ommlds 0.0.0.dev466__py3-none-any.whl → 0.0.0.dev468__py3-none-any.whl

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/backends/impls/transformers/transformers.py

@@ -16,23 +16,42 @@ 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 ChatChoicesOutputs
 from ....chat.messages import AiMessage
 from ....chat.messages import Message
 from ....chat.messages import SystemMessage
 from ....chat.messages import ToolUseMessage
 from ....chat.messages import ToolUseResultMessage
 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  # noqa
+from ....chat.stream.types import AiChoicesDeltas  # noqa
+from ....chat.stream.types import ContentAiChoiceDelta  # noqa
 from ....completion import CompletionRequest
 from ....completion import CompletionResponse
 from ....completion import static_check_is_completion_service
 from ....configs import Config
 from ....models.configs import ModelPath
+from ....resources import UseResources
+from ....stream.services import StreamResponseSink
+from ....stream.services import new_stream_response
 from ...impls.huggingface.configs import HuggingfaceHubToken
 
 
 ##
 
 
+# @omlish-manifest $.minichain.backends.strings.manifests.BackendStringsManifest(
+#     ['ChatChoicesService', 'ChatChoicesStreamService'],
+#     'transformers',
+# )
+
+
+##
+
+
 class TransformersPipelineKwargs(Config, tv.ScalarTypedValue[ta.Mapping[str, ta.Any]]):
     pass
 
@@ -128,13 +147,10 @@ def build_chat_message(m: Message) -> ta.Mapping[str, ta.Any]:
         raise TypeError(m)
 
 
-# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
-#     name='transformers',
-#     aliases=['tfm'],
-#     type='ChatChoicesService',
-# )
-@static_check_is_chat_choices_service
-class TransformersChatChoicesService(lang.ExitStacked):
+##
+
+
+class BaseTransformersChatChoicesService(lang.ExitStacked):
     DEFAULT_MODEL: ta.ClassVar[str] = (
         'meta-llama/Llama-3.2-1B-Instruct'
     )
@@ -166,16 +182,79 @@ class TransformersChatChoicesService(lang.ExitStacked):
             **pkw,
         )
 
+
+##
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='transformers',
+#     aliases=['tfm'],
+#     type='ChatChoicesService',
+# )
+@static_check_is_chat_choices_service
+class TransformersChatChoicesService(BaseTransformersChatChoicesService):
     async def invoke(self, request: ChatChoicesRequest) -> ChatChoicesResponse:
         check.empty(request.options)
 
         pipeline = self._load_pipeline()
 
-        output = pipeline(
-            [
-                build_chat_message(m)
-                for m in request.v
-            ],
-        )
+        inputs = [
+            build_chat_message(m)
+            for m in request.v
+        ]
+
+        outputs = pipeline(inputs)
+
+        gts = check.single(outputs)['generated_text']
+        ugt, agt = gts
+        check.state(ugt['role'] == 'user')
+        check.state(agt['role'] == 'assistant')
+
+        return ChatChoicesResponse([AiChoice([AiMessage(agt['content'])])])
+
+
+##
+
+
+# @omlish-manifest $.minichain.registries.manifests.RegistryManifest(
+#     name='transformers',
+#     type='ChatChoicesStreamService',
+# )
+@static_check_is_chat_choices_stream_service
+class TransformersChatChoicesStreamService(BaseTransformersChatChoicesService):
+    async def invoke(self, request: ChatChoicesStreamRequest) -> ChatChoicesStreamResponse:
+        check.empty(request.options)
 
-        return ChatChoicesResponse([AiChoice([output])])
+        pipeline = self._load_pipeline()  # noqa
+
+        inputs = [  # noqa
+            build_chat_message(m)
+            for m in request.v
+        ]
+
+        async with UseResources.or_new(request.options) as rs:
+            async def inner(sink: StreamResponseSink[AiChoicesDeltas]) -> ta.Sequence[ChatChoicesOutputs] | None:
+                # last_role: ta.Any = None
+                #
+                # for chunk in output:
+                #     check.state(chunk['object'] == 'chat.completion.chunk')
+                #
+                #     choice = check.single(chunk['choices'])
+                #
+                #     if not (delta := choice.get('delta', {})):
+                #         continue
+                #
+                #     # FIXME: check role is assistant
+                #     if (role := delta.get('role')) != last_role:
+                #         last_role = role
+                #
+                #     # FIXME: stop reason
+                #
+                #     if (content := delta.get('content', '')):
+                #         await sink.emit(AiChoicesDeltas([AiChoiceDeltas([ContentAiChoiceDelta(content)])]))
+                #
+                # return None
+
+                raise NotImplementedError
+
+            return await new_stream_response(rs, inner)

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/minichain/standard.py

@@ -25,6 +25,13 @@ class Device(tv.UniqueScalarTypedValue[ta.Any], Config):
 ##
 
 
+class ApiUrl(tv.UniqueScalarTypedValue[str], Config):
+    pass
+
+
+##
+
+
 @dc.dataclass(frozen=True)
 class SecretConfig(Config, lang.Abstract):
     v: sec.SecretRefOrStr = dc.field() | sec.secret_field

{ommlds-0.0.0.dev466.dist-info → ommlds-0.0.0.dev468.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ommlds
-Version: 0.0.0.dev466
+Version: 0.0.0.dev468
 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.dev468
+Requires-Dist: omlish==0.0.0.dev468
 Provides-Extra: all
 Requires-Dist: llama-cpp-python~=0.3; extra == "all"
 Requires-Dist: mlx~=0.29; extra == "all"
@@ -26,8 +26,8 @@ Requires-Dist: tokenizers~=0.22; extra == "all"
 Requires-Dist: torch~=2.9; extra == "all"
 Requires-Dist: transformers~=4.57; extra == "all"
 Requires-Dist: sentence-transformers~=5.1; extra == "all"
-Requires-Dist: huggingface-hub~=0.35; extra == "all"
-Requires-Dist: datasets~=4.2; extra == "all"
+Requires-Dist: huggingface-hub~=0.36; extra == "all"
+Requires-Dist: datasets~=4.3; extra == "all"
 Requires-Dist: numpy>=1.26; extra == "all"
 Requires-Dist: pytesseract~=0.3; extra == "all"
 Requires-Dist: rapidocr-onnxruntime~=1.4; extra == "all"
@@ -47,8 +47,8 @@ Requires-Dist: torch~=2.9; extra == "backends"
 Requires-Dist: transformers~=4.57; extra == "backends"
 Requires-Dist: sentence-transformers~=5.1; extra == "backends"
 Provides-Extra: huggingface
-Requires-Dist: huggingface-hub~=0.35; extra == "huggingface"
-Requires-Dist: datasets~=4.2; extra == "huggingface"
+Requires-Dist: huggingface-hub~=0.36; extra == "huggingface"
+Requires-Dist: datasets~=4.3; extra == "huggingface"
 Provides-Extra: numpy
 Requires-Dist: numpy>=1.26; extra == "numpy"
 Provides-Extra: ocr

{ommlds-0.0.0.dev466.dist-info → ommlds-0.0.0.dev468.dist-info}/RECORD

@@ -1,5 +1,5 @@
-ommlds/.omlish-manifests.json,sha256=ZrDlaAwG8hoshkjW-up0pk0dMvDI3g5dW1M92uaf5KI,17930
-ommlds/__about__.py,sha256=uAJgr2I_m_oZPlV5P8XLFeYpBlEM-DdzeyF6O5OK_qs,1759
+ommlds/.omlish-manifests.json,sha256=u1WF90X6xpzZW21a4h5zzPyP4a3T30V08RjQz5HGABM,21555
+ommlds/__about__.py,sha256=t2rQF0yXpWFcCb2dvgzGR3I35HKGvGSn-EfhaUWVl5s,1759
 ommlds/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/huggingface.py,sha256=JfEyfKOxU3-SY_ojtXBJFNeD-NIuKjvMe3GL3e93wNA,1175
 ommlds/_hacks/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -37,6 +37,8 @@ ommlds/backends/mlx/tokenization/detokenization/base.py,sha256=Tezf8Anh-w7BxpNQs
 ommlds/backends/mlx/tokenization/detokenization/bpe.py,sha256=cIw6-r-cyXTfZdyfGRgohrElMIqeLKfMRb8R1H_56nY,3659
 ommlds/backends/mlx/tokenization/detokenization/naive.py,sha256=6L-SvphzP1z16cmVB4QC9VraF7khE8ZcvKqIwwFqN6U,1779
 ommlds/backends/mlx/tokenization/detokenization/spm.py,sha256=IYSnEm-C0z_o5TKLJE_Rj6P0nNd-prT6psVPKsERWAE,1751
+ommlds/backends/ollama/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+ommlds/backends/ollama/protocol.py,sha256=1rBZOIb080MsWMfgU4d59wDQhW5EiyBYKgnFbBnLatg,4437
 ommlds/backends/openai/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/backends/openai/protocol/__init__.py,sha256=gYPUQ94GfoIAgU5TGoaC0OVGkWuplrHg-s83ynT9f-4,1750
 ommlds/backends/openai/protocol/_common.py,sha256=r4EXmw1fBFHjU5vbWTDvlM_fsafdIVg3d3PNw4F9m-Q,313
@@ -75,6 +77,8 @@ ommlds/backends/torch/__init__.py,sha256=Id8dKbxMLlp3ux62ohu9JKoXPSrM0ZXUK0eCDTY
 ommlds/backends/torch/backends.py,sha256=Bo-ZdW1n9NswvptT8bL9CssEOKwusDuBMaXVjRS8zrA,3528
 ommlds/backends/torch/devices.py,sha256=KWkeyArPdUwVqckQTJPkN-4GQdv39cpOgCMv_XfkLkQ,776
 ommlds/backends/torch/purge.py,sha256=sp6XUxNLoVCepxIPKw3tevHn-cQqgorILvIQzixauiI,1834
+ommlds/backends/transformers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+ommlds/backends/transformers/streamers.py,sha256=Hu_9lp_kUilKjOfs7Ixqr2NoA5FuRn2eRh8JdvaBDYc,1688
 ommlds/cli/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/cli/__main__.py,sha256=1ffCb0fcUOJMzxROJmJRXQ8PSOVYv7KrcuBtT95cf0c,140
 ommlds/cli/inject.py,sha256=WhTDabJz9b1NRRHVH-UyVN5nj6UncvIeTvgkGrcE9vc,666
@@ -91,7 +95,7 @@ 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
@@ -147,7 +151,7 @@ ommlds/cli/state/storage.py,sha256=tRPmgCANRrw7A5Qr700OaH58F6S96O37I8Ivrbo7_gI,3
 ommlds/datasets/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/datasets/lib/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/datasets/lib/movies.py,sha256=LmdfoXsZU9XMM_r-sxCLv_s06BFzwWO4xUj6sc9XVcI,1961
-ommlds/minichain/__init__.py,sha256=EqNJpuMwqkkdlNmipjaoC30yAqH7c8oziszlkCcXBrQ,10982
+ommlds/minichain/__init__.py,sha256=5S2GfZW4qWF1fin2Ee8YHT4XuD_vEvtCVWTIXHOGrwo,11016
 ommlds/minichain/_marshal.py,sha256=n9PGWrHhvAmGIc7KDOYt3IF9Z6G0ncXskyICTp3Ji6k,1923
 ommlds/minichain/_typedvalues.py,sha256=Vl1Edt5khC0e5RPFBPmPCxn0IzrfVd0NHzAjAN2E6Kc,2183
 ommlds/minichain/completion.py,sha256=lQ0LfCIYZsvDqteHhhDIv16D2_gn_xMfEL0ouywE5Yo,1033
@@ -157,7 +161,7 @@ ommlds/minichain/json.py,sha256=0_5rV5Zi2qPOvXi2CLAc5DF7FN3jK3ABbjoKdjtTuVo,360
 ommlds/minichain/metadata.py,sha256=2jik8gEm_VMnknPuPwqRssTg0MClRFUrXz_IsyEgUt4,878
 ommlds/minichain/resources.py,sha256=HfcydnyFmXVRspYw-32-lvM_OfrZQdPEebAt3ivLev0,4436
 ommlds/minichain/search.py,sha256=azRzWcYhcm9IgSHquqLwtbwowtYCRAtPLSm7Gvt9iNo,1262
-ommlds/minichain/standard.py,sha256=uKXvdUNLxdUu7suCBsVOjJtnYVC2hjD_tmz3Ra7H6Jg,2510
+ommlds/minichain/standard.py,sha256=cGXaGtC5iM9Q2lCcbhLtvEcPGKhcJUIh3UWyNgOssRM,2580
 ommlds/minichain/types.py,sha256=K6RRjpUi17UEG0cqPrrvbVANU0iRVh3WLiH-y6oEWFI,414
 ommlds/minichain/utils.py,sha256=NTsBu_pSZnLdZc1R1Se70rb_9J-IoB6VRwjhwzh3PwY,490
 ommlds/minichain/backends/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -185,20 +189,21 @@ ommlds/minichain/backends/impls/huggingface/__init__.py,sha256=47DEQpj8HBSa-_TIm
 ommlds/minichain/backends/impls/huggingface/configs.py,sha256=6jsBtPNXOP57PcpxNTVLGWLc-18Iwn_lDbGouwCJTIQ,258
 ommlds/minichain/backends/impls/huggingface/repos.py,sha256=8BDxJmra9elSQL2vzp2nr2p4Hpq56A3zTk7hTTnfJU4,861
 ommlds/minichain/backends/impls/llamacpp/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ommlds/minichain/backends/impls/llamacpp/chat.py,sha256=YeBzlA_3gcuF6KF0HIE7abUp28_o1Kil-SujyQNAHyE,5508
+ommlds/minichain/backends/impls/llamacpp/chat.py,sha256=J6Jslx9atAtWvLdrVtvRboQUBzRX7Z5aHlo0dK5X78A,5649
 ommlds/minichain/backends/impls/llamacpp/completion.py,sha256=oJ2I6wUoIPXYLm9Vc7dwOPgqbevatTjNBZ-jXeM24tQ,2372
 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/llamacpp/stream.py,sha256=uzrXr2HhshgFe3Z0g8KTPc6Dr2kPsyxZabIy2d6IOBg,3547
 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/ollama/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+ommlds/minichain/backends/impls/ollama/chat.py,sha256=UK19riOph-ptIz9zW7PucGWvVEtWHOHvwp7hoKurDNw,6393
 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
@@ -208,7 +213,7 @@ ommlds/minichain/backends/impls/tokenizers/tokens.py,sha256=_8Q49k5YroG5wQI0cuK6
 ommlds/minichain/backends/impls/transformers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/impls/transformers/sentence.py,sha256=1bFJ-ND3MOkj7mNsPuISrQCpqTs7npmmNmYcc2go-Fk,1393
 ommlds/minichain/backends/impls/transformers/tokens.py,sha256=uS3-IWOJRUMBfPDVRrp3SCaXdE1yzEdKHQcyv0JZQIw,2089
-ommlds/minichain/backends/impls/transformers/transformers.py,sha256=Bb1RnvDlo8bzu24ByhDacDC0sN7R7KYZnPZ9hjbViBg,5287
+ommlds/minichain/backends/impls/transformers/transformers.py,sha256=U4O-MiVH3dRXf-UNSoKZueZVM8XvAm2mMr30qQUHhFY,8000
 ommlds/minichain/backends/strings/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 ommlds/minichain/backends/strings/manifests.py,sha256=kmlanVUAZqIh0P95Mm8H20e8ib3gEgYHHUlkCXDQGFk,413
 ommlds/minichain/backends/strings/parsing.py,sha256=2wChk9Z8fhqJTk8_91f8QFjKcSZygOQM_rVk-P4NnKw,1772
@@ -231,7 +236,7 @@ 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 +373,9 @@ 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.dev468.dist-info/licenses/LICENSE,sha256=B_hVtavaA8zCYDW99DYdcpDLKz1n3BBRjZrcbv8uG8c,1451
+ommlds-0.0.0.dev468.dist-info/METADATA,sha256=k1H1yGwCqmZETx1eTEKBsNZ_whn2QWxWap9pdErGIkw,3224
+ommlds-0.0.0.dev468.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+ommlds-0.0.0.dev468.dist-info/entry_points.txt,sha256=Z5YWtX7ClfiCKdW-dd_CSVvM0h4yQpJPi-2G3q6gNFo,35
+ommlds-0.0.0.dev468.dist-info/top_level.txt,sha256=Rbnk5d5wi58vnAXx13WFZqdQ4VX8hBCS2hEL3WeXOhY,7
+ommlds-0.0.0.dev468.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,
-        )