model-library 0.1.2__py3-none-any.whl → 0.1.3__py3-none-any.whl

model_library/providers/together.py

@@ -1,51 +1,27 @@
-import io
-from typing import Any, Literal, Sequence, cast
+from typing import Literal
 
-from together import AsyncTogether
-from together.types.chat_completions import (
-    ChatCompletionMessage,
-    ChatCompletionResponse,
-)
 from typing_extensions import override
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithBase64,
-    FileWithId,
-    FileWithUrl,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
+    ProviderConfig,
     QueryResultCost,
     QueryResultMetadata,
-    TextInput,
-    ToolDefinition,
-)
-from model_library.exceptions import (
-    BadInputError,
-    MaxOutputTokensExceededError,
-    ModelNoOutputError,
 )
-from model_library.file_utils import trim_images
-from model_library.model_utils import get_reasoning_in_tag
 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
 
 
-@register_provider("together")
-class TogetherModel(LLM):
-    _client: AsyncTogether | None = None
+class TogetherConfig(ProviderConfig):
+    serverless: bool = True
 
-    @override
-    def get_client(self) -> AsyncTogether:
-        if not TogetherModel._client:
-            TogetherModel._client = AsyncTogether(
-                api_key=model_library_settings.TOGETHER_API_KEY,
-            )
-        return TogetherModel._client
+
+@register_provider("together")
+class TogetherModel(DelegateOnly):
+    provider_config = TogetherConfig()
 
     def __init__(
         self,
@@ -55,187 +31,18 @@ class TogetherModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-
         # https://docs.together.ai/docs/openai-api-compatibility
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.TOGETHER_API_KEY,
-                    base_url="https://api.together.xyz/v1",
-                ),
-                use_completions=False,
-            )
-        )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> list[dict[str, Any] | Any]:
-        new_input: list[dict[str, Any] | Any] = []
-        content_user: list[dict[str, Any]] = []
-
-        def flush_content_user():
-            nonlocal content_user
-
-            if content_user:
-                new_input.append({"role": "user", "content": content_user})
-                content_user = []
-
-        for item in input:
-            match item:
-                case TextInput():
-                    content_user.append({"type": "text", "text": item.text})
-                case FileWithBase64() | FileWithUrl() | FileWithId():
-                    match item.type:
-                        case "image":
-                            content_user.append(await self.parse_image(item))
-                        case "file":
-                            content_user.append(await self.parse_file(item))
-                case ChatCompletionMessage():
-                    flush_content_user()
-                    new_input.append(item)
-                case _:
-                    raise BadInputError("Unsupported input type")
-
-        flush_content_user()
-
-        return new_input
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> dict[str, Any]:
-        match image:
-            case FileWithBase64():
-                return {
-                    "type": "image_url",
-                    "image_url": {
-                        "url": f"data:image/{image.mime};base64,{image.base64}"
-                    },
-                }
-            case _:
-                # docs show that we can pass in s3 location somehow
-                raise BadInputError("Unsupported image type")
-
-    @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)
-
-        # llama supports max 5 images
-        if "lama-4" in self.model_name:
-            input = trim_images(input, max_images=5)
-
-        messages: list[dict[str, Any]] = []
-
-        if "nemotron-super" in self.model_name:
-            # move system prompt to prompt
-            if "system_prompt" in kwargs:
-                first_text_item = next(
-                    (item for item in input if isinstance(item, TextInput)), None
-                )
-                if not first_text_item:
-                    raise Exception(
-                        "Given system prompt for nemotron-super model, but no text input found"
-                    )
-                system_prompt = kwargs.pop("system_prompt")
-                first_text_item.text = f"SYSTEM PROMPT: {system_prompt}\nUSER PROMPT: {first_text_item.text}"
-
-            # set system prompt to detailed thinking
-            mode = "on" if self.reasoning else "off"
-            kwargs["system_prompt"] = f"detailed thinking {mode}"
-            messages.append(
-                {
-                    "role": "system",
-                    "content": f"detailed thinking {mode}",
-                }
-            )
-
-        if "system_prompt" in kwargs:
-            messages.append({"role": "system", "content": kwargs.pop("system_prompt")})
-
-        messages.extend(await self.parse_input(input))
-
-        body: dict[str, Any] = {
-            "max_tokens": self.max_tokens,
-            "model": self.model_name,
-            "messages": messages,
-        }
-
-        if self.supports_temperature:
-            if self.temperature is not None:
-                body["temperature"] = self.temperature
-            if self.top_p is not None:
-                body["top_p"] = self.top_p
-
-        body.update(kwargs)
-
-        response = await self.get_client().chat.completions.create(**body, stream=False)  # pyright: ignore[reportAny]
-
-        response = cast(ChatCompletionResponse, response)
-
-        if not response or not response.choices or not response.choices[0].message:
-            raise ModelNoOutputError("Model returned no completions")
-
-        text = str(response.choices[0].message.content)
-        reasoning = None
-
-        if response.choices[0].finish_reason == "length" and not text:
-            raise MaxOutputTokensExceededError()
-
-        if self.reasoning:
-            text, reasoning = get_reasoning_in_tag(text)
-
-        output = QueryResult(
-            output_text=text,
-            reasoning=reasoning,
-            history=[*input, response.choices[0].message],
+        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.TOGETHER_API_KEY,
+                base_url="https://api.together.xyz/v1",
+            ),
+            use_completions=True,
         )
 
-        if response.usage:
-            output.metadata.in_tokens = response.usage.prompt_tokens
-            output.metadata.out_tokens = response.usage.completion_tokens
-            # no cache tokens it seems
-        return output
-
     @override
     async def _calculate_cost(
         self,

model_library/register_models.py

@@ -143,8 +143,6 @@ This has no effect on runtime use of ProviderConfig, only used to load the yaml
 class BaseProviderProperties(BaseModel):
     """Static base class for dynamic ProviderProperties."""
 
-    pass
-
 
 def all_subclasses(cls: type) -> list[type]:
     """Recursively find all subclasses of a class."""

{model_library-0.1.2.dist-info → model_library-0.1.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: model-library
-Version: 0.1.2
+Version: 0.1.3
 Summary: Model Library for vals.ai 
 Author-email: "Vals AI, Inc." <contact@vals.ai>
 License: MIT
@@ -17,12 +17,11 @@ Requires-Dist: tiktoken==0.11.0
 Requires-Dist: pillow
 Requires-Dist: openai<2.0,>=1.97.1
 Requires-Dist: anthropic<1.0,>=0.57.1
-Requires-Dist: together<2.0,>=1.5.25
 Requires-Dist: mistralai<2.0,>=1.9.10
 Requires-Dist: xai-sdk<2.0,>=1.0.0
 Requires-Dist: ai21<5.0,>=4.0.3
 Requires-Dist: boto3<2.0,>=1.38.27
-Requires-Dist: google-genai[aiohttp]<2.0,>=1.48.0
+Requires-Dist: google-genai[aiohttp]>=1.51.0
 Requires-Dist: google-cloud-storage>=1.26.0
 Dynamic: license-file
 

{model_library-0.1.2.dist-info → model_library-0.1.3.dist-info}/RECORD

@@ -1,43 +1,43 @@
 model_library/__init__.py,sha256=AKc_15aklOf-LbcS9z1Xer_moRWNpG6Dh3kqvSQ0nOI,714
-model_library/exceptions.py,sha256=T_CEX6STyGPMFFAz4kXZg7fv6YfvPi8UJjRSeogP1fk,8845
-model_library/file_utils.py,sha256=vLxYWI0-kwp67UONcFdZw2qDTV38P7IZLBaXFJDNtO4,3666
+model_library/exceptions.py,sha256=BWM3Gk7Lh_H4ttKtkvr4zcj01_UXb8e1PbKn8bl5Fgg,8953
+model_library/file_utils.py,sha256=FAZRRtDT8c4Rjfoj64Te3knEHggXAAfRRuS8WLCsSe8,3682
 model_library/logging.py,sha256=McyaPHUk7RkB38-LrfnudrrU1B62ta8wAbbIBwLRmj0,853
 model_library/model_utils.py,sha256=l8oCltGeimMGtnne_3Q1EguVtzCj61UMsLsma-1czwg,753
 model_library/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-model_library/register_models.py,sha256=kAttwWA4tpb8WPnJSoANpQ0sa1ERWWgGx1EQcuuCaHI,13863
+model_library/register_models.py,sha256=iVmzLDraUfzwF_bPz6DI2zMIeSUG1UQO9IdmGZTTui4,13853
 model_library/registry_utils.py,sha256=Op1NnQGD0XZyP032pGJdWdHgdEoHwKOcoWNCihT69SE,6977
 model_library/settings.py,sha256=QyeUqzWBpexFi014L_mZkoXP49no3SAQNJRObATXrL8,873
 model_library/utils.py,sha256=jQqBbP9vafpuFxp7kb53XYvCAtW79FtFJelnGGPn-pQ,4011
 model_library/base/__init__.py,sha256=TtxCXGUtkEqWZNMMofLPuC4orN7Ja2hemtbtHitt_UA,266
-model_library/base/base.py,sha256=gatrvBdSt2z2wOncgi4FtlPqvlOqmVfkRTh2eZcqgKk,13937
+model_library/base/base.py,sha256=AyUt3XeFX3l1mVPf57IfqOnINdF3p2ziv5R3wz44VkM,14064
 model_library/base/batch.py,sha256=-jd6L0ECc5pkj73zoX2ZYcv_9iQdqxEi1kEilwaXWSA,2895
 model_library/base/delegate_only.py,sha256=V2MzENtvBg0pySKncgE-mfCLBhhRZ0y4BntQwQsxbqU,2111
 model_library/base/input.py,sha256=Nhg8Ril1kFau1DnE8u102JC1l-vxNd-v9e3SjovR-Do,1876
-model_library/base/output.py,sha256=5DG077lU-CNXVVfJAELaR4gJh0_bw3V5MwbgZIQphiY,5504
-model_library/base/utils.py,sha256=6okObx8VJc7xKmPR-tBFLTWFbnZTNhTdCPbuoV-Mef8,1246
-model_library/config/ai21labs_models.yaml,sha256=su8YrHwLTVfIvRrk4AFaLi_Xg1BU_-g8AK1ExKZUXSk,2547
+model_library/base/output.py,sha256=v4s8Q_gYG2zxhhHQnDsAWFNPZHDx_kyLAhK7kxRhPjA,6673
+model_library/base/utils.py,sha256=KJZRVWr38Tik3yNJvTXnBy62ccilzzmSxHZFpQBJMPo,1330
+model_library/config/ai21labs_models.yaml,sha256=ykOwxkeWrhO98-V3sC5-FwXs0WLEeD1GZrM-VdTpF9o,2577
 model_library/config/alibaba_models.yaml,sha256=2tIdj_7Qiw_-jZmutdtdJWyAXPl4WEFh1ij3ubymov0,2252
-model_library/config/all_models.json,sha256=8XKNQbk0Zgmmkfio1kRDYs3MhFmaRQsVS30c0EENB4Y,488427
+model_library/config/all_models.json,sha256=0i0aND49z9aBG19JpZp5Kk3Kt2nuj_mHaBJ9ZexYkwQ,497215
 model_library/config/amazon_models.yaml,sha256=RGj7DH0IzXNs4JmAk6adC2jUEefVxBJNVQIB-n-fXbc,8988
-model_library/config/anthropic_models.yaml,sha256=VHfmm2mRjKa0Ieusk0N4tRT_oy5BjIoNy1BkZJpEcnU,10490
+model_library/config/anthropic_models.yaml,sha256=HszgDVQpv9EwSKwXd7T-eYIdQ3Ue07uNm8QjeRirFY0,11074
 model_library/config/cohere_models.yaml,sha256=BQSqIsGUXvULeFLGDFJNBxen-6CC5hKNW2s4lSC8Np0,5186
 model_library/config/deepseek_models.yaml,sha256=42bfyNZtaiOYx188FMqkfuCBSLNM1EBTYzf7hwzsjHw,1318
 model_library/config/dummy_model.yaml,sha256=CTnSdFYC-KTsHt5TvS0echWwUlb_H3eATZL1tVIkXkM,915
 model_library/config/fireworks_models.yaml,sha256=VG_Fo8X6qSA3VALn1SKehjj3B1HP0XO1fCYDdaCUdnM,5905
-model_library/config/google_models.yaml,sha256=QE9QjBrVlRDvuQ5tZ041VktcMJVGF7aAcYEJoBsimJQ,15661
+model_library/config/google_models.yaml,sha256=kymR44OrrTygVSqI2PH3ffbpVYRuoxsAVP9ERq2fS18,16855
 model_library/config/inception_models.yaml,sha256=g6PC0qjEC2SUPTo2Rad34Dl8dt8ZBv1svaaP2_PIrYg,660
 model_library/config/kimi_models.yaml,sha256=BySTLTc0m24oBC94VegosQgxpHglthe5dGRwF-fyduo,840
 model_library/config/mistral_models.yaml,sha256=MjKEYFYcGBsFd6iXekE_1oGa3CmEWAVPABJR94gV6SE,3839
-model_library/config/openai_models.yaml,sha256=6ZpqIoSGbR-K0JQzOq8kjnQT-I78ap7psRpj_D-iUas,25283
+model_library/config/openai_models.yaml,sha256=iSrX7WAZoNJOmVxrtEStnvRc-Md9z0CFYAIcC7GC_w8,25130
 model_library/config/perplexity_models.yaml,sha256=avTBrwnG-5Y6kle9t9vBrwcImhSzw-dgoYQuaw7K7Rs,2962
-model_library/config/together_models.yaml,sha256=9FtYtEyPFiuInfamuncg5b7PjSh-tc6k1448QulEIg4,24422
-model_library/config/xai_models.yaml,sha256=iNVTlpBXHF__KkvAQHgoyDPWrDvyIcOFHZOF8crAMVM,7798
+model_library/config/together_models.yaml,sha256=VA2cuo9hgrQW3yOz5sYfUcMoU_AgNifQ6k1MjiXO3jk,24454
+model_library/config/xai_models.yaml,sha256=xqXhXT3vpYC1iMzdnotj75oJ98nOwY6oFQY4gobAOmg,9328
 model_library/config/zai_models.yaml,sha256=lyAUPp4qOxkAAKCcbX48IKLaYYPAkp-Jn1wyCjLqmeA,1396
 model_library/providers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 model_library/providers/ai21labs.py,sha256=7PnXKl-Fv8KlE95eBv2izbFg1u7utDRQPdWXYVl_-as,5832
 model_library/providers/alibaba.py,sha256=k6LZErV_l9oTFTdKTwyw1SXD509Rl3AqFbN8umCryEE,2941
 model_library/providers/amazon.py,sha256=Qd5zAEn71WVo6c8mpPW0gE7WHLarzKzh6r4b_R4FaYk,13137
-model_library/providers/anthropic.py,sha256=iEQ5qRctp3iR_i4FZMWjKOIIifN2aGgW10j7Q8H4LhQ,22237
+model_library/providers/anthropic.py,sha256=6YI04jdDDtDjLS17jThVYlNvLbqd9THrKAtaVTYL6eg,22194
 model_library/providers/azure.py,sha256=brQNCED-zHvYjL5K5hdjFBNso6hJZg0HTHNnAgJPPG0,1408
 model_library/providers/cohere.py,sha256=lCBm1PP1l_UOa1pKFMIZM3C0wCv3QWB6UP0-jvjkFa4,1066
 model_library/providers/deepseek.py,sha256=7T4lxDiV5wmWUK7TAKwr332_T6uyXNCOiirZOCCETL0,1159
@@ -45,17 +45,17 @@ model_library/providers/fireworks.py,sha256=w-5mOF5oNzqx_0ijCoTm1lSn2ZHwhp6fURKh
 model_library/providers/inception.py,sha256=Nrky53iujIM9spAWoNRtoJg2inFiL0li6E75vT3b6V8,1107
 model_library/providers/kimi.py,sha256=zzvcKpZLsM1xPebpLeMxNKTt_FRiLN1rFWrIly7wfXA,1092
 model_library/providers/mistral.py,sha256=DHl0BYUZOrCvD4La5cyzcpQKHh4RbTbgMORWFbU_TuQ,9536
-model_library/providers/openai.py,sha256=2hEJIgv5HnfrNhs_E3xzDpjV64QATJxu6R0R8JMMAV0,33447
+model_library/providers/openai.py,sha256=ztJgx17e5PbbLVKsRA5Op0rczMsY4YNcgvRGKHRGlUQ,33572
 model_library/providers/perplexity.py,sha256=eIzzkaZ4ZMlRKFVI9bnwyo91iJkh7aEmJ-0_4OKeAWc,1083
-model_library/providers/together.py,sha256=ElE9k2H6kkMiK23yxz5-Czg6sOKDy80uyKAI1_oPM_4,8178
+model_library/providers/together.py,sha256=7Y4QLnX8c_fyXUud-W_C1gidmROQainTgODBwbvFyXQ,2033
 model_library/providers/vals.py,sha256=VLF1rsCR13a_kmtZfboDzJJ64Io_tBFe60vf-0BdYPc,9830
 model_library/providers/xai.py,sha256=oJiMICYLkybHpLv77PmMbi1Xj9IUZmKX3kANksjjFEQ,10828
 model_library/providers/zai.py,sha256=O_GM6KlJ0fM2wYoxO9xrCWfnpYH7IpoKEzjiD4jB8Kc,1050
 model_library/providers/google/__init__.py,sha256=ypuLVL_QJEQ7C3S47FhC9y4wyawYOdGikAViJmACI0U,115
 model_library/providers/google/batch.py,sha256=4TE90Uo1adi54dVtGcGyUAxw11YExJq-Y4KmkQ-cyHA,9978
-model_library/providers/google/google.py,sha256=WKuQr4C-ZVcK6ew50YEHgcPA91pEafFXSI7cNi6nRlQ,16436
-model_library-0.1.2.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
-model_library-0.1.2.dist-info/METADATA,sha256=dEDbkl76pIIbzBEZ6qtTXMUxr3sX_-AO9vC76y8OSho,7034
-model_library-0.1.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-model_library-0.1.2.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
-model_library-0.1.2.dist-info/RECORD,,
+model_library/providers/google/google.py,sha256=s9vky9r5SVNhBvMXcIr0_h0MlKLXwx_tQlZzs57xXYo,16507
+model_library-0.1.3.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
+model_library-0.1.3.dist-info/METADATA,sha256=qnQqa4o9vNXB74vfgytR3qYSNr16HBOT7_iSDeAx0zw,6992
+model_library-0.1.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+model_library-0.1.3.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
+model_library-0.1.3.dist-info/RECORD,,