model-library 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

model_library/providers/deepseek.py

@@ -3,31 +3,20 @@ See deepseek data retention policy
 https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html
 """
 
-import io
-from typing import Any, Literal, Sequence
-
-from typing_extensions import override
+from typing import Literal
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
-    ToolDefinition,
 )
-from model_library.exceptions import ToolCallingNotSupportedError
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
-class DeepSeekModel(LLM):
-    @override
-    def get_client(self) -> None:
-        raise NotImplementedError("Not implemented")
-
+@register_provider("deepseek")
+class DeepSeekModel(DelegateOnly):
     def __init__(
         self,
         model_name: str,
@@ -36,80 +25,15 @@ class DeepSeekModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-        self.model_name: str = model_name
-        self.native: bool = False
 
         # https://api-docs.deepseek.com/
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=self.model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.DEEPSEEK_API_KEY,
-                    base_url="https://api.deepseek.com",
-                ),
-                use_completions=True,
-            )
+        self.delegate = OpenAIModel(
+            model_name=self.model_name,
+            provider=self.provider,
+            config=config,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.DEEPSEEK_API_KEY,
+                base_url="https://api.deepseek.com",
+            ),
+            use_completions=True,
         )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        # DeepSeek reasoning models don't support tools - they auto-route to chat model
-        # which loses reasoning capability
-        if tools and not self.supports_tools:
-            raise ToolCallingNotSupportedError(
-                f"DeepSeek model ({self.model_name}) does not support tools. "
-                f"Use deepseek/deepseek-chat for tool calls."
-            )
-
-        if self.delegate:
-            return await self.delegate_query(input, tools=tools, **kwargs)
-        raise NotImplementedError()

model_library/providers/fireworks.py

@@ -1,22 +1,17 @@
-import io
-from typing import Any, Literal, Sequence
+from typing import Literal
 
 from typing_extensions import override
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
     LLMConfig,
     ProviderConfig,
-    QueryResult,
     QueryResultCost,
     QueryResultMetadata,
-    ToolDefinition,
 )
+from model_library.base.delegate_only import DelegateOnly
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
@@ -24,13 +19,10 @@ class FireworksConfig(ProviderConfig):
     serverless: bool = True
 
 
-class FireworksModel(LLM):
+@register_provider("fireworks")
+class FireworksModel(DelegateOnly):
     provider_config = FireworksConfig()
 
-    @override
-    def get_client(self) -> None:
-        raise NotImplementedError("Not implemented")
-
     def __init__(
         self,
         model_name: str,
@@ -45,76 +37,18 @@ class FireworksModel(LLM):
         else:
             self.model_name = "accounts/rayan-936e28/deployedModels/" + self.model_name
 
-        # not using Fireworks SDK
-        self.native: bool = False
-
         # https://docs.fireworks.ai/tools-sdks/openai-compatibility
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=self.model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.FIREWORKS_API_KEY,
-                    base_url="https://api.fireworks.ai/inference/v1",
-                ),
-                use_completions=True,
-            )
+        self.delegate = OpenAIModel(
+            model_name=self.model_name,
+            provider=self.provider,
+            config=config,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.FIREWORKS_API_KEY,
+                base_url="https://api.fireworks.ai/inference/v1",
+            ),
+            use_completions=True,
         )
 
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        if self.delegate:
-            return await self.delegate_query(input, tools=tools, **kwargs)
-        raise NotImplementedError()
-
     @override
     async def _calculate_cost(
         self,
@@ -130,4 +64,6 @@ class FireworksModel(LLM):
         # https://docs.fireworks.ai/faq-new/billing-pricing/is-prompt-caching-billed-differently
         # prompt caching does not affect billing for serverless models
 
-        return await super()._calculate_cost(metadata, batch, bill_reasoning=True)
+        return await super()._calculate_cost(
+            metadata, batch, bill_reasoning=bill_reasoning
+        )

model_library/providers/google/google.py

@@ -2,6 +2,26 @@ import base64
 import io
 from typing import Any, Literal, Sequence, cast
 
+from typing_extensions import override
+
+from google.genai import Client
+from google.genai import errors as genai_errors
+from google.genai.types import (
+    Content,
+    File,
+    FunctionDeclaration,
+    GenerateContentConfig,
+    GenerateContentResponse,
+    GenerateContentResponseUsageMetadata,
+    HarmBlockThreshold,
+    HarmCategory,
+    Part,
+    SafetySetting,
+    ThinkingConfig,
+    Tool,
+    ToolListUnion,
+    UploadFileConfig,
+)
 from model_library import model_library_settings
 from model_library.base import (
     LLM,
@@ -30,27 +50,8 @@ from model_library.exceptions import (
     ModelNoOutputError,
 )
 from model_library.providers.google.batch import GoogleBatchMixin
+from model_library.register_models import register_provider
 from model_library.utils import normalize_tool_result
-from typing_extensions import override
-
-from google.genai import Client
-from google.genai import errors as genai_errors
-from google.genai.types import (
-    Content,
-    File,
-    FunctionDeclaration,
-    GenerateContentConfig,
-    GenerateContentResponse,
-    GenerateContentResponseUsageMetadata,
-    HarmBlockThreshold,
-    HarmCategory,
-    Part,
-    SafetySetting,
-    ThinkingConfig,
-    Tool,
-    ToolListUnion,
-    UploadFileConfig,
-)
 
 
 class GoogleConfig(ProviderConfig):
@@ -58,6 +59,7 @@ class GoogleConfig(ProviderConfig):
     use_interactions: bool = False
 
 
+@register_provider("google")
 class GoogleModel(LLM):
     provider_config = GoogleConfig()
 

model_library/providers/inception.py

@@ -1,27 +1,17 @@
-import io
-from typing import Any, Literal, Sequence
-
-from typing_extensions import override
+from typing import Literal
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
-    ToolDefinition,
 )
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
-class MercuryModel(LLM):
-    @override
-    def get_client(self) -> None:
-        raise NotImplementedError("Not implemented")
-
+@register_provider("inception")
+class MercuryModel(DelegateOnly):
     def __init__(
         self,
         model_name: str,
@@ -30,73 +20,15 @@ class MercuryModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-        self.model_name: str = model_name
-        self.native: bool = False
 
-        # https://docs.inceptionlabs.ai/get-started/models
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=self.model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.MERCURY_API_KEY,
-                    base_url="https://api.inceptionlabs.ai/v1/",
-                ),
-                use_completions=True,
-            )
+        # https://docs.inceptionlabs.ai/get-started/get-started#external-libraries-compatibility
+        self.delegate = OpenAIModel(
+            model_name=self.model_name,
+            provider=self.provider,
+            config=config,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.MERCURY_API_KEY,
+                base_url="https://api.inceptionlabs.ai/v1/",
+            ),
+            use_completions=True,
         )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        # relies on OpenAI delegate
-        if self.delegate:
-            return await self.delegate_query(input, tools=tools, **kwargs)
-        raise NotImplementedError()

model_library/providers/kimi.py

@@ -1,27 +1,17 @@
-import io
-from typing import Any, Literal, Sequence
-
-from typing_extensions import override
+from typing import Literal
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
-    ToolDefinition,
 )
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
-class KimiModel(LLM):
-    @override
-    def get_client(self) -> None:
-        raise NotImplementedError("Not implemented")
-
+@register_provider("kimi")
+class KimiModel(DelegateOnly):
     def __init__(
         self,
         model_name: str,
@@ -30,73 +20,15 @@ class KimiModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-        self.model_name: str = model_name
-        self.native: bool = False
 
-        # https://platform.moonshot.ai/docs
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=self.model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.KIMI_API_KEY,
-                    base_url="https://api.moonshot.ai/v1/",
-                ),
-                use_completions=True,
-            )
+        # https://platform.moonshot.ai/docs/guide/migrating-from-openai-to-kimi#about-api-compatibility
+        self.delegate = OpenAIModel(
+            model_name=self.model_name,
+            provider=self.provider,
+            config=config,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.KIMI_API_KEY,
+                base_url="https://api.moonshot.ai/v1/",
+            ),
+            use_completions=True,
         )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        # relies on oAI delegate
-        if self.delegate:
-            return await self.delegate_query(input, tools=tools, **kwargs)
-        raise NotImplementedError()

model_library/providers/mistral.py

@@ -31,9 +31,11 @@ from model_library.exceptions import (
     MaxOutputTokensExceededError,
 )
 from model_library.file_utils import trim_images
+from model_library.register_models import register_provider
 from model_library.utils import default_httpx_client
 
 
+@register_provider("mistralai")
 class MistralModel(LLM):
     _client: Mistral | None = None
 

model_library/providers/openai.py

@@ -50,6 +50,7 @@ from model_library.exceptions import (
     ModelNoOutputError,
 )
 from model_library.model_utils import get_reasoning_in_tag
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
@@ -226,6 +227,7 @@ class OpenAIConfig(ProviderConfig):
     deep_research: bool = False
 
 
+@register_provider("openai")
 class OpenAIModel(LLM):
     provider_config = OpenAIConfig()
 

model_library/providers/perplexity.py

@@ -1,30 +1,17 @@
-import io
-from typing import Any, Literal, Sequence
-
-from openai import AsyncOpenAI
-from typing_extensions import override
+from typing import Literal
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
-    ToolDefinition,
 )
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
-class PerplexityModel(LLM):
-    """Perplexity Sonar models via OpenAI-compatible API."""
-
-    @override
-    def get_client(self) -> None:  # pragma: no cover - delegate is used instead
-        raise NotImplementedError("Perplexity models are accessed via delegate client")
-
+@register_provider("perplexity")
+class PerplexityModel(DelegateOnly):
     def __init__(
         self,
         model_name: str,
@@ -33,69 +20,15 @@ class PerplexityModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-        self.native = False
-        client: AsyncOpenAI = create_openai_client_with_defaults(
-            api_key=model_library_settings.PERPLEXITY_API_KEY,
-            base_url="https://api.perplexity.ai",
-        )
 
+        # https://docs.perplexity.ai/guides/chat-completions-guide
         self.delegate = OpenAIModel(
-            model_name=model_name,
-            provider=provider,
+            model_name=self.model_name,
+            provider=self.provider,
             config=config,
-            custom_client=client,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.PERPLEXITY_API_KEY,
+                base_url="https://api.perplexity.ai",
+            ),
             use_completions=True,
         )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: object,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        return await self.delegate_query(
-            input,
-            tools=tools,
-            **kwargs,
-        )