOption, tv.UniqueScalarTypedValue[int]):
+    pass
+
+
 ##
 
 

ommlds/minichain/search.py

@@ -19,7 +19,7 @@ class SearchHit(lang.Final):
     title: str | None
     url: str | None
     description: str | None = None
-    snippets: lang.SequenceNotStr[str] | None
+    snippets: lang.SequenceNotStr[str] | None = None
 
 
 @dc.dataclass(frozen=True, kw_only=True)

ommlds/minichain/standard.py

@@ -25,6 +25,7 @@ class Device(tv.UniqueScalarTypedValue[ta.Any], Config):
 ##
 
 
+# TODO: ApiEndpointPath, ApiEndpointUrl, ApiBaseUrl, ...
 class ApiUrl(tv.UniqueScalarTypedValue[str], Config):
     pass
 

ommlds/specs/mcp/_marshal.py

@@ -0,0 +1,23 @@
+from omlish import lang
+from omlish import marshal as msh
+
+from .protocol import ContentBlock
+
+
+##
+
+
+@lang.static_init
+def _install_standard_marshaling() -> None:
+    for root_cls, tag_field in [
+        (ContentBlock, 'type'),
+    ]:
+        msh.install_standard_factories(*msh.standard_polymorphism_factories(
+            msh.polymorphism_from_subclasses(
+                root_cls,
+                naming=msh.Naming.SNAKE,
+                strip_suffix=msh.AutoStripSuffix,
+            ),
+            msh.FieldTypeTagging(tag_field),
+            unions='partial',
+        ))