kiln-ai 0.8.1__py3-none-any.whl → 0.12.0__py3-none-any.whl

kiln_ai/adapters/langchain_adapters.py

@@ -1,256 +0,0 @@
-import os
-from os import getenv
-from typing import Any, Dict
-
-from langchain_aws import ChatBedrockConverse
-from langchain_core.language_models import LanguageModelInput
-from langchain_core.language_models.chat_models import BaseChatModel
-from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
-from langchain_core.messages.base import BaseMessage
-from langchain_core.runnables import Runnable
-from langchain_fireworks import ChatFireworks
-from langchain_groq import ChatGroq
-from langchain_ollama import ChatOllama
-from langchain_openai import ChatOpenAI
-from pydantic import BaseModel
-
-import kiln_ai.datamodel as datamodel
-from kiln_ai.adapters.ollama_tools import (
-    get_ollama_connection,
-    ollama_base_url,
-    ollama_model_installed,
-)
-from kiln_ai.utils.config import Config
-
-from .base_adapter import AdapterInfo, BaseAdapter, BasePromptBuilder, RunOutput
-from .ml_model_list import KilnModelProvider, ModelProviderName
-from .provider_tools import kiln_model_provider_from
-
-LangChainModelType = BaseChatModel | Runnable[LanguageModelInput, Dict | BaseModel]
-
-
-class LangchainAdapter(BaseAdapter):
-    _model: LangChainModelType | None = None
-
-    def __init__(
-        self,
-        kiln_task: datamodel.Task,
-        custom_model: BaseChatModel | None = None,
-        model_name: str | None = None,
-        provider: str | None = None,
-        prompt_builder: BasePromptBuilder | None = None,
-        tags: list[str] | None = None,
-    ):
-        super().__init__(kiln_task, prompt_builder=prompt_builder, tags=tags)
-        if custom_model is not None:
-            self._model = custom_model
-
-            # Attempt to infer model provider and name from custom model
-            self.model_provider = "custom.langchain:" + custom_model.__class__.__name__
-            self.model_name = "custom.langchain:unknown_model"
-            if hasattr(custom_model, "model_name") and isinstance(
-                getattr(custom_model, "model_name"), str
-            ):
-                self.model_name = "custom.langchain:" + getattr(
-                    custom_model, "model_name"
-                )
-            if hasattr(custom_model, "model") and isinstance(
-                getattr(custom_model, "model"), str
-            ):
-                self.model_name = "custom.langchain:" + getattr(custom_model, "model")
-        elif model_name is not None:
-            self.model_name = model_name
-            self.model_provider = provider or "custom.langchain.default_provider"
-        else:
-            raise ValueError(
-                "model_name and provider must be provided if custom_model is not provided"
-            )
-
-    async def model(self) -> LangChainModelType:
-        # cached model
-        if self._model:
-            return self._model
-
-        self._model = await langchain_model_from(self.model_name, self.model_provider)
-
-        if self.has_structured_output():
-            if not hasattr(self._model, "with_structured_output") or not callable(
-                getattr(self._model, "with_structured_output")
-            ):
-                raise ValueError(
-                    f"model {self._model} does not support structured output, cannot use output_json_schema"
-                )
-            # Langchain expects title/description to be at top level, on top of json schema
-            output_schema = self.kiln_task.output_schema()
-            if output_schema is None:
-                raise ValueError(
-                    f"output_json_schema is not valid json: {self.kiln_task.output_json_schema}"
-                )
-            output_schema["title"] = "task_response"
-            output_schema["description"] = "A response from the task"
-            with_structured_output_options = await get_structured_output_options(
-                self.model_name, self.model_provider
-            )
-            self._model = self._model.with_structured_output(
-                output_schema,
-                include_raw=True,
-                **with_structured_output_options,
-            )
-        return self._model
-
-    async def _run(self, input: Dict | str) -> RunOutput:
-        model = await self.model()
-        chain = model
-        intermediate_outputs = {}
-
-        prompt = self.build_prompt()
-        user_msg = self.prompt_builder.build_user_message(input)
-        messages = [
-            SystemMessage(content=prompt),
-            HumanMessage(content=user_msg),
-        ]
-
-        # COT with structured output
-        cot_prompt = self.prompt_builder.chain_of_thought_prompt()
-        if cot_prompt and self.has_structured_output():
-            # Base model (without structured output) used for COT message
-            base_model = await langchain_model_from(
-                self.model_name, self.model_provider
-            )
-            messages.append(
-                SystemMessage(content=cot_prompt),
-            )
-
-            cot_messages = [*messages]
-            cot_response = await base_model.ainvoke(cot_messages)
-            intermediate_outputs["chain_of_thought"] = cot_response.content
-            messages.append(AIMessage(content=cot_response.content))
-            messages.append(
-                SystemMessage(content="Considering the above, return a final result.")
-            )
-        elif cot_prompt:
-            messages.append(SystemMessage(content=cot_prompt))
-
-        response = await chain.ainvoke(messages)
-
-        if self.has_structured_output():
-            if (
-                not isinstance(response, dict)
-                or "parsed" not in response
-                or not isinstance(response["parsed"], dict)
-            ):
-                raise RuntimeError(f"structured response not returned: {response}")
-            structured_response = response["parsed"]
-            return RunOutput(
-                output=self._munge_response(structured_response),
-                intermediate_outputs=intermediate_outputs,
-            )
-        else:
-            if not isinstance(response, BaseMessage):
-                raise RuntimeError(f"response is not a BaseMessage: {response}")
-            text_content = response.content
-            if not isinstance(text_content, str):
-                raise RuntimeError(f"response is not a string: {text_content}")
-            return RunOutput(
-                output=text_content,
-                intermediate_outputs=intermediate_outputs,
-            )
-
-    def adapter_info(self) -> AdapterInfo:
-        return AdapterInfo(
-            model_name=self.model_name,
-            model_provider=self.model_provider,
-            adapter_name="kiln_langchain_adapter",
-            prompt_builder_name=self.prompt_builder.__class__.prompt_builder_name(),
-        )
-
-    def _munge_response(self, response: Dict) -> Dict:
-        # Mistral Large tool calling format is a bit different. Convert to standard format.
-        if (
-            "name" in response
-            and response["name"] == "task_response"
-            and "arguments" in response
-        ):
-            return response["arguments"]
-        return response
-
-
-async def get_structured_output_options(
-    model_name: str, model_provider: str
-) -> Dict[str, Any]:
-    finetune_provider = await kiln_model_provider_from(model_name, model_provider)
-    if finetune_provider and finetune_provider.adapter_options.get("langchain"):
-        return finetune_provider.adapter_options["langchain"].get(
-            "with_structured_output_options", {}
-        )
-    return {}
-
-
-async def langchain_model_from(
-    name: str, provider_name: str | None = None
-) -> BaseChatModel:
-    provider = await kiln_model_provider_from(name, provider_name)
-    return await langchain_model_from_provider(provider, name)
-
-
-async def langchain_model_from_provider(
-    provider: KilnModelProvider, model_name: str
-) -> BaseChatModel:
-    if provider.name == ModelProviderName.openai:
-        api_key = Config.shared().open_ai_api_key
-        return ChatOpenAI(**provider.provider_options, openai_api_key=api_key)  # type: ignore[arg-type]
-    elif provider.name == ModelProviderName.openai_compatible:
-        # See provider_tools.py for how base_url, key and other parameters are set
-        return ChatOpenAI(**provider.provider_options)  # type: ignore[arg-type]
-    elif provider.name == ModelProviderName.groq:
-        api_key = Config.shared().groq_api_key
-        if api_key is None:
-            raise ValueError(
-                "Attempted to use Groq without an API key set. "
-                "Get your API key from https://console.groq.com/keys"
-            )
-        return ChatGroq(**provider.provider_options, groq_api_key=api_key)  # type: ignore[arg-type]
-    elif provider.name == ModelProviderName.amazon_bedrock:
-        api_key = Config.shared().bedrock_access_key
-        secret_key = Config.shared().bedrock_secret_key
-        # langchain doesn't allow passing these, so ugly hack to set env vars
-        os.environ["AWS_ACCESS_KEY_ID"] = api_key
-        os.environ["AWS_SECRET_ACCESS_KEY"] = secret_key
-        return ChatBedrockConverse(
-            **provider.provider_options,
-        )
-    elif provider.name == ModelProviderName.fireworks_ai:
-        api_key = Config.shared().fireworks_api_key
-        return ChatFireworks(**provider.provider_options, api_key=api_key)
-    elif provider.name == ModelProviderName.ollama:
-        # Ollama model naming is pretty flexible. We try a few versions of the model name
-        potential_model_names = []
-        if "model" in provider.provider_options:
-            potential_model_names.append(provider.provider_options["model"])
-        if "model_aliases" in provider.provider_options:
-            potential_model_names.extend(provider.provider_options["model_aliases"])
-
-        # Get the list of models Ollama supports
-        ollama_connection = await get_ollama_connection()
-        if ollama_connection is None:
-            raise ValueError("Failed to connect to Ollama. Ensure Ollama is running.")
-
-        for model_name in potential_model_names:
-            if ollama_model_installed(ollama_connection, model_name):
-                return ChatOllama(model=model_name, base_url=ollama_base_url())
-
-        raise ValueError(f"Model {model_name} not installed on Ollama")
-    elif provider.name == ModelProviderName.openrouter:
-        api_key = Config.shared().open_router_api_key
-        base_url = getenv("OPENROUTER_BASE_URL") or "https://openrouter.ai/api/v1"
-        return ChatOpenAI(
-            **provider.provider_options,
-            openai_api_key=api_key,  # type: ignore[arg-type]
-            openai_api_base=base_url,  # type: ignore[arg-type]
-            default_headers={
-                "HTTP-Referer": "https://getkiln.ai/openrouter",
-                "X-Title": "KilnAI",
-            },
-        )
-    else:
-        raise ValueError(f"Invalid model or provider: {model_name} - {provider.name}")

kiln_ai-0.8.1.dist-info/RECORD

@@ -1,58 +0,0 @@
-kiln_ai/__init__.py,sha256=Sc4z8LRVFMwJUoc_DPVUriSXTZ6PO9MaJ80PhRbKyB8,34
-kiln_ai/adapters/__init__.py,sha256=8-YlnTh3gsaPeEArFVLIqGE7-tbssI42fub4OQBp_DA,970
-kiln_ai/adapters/adapter_registry.py,sha256=zO-0_CWF3ZGA-1420_0Uwq976o3-7WXxEY_aTeu0PzQ,688
-kiln_ai/adapters/base_adapter.py,sha256=POSdMrZFqd0IJnLpVoyc1w9CGhdNtePZyQPgdBBRUpQ,6276
-kiln_ai/adapters/langchain_adapters.py,sha256=S9VZ9JLBDEue-vh00iNv4wM1rdBQRNnF0ubeOFLAdZc,10861
-kiln_ai/adapters/ml_model_list.py,sha256=Fl8PUlecibRjcWkKFwfge4cFz7jusVMeK35ewaWw8ac,25446
-kiln_ai/adapters/ollama_tools.py,sha256=0Of6ySbJ2d4j--9laOL6QKgRUQSrqX8dJUIrz20n59s,3561
-kiln_ai/adapters/prompt_builders.py,sha256=Mdu-f1mC9hWIDwoF7Qwd9F99GDx6oNGvtEZN-SrOsNM,10325
-kiln_ai/adapters/provider_tools.py,sha256=m7X93DFbnYnw5H2HDumFJKpTKmeau-GZLv-SUmssJZ0,12381
-kiln_ai/adapters/test_langchain_adapter.py,sha256=QiVdCUJJ_uEzD0uA0jYMC3ZO4NTGJLm9iWTwvQfdFxI,12037
-kiln_ai/adapters/test_ollama_tools.py,sha256=2KwYVaj3ySV3ld-z51TCGbJEMdb3MZj2eoEicIWz3Q4,2552
-kiln_ai/adapters/test_prompt_adaptors.py,sha256=Mc0oSYgDLxfP2u3GVR_iDWaYctTQ8Ug1u6UGvWA90lM,7494
-kiln_ai/adapters/test_prompt_builders.py,sha256=sU0bSBZa9Y4Q-mmkDf3HbQ0MNSWk5o9bC9sNgtnBokk,14598
-kiln_ai/adapters/test_provider_tools.py,sha256=S1PSXd5MJnPvBe7Hq4FijptB0lbmym2E6iztncAvuUg,20752
-kiln_ai/adapters/test_saving_adapter_results.py,sha256=SYYh2xY1zmeKhFHfWAuEY4pEiLd8SitSV5ewGOTmaOI,6447
-kiln_ai/adapters/test_structured_output.py,sha256=9Mgng-HOXiZ_WcJG5cpMWhtsdJt8Rn-7qIouBWvWVoU,9324
-kiln_ai/adapters/data_gen/__init__.py,sha256=QTZWaf7kq5BorhPvexJfwDEKmjRmIbhwW9ei8LW2SIs,276
-kiln_ai/adapters/data_gen/data_gen_prompts.py,sha256=kudjHnAz7L3q0k_NLyTlaIV7M0uRFrxXNcfcnjOE2uc,5810
-kiln_ai/adapters/data_gen/data_gen_task.py,sha256=vwjC47YDrsl4GtBJpK6FWh07TGd8CalhZOX4p4YBX8w,5904
-kiln_ai/adapters/data_gen/test_data_gen_task.py,sha256=TC_n1iWgfLp87q7eNE3ZunVCuk_J25vfw-ohi2qtnp0,9668
-kiln_ai/adapters/fine_tune/__init__.py,sha256=DxdTR60chwgck1aEoVYWyfWi6Ed2ZkdJj0lar-SEAj4,257
-kiln_ai/adapters/fine_tune/base_finetune.py,sha256=-3hyWZXImJomaZeAME6mxbjifQDAn7hwlgTm8VVkxkg,5861
-kiln_ai/adapters/fine_tune/dataset_formatter.py,sha256=DzmUaCaUalTYaX2aNtnb_oucb5ZghI13RDVwtxECMUU,6340
-kiln_ai/adapters/fine_tune/finetune_registry.py,sha256=H1B-opCTlIyd9JlIFTKsY_ctxUX9ziEc49_gnmg1SZg,483
-kiln_ai/adapters/fine_tune/fireworks_finetune.py,sha256=B5o_-A0_Y_QYtgUXZWhKAjR1MeCXvZWz5scZZuK3pMg,13303
-kiln_ai/adapters/fine_tune/openai_finetune.py,sha256=WJKczDN7CA1TJnIokzZu7hbcZiOv9JIRA1scv1zDe8o,8312
-kiln_ai/adapters/fine_tune/test_base_finetune.py,sha256=YOCdQCL5Q0kpBiaU3hccafknCg0kIFRyp16lttR2Io0,9843
-kiln_ai/adapters/fine_tune/test_dataset_formatter.py,sha256=7atbHb4kFtgSmHQMNrSnNpH2ZO8drpnfwKWCsx1p8mM,11127
-kiln_ai/adapters/fine_tune/test_fireworks_tinetune.py,sha256=Y6r5BxsevFeEUHJikfFLeeG6fbPvLOxQpqIMpn-SpvU,15272
-kiln_ai/adapters/fine_tune/test_openai_finetune.py,sha256=EF-f0JbVaPiVXF0eBYbwTKdi5thA45s-XbVB0iUBI00,16629
-kiln_ai/adapters/repair/__init__.py,sha256=dOO9MEpEhjiwzDVFg3MNfA2bKMPlax9iekDatpTkX8E,217
-kiln_ai/adapters/repair/repair_task.py,sha256=L7WTFEpfaGpWXHPQf7BTNL0wiDPbeBIVqn7qNV_SeZc,3354
-kiln_ai/adapters/repair/test_repair_task.py,sha256=JBcyqyQYWniiUo4FSle9kUEsnbTsl5JN1LTRN1SRnrE,7940
-kiln_ai/datamodel/__init__.py,sha256=zhiyzeEgaE3IYunmv6KEjX43Biby6uxBjBMMEMAjE1g,28337
-kiln_ai/datamodel/basemodel.py,sha256=zWyoYgsA2tmP55jl9H18xQ0yl9vM98aTOFJTUnW5ulU,20984
-kiln_ai/datamodel/json_schema.py,sha256=l4BIq1ItLHgcSHqsqDOchegLLHY48U4yR0SP2aMb4i0,2449
-kiln_ai/datamodel/model_cache.py,sha256=d8VjPp0p5BhrGSkx9soKyxO6VWW-bcesNSJI21ySvmA,4369
-kiln_ai/datamodel/registry.py,sha256=XwGFXJFKZtOpR1Z9ven6SftggfADdZRm8TFxCEVtfUQ,957
-kiln_ai/datamodel/test_basemodel.py,sha256=r40jWaW1073ZdIhHe-GHFE8jJDD9ocauItInOsK8pWU,15234
-kiln_ai/datamodel/test_dataset_split.py,sha256=Ug-vbga-opGN_LF51Mszx5NN4wXbx3MIP1LiNzIn5Nw,7264
-kiln_ai/datamodel/test_datasource.py,sha256=GAiZz31qezVVPwFqnt8wHMu15WvtlV89jw8C1Ue6YNI,3165
-kiln_ai/datamodel/test_example_models.py,sha256=9Jhc0bvbM4hCjJGiQNgWH5rwyIsGuneAD8h4o1P3zAY,20356
-kiln_ai/datamodel/test_json_schema.py,sha256=vdLnTQxxrcmuSrf6iOmkrmpfh7JnxqIw4B4dbDAAcZ4,3199
-kiln_ai/datamodel/test_model_cache.py,sha256=9HvK2etVZJyepdlRz5ja7u1CnyzhsV4_BupJF77yBxE,7285
-kiln_ai/datamodel/test_models.py,sha256=t2Uthl559QioTyFAbQUk4BD3PqAywl3u1RSh4tHiMP0,15071
-kiln_ai/datamodel/test_nested_save.py,sha256=xciCddqvPyKyoyjC5Lx_3Kh1t4LJv1xYRAPazR3SRcs,5588
-kiln_ai/datamodel/test_output_rating.py,sha256=zvPIp2shAgCs2RQBgwYoL09fRA3krHvgAqUa91RlWR0,15125
-kiln_ai/datamodel/test_registry.py,sha256=PhS4anLi5Bf_023obuTlO5DALhtPB8WIc_bX12Yg6Po,2705
-kiln_ai/utils/__init__.py,sha256=PTD0MwBCKAMIOGsTAwsFaJOusTJJoRFTfOGqRvCaU-E,142
-kiln_ai/utils/config.py,sha256=u289b2AHuQoPup_vILTSpgsO29fxJyU8zy8BwADAtvs,6859
-kiln_ai/utils/formatting.py,sha256=VtB9oag0lOGv17dwT7OPX_3HzBfaU9GsLH-iLete0yM,97
-kiln_ai/utils/name_generator.py,sha256=v26TgpCwQbhQFcZvzgjZvURinjrOyyFhxpsI6NQrHKc,1914
-kiln_ai/utils/test_config.py,sha256=Jw3nMFeIgZUsZDRJJY2HpB-2EkR2NoZ-rDe_o9oA7ws,9174
-kiln_ai/utils/test_name_geneator.py,sha256=9-hSTBshyakqlPbFnNcggwLrL7lcPTitauBYHg9jFWI,1513
-kiln_ai-0.8.1.dist-info/METADATA,sha256=hcnd6e-iYHbE9GsI0W2deM8O6UgZOXl70VI1j8kIJrg,9050
-kiln_ai-0.8.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-kiln_ai-0.8.1.dist-info/licenses/LICENSE.txt,sha256=_NA5pnTYgRRr4qH6lE3X-TuZJ8iRcMUi5ASoGr-lEx8,1209
-kiln_ai-0.8.1.dist-info/RECORD,,