Revert "Add registry functions to instantiate models by provider (#428)" (#469)

This reverts commit ff07682.

Author: svlandeg

README.md

@@ -119,8 +119,7 @@ factory = "llm"
 labels = ["COMPLIMENT", "INSULT"]
 
 [components.llm.model]
-@llm_models = "spacy.OpenAI.v1"
-name = "gpt-4"
+@llm_models = "spacy.GPT-4.v2"
 ```
 
 Now run:

pyproject.toml

@@ -27,9 +27,7 @@ filterwarnings = [
     "ignore:^.*The `construct` method is deprecated.*",
     "ignore:^.*Skipping device Apple Paravirtual device that does not support Metal 2.0.*",
     "ignore:^.*Pydantic V1 style `@validator` validators are deprecated.*",
-    "ignore:^.*was deprecated in langchain-community.*",
-    "ignore:^.*was deprecated in LangChain 0.0.1.*",
-    "ignore:^.*the load_module() method is deprecated and slated for removal in Python 3.12.*"
+    "ignore:^.*was deprecated in langchain-community.*"
 ]
 markers = [
     "external: interacts with a (potentially cost-incurring) third-party API",

requirements-dev.txt

@@ -13,8 +13,7 @@ langchain>=0.1,<0.2; python_version>="3.9"
 openai>=0.27,<=0.28.1; python_version>="3.9"
 
 # Necessary for running all local models on GPU.
-# TODO: transformers > 4.38 causes bug in model handling due to unknown factors. To be investigated.
-transformers[sentencepiece]>=4.0.0,<=4.38
+transformers[sentencepiece]>=4.0.0
 torch
 einops>=0.4
 

spacy_llm/models/hf/__init__.py

@@ -4,14 +4,12 @@
 from .llama2 import llama2_hf
 from .mistral import mistral_hf
 from .openllama import openllama_hf
-from .registry import huggingface_v1
 from .stablelm import stablelm_hf
 
 __all__ = [
     "HuggingFace",
     "dolly_hf",
     "falcon_hf",
-    "huggingface_v1",
     "llama2_hf",
     "mistral_hf",
     "openllama_hf",

spacy_llm/models/hf/mistral.py

@@ -99,7 +99,8 @@ def mistral_hf(
     name (Literal): Name of the Mistral model. Has to be one of Mistral.get_model_names().
     config_init (Optional[Dict[str, Any]]): HF config for initializing the model.
     config_run (Optional[Dict[str, Any]]): HF config for running the model.
-    RETURNS (Mistral): Mistral instance that can execute a set of prompts and return the raw responses.
+    RETURNS (Callable[[Iterable[str]], Iterable[str]]): Mistral instance that can execute a set of prompts and return
+        the raw responses.
     """
     return Mistral(
         name=name, config_init=config_init, config_run=config_run, context_length=8000

spacy_llm/models/hf/registry.py

@@ -99,7 +99,6 @@ def query_langchain(
         prompts (Iterable[Iterable[Any]]): Prompts to execute.
         RETURNS (Iterable[Iterable[Any]]): LLM responses.
         """
-        assert callable(model)
         return [
             [model.invoke(pr) for pr in prompts_for_doc] for prompts_for_doc in prompts
         ]

spacy_llm/models/langchain/model.py

@@ -7,43 +7,6 @@
 from .model import Anthropic, Endpoints
 
 
-@registry.llm_models("spacy.Anthropic.v1")
-def anthropic_v1(
-    name: str,
-    config: Dict[Any, Any] = SimpleFrozenDict(),
-    strict: bool = Anthropic.DEFAULT_STRICT,
-    max_tries: int = Anthropic.DEFAULT_MAX_TRIES,
-    interval: float = Anthropic.DEFAULT_INTERVAL,
-    max_request_time: float = Anthropic.DEFAULT_MAX_REQUEST_TIME,
-    context_length: Optional[int] = None,
-) -> Anthropic:
-    """Returns Anthropic model instance using REST to prompt API.
-    config (Dict[Any, Any]): LLM config arguments passed on to the initialization of the model instance.
-    name (str): Name of model to use.
-    strict (bool): If True, ValueError is raised if the LLM API returns a malformed response (i. e. any kind of JSON
-        or other response object that does not conform to the expectation of how a well-formed response object from
-        this API should look like). If False, the API error responses are returned by __call__(), but no error will
-        be raised.
-    max_tries (int): Max. number of tries for API request.
-    interval (float): Time interval (in seconds) for API retries in seconds. We implement a base 2 exponential backoff
-        at each retry.
-    max_request_time (float): Max. time (in seconds) to wait for request to terminate before raising an exception.
-    context_length (Optional[int]): Context length for this model. Only necessary for sharding and if no context length
-        natively provided by spacy-llm.
-    RETURNS (Anthropic): Instance of Anthropic model.
-    """
-    return Anthropic(
-        name=name,
-        endpoint=Endpoints.COMPLETIONS.value,
-        config=config,
-        strict=strict,
-        max_tries=max_tries,
-        interval=interval,
-        max_request_time=max_request_time,
-        context_length=context_length,
-    )
-
-
 @registry.llm_models("spacy.Claude-2.v2")
 def anthropic_claude_2_v2(
     config: Dict[Any, Any] = SimpleFrozenDict(),

spacy_llm/models/rest/anthropic/registry.py

@@ -7,43 +7,6 @@
 from .model import Cohere, Endpoints
 
 
-@registry.llm_models("spacy.Cohere.v1")
-def cohere_v1(
-    name: str,
-    config: Dict[Any, Any] = SimpleFrozenDict(),
-    strict: bool = Cohere.DEFAULT_STRICT,
-    max_tries: int = Cohere.DEFAULT_MAX_TRIES,
-    interval: float = Cohere.DEFAULT_INTERVAL,
-    max_request_time: float = Cohere.DEFAULT_MAX_REQUEST_TIME,
-    context_length: Optional[int] = None,
-) -> Cohere:
-    """Returns Cohere model instance using REST to prompt API.
-    config (Dict[Any, Any]): LLM config arguments passed on to the initialization of the model instance.
-    name (str): Name of model to use.
-    strict (bool): If True, ValueError is raised if the LLM API returns a malformed response (i. e. any kind of JSON
-        or other response object that does not conform to the expectation of how a well-formed response object from
-        this API should look like). If False, the API error responses are returned by __call__(), but no error will
-        be raised.
-    max_tries (int): Max. number of tries for API request.
-    interval (float): Time interval (in seconds) for API retries in seconds. We implement a base 2 exponential backoff
-        at each retry.
-    max_request_time (float): Max. time (in seconds) to wait for request to terminate before raising an exception.
-    context_length (Optional[int]): Context length for this model. Only necessary for sharding and if no context length
-        natively provided by spacy-llm.
-    RETURNS (Cohere): Instance of Cohere model.
-    """
-    return Cohere(
-        name=name,
-        endpoint=Endpoints.COMPLETION.value,
-        config=config,
-        strict=strict,
-        max_tries=max_tries,
-        interval=interval,
-        max_request_time=max_request_time,
-        context_length=context_length,
-    )
-
-
 @registry.llm_models("spacy.Command.v2")
 def cohere_command_v2(
     config: Dict[Any, Any] = SimpleFrozenDict(),
@@ -93,7 +56,7 @@ def cohere_command(
     max_request_time: float = Cohere.DEFAULT_MAX_REQUEST_TIME,
 ) -> Callable[[Iterable[Iterable[str]]], Iterable[Iterable[str]]]:
     """Returns Cohere instance for 'command' model using REST to prompt API.
-    name (Literal["command", "command-light", "command-light-nightly", "command-nightly"]): Name of model to use.
+    name (Literal["command", "command-light", "command-light-nightly", "command-nightly"]): Model  to use.
     config (Dict[Any, Any]): LLM config arguments passed on to the initialization of the model instance.
     strict (bool): If True, ValueError is raised if the LLM API returns a malformed response (i. e. any kind of JSON
         or other response object that does not conform to the expectation of how a well-formed response object from

spacy_llm/models/rest/cohere/registry.py

@@ -8,47 +8,6 @@
 
 _DEFAULT_TEMPERATURE = 0.0
 
-
-@registry.llm_models("spacy.OpenAI.v1")
-def openai_v1(
-    name: str,
-    config: Dict[Any, Any] = SimpleFrozenDict(temperature=_DEFAULT_TEMPERATURE),
-    strict: bool = OpenAI.DEFAULT_STRICT,
-    max_tries: int = OpenAI.DEFAULT_MAX_TRIES,
-    interval: float = OpenAI.DEFAULT_INTERVAL,
-    max_request_time: float = OpenAI.DEFAULT_MAX_REQUEST_TIME,
-    endpoint: Optional[str] = None,
-    context_length: Optional[int] = None,
-) -> OpenAI:
-    """Returns OpenAI model instance using REST to prompt API.
-
-    config (Dict[Any, Any]): LLM config passed on to the model's initialization.
-    name (str): Model name to use. Can be any model name supported by the OpenAI API.
-    strict (bool): If True, ValueError is raised if the LLM API returns a malformed response (i. e. any kind of JSON
-        or other response object that does not conform to the expectation of how a well-formed response object from
-        this API should look like). If False, the API error responses are returned by __call__(), but no error will
-        be raised.
-    max_tries (int): Max. number of tries for API request.
-    interval (float): Time interval (in seconds) for API retries in seconds. We implement a base 2 exponential backoff
-        at each retry.
-    max_request_time (float): Max. time (in seconds) to wait for request to terminate before raising an exception.
-    endpoint (Optional[str]): Endpoint to set. Defaults to standard endpoint.
-    context_length (Optional[int]): Context length for this model. Only necessary for sharding and if no context length
-        natively provided by spacy-llm.
-    RETURNS (OpenAI): OpenAI model instance.
-    """
-    return OpenAI(
-        name=name,
-        endpoint=endpoint or Endpoints.CHAT.value,
-        config=config,
-        strict=strict,
-        max_tries=max_tries,
-        interval=interval,
-        max_request_time=max_request_time,
-        context_length=context_length,
-    )
-
-
 """
 Parameter explanations:
     strict (bool): If True, ValueError is raised if the LLM API returns a malformed response (i. e. any kind of JSON