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

* Add provider-specific registry functions.

* Update model registry handles used in tests.

* Update readme and usage examples.

* Update spacy_llm/models/rest/openai/registry.py

Co-authored-by: Sofie Van Landeghem <[email protected]>

* Fix HF registry return type.

* Fix GPU test error message regexes.

* Fix tests. Bump default OAI model to GPT-4.

* Fix external tests.

* Format.

* Ignore LangChain deprecation warning. Ease sentiment tests.

* Use GPT-4 for sharding spancat test case.

* Relax EL test. Remove unnecessary warning contexts.

* Fix comparison in EL test.

* Fix GPU tests.

---------

Co-authored-by: Sofie Van Landeghem <[email protected]>

Author: rmitsch

README.md

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

pyproject.toml

@@ -27,7 +27,9 @@ 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-community.*",
+    "ignore:^.*was deprecated in LangChain 0.0.1.*",
+    "ignore:^.*the load_module() method is deprecated and slated for removal in Python 3.12.*"
 ]
 markers = [
     "external: interacts with a (potentially cost-incurring) third-party API",

requirements-dev.txt

@@ -13,7 +13,8 @@ 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.
-transformers[sentencepiece]>=4.0.0
+# TODO: transformers > 4.38 causes bug in model handling due to unknown factors. To be investigated.
+transformers[sentencepiece]>=4.0.0,<=4.38
 torch
 einops>=0.4
 

spacy_llm/models/hf/__init__.py

@@ -4,12 +4,14 @@
 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

@@ -99,8 +99,7 @@ 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 (Callable[[Iterable[str]], Iterable[str]]): Mistral instance that can execute a set of prompts and return
-        the raw responses.
+    RETURNS (Mistral): 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

@@ -0,0 +1,51 @@
+from typing import Any, Dict, Optional
+
+from confection import SimpleFrozenDict
+
+from ...registry import registry
+from .base import HuggingFace
+from .dolly import Dolly
+from .falcon import Falcon
+from .llama2 import Llama2
+from .mistral import Mistral
+from .openllama import OpenLLaMA
+from .stablelm import StableLM
+
+
+@registry.llm_models("spacy.HF.v1")
+@registry.llm_models("spacy.HuggingFace.v1")
+def huggingface_v1(
+    name: str,
+    config_init: Optional[Dict[str, Any]] = SimpleFrozenDict(),
+    config_run: Optional[Dict[str, Any]] = SimpleFrozenDict(),
+) -> HuggingFace:
+    """Returns HuggingFace model instance.
+    name (str): Name of model to use.
+    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 (Callable[[Iterable[str]], Iterable[str]]): Model instance that can execute a set of prompts and return
+        the raw responses.
+    """
+    model_context_lengths = {
+        Dolly: 2048,
+        Falcon: 2048,
+        Llama2: 4096,
+        Mistral: 8000,
+        OpenLLaMA: 2048,
+        StableLM: 4096,
+    }
+
+    for model_cls, context_length in model_context_lengths.items():
+        model_names = getattr(model_cls, "MODEL_NAMES")
+        if model_names and name in model_names.__args__:
+            return model_cls(
+                name=name,
+                config_init=config_init,
+                config_run=config_run,
+                context_length=context_length,
+            )
+
+    raise ValueError(
+        f"Name {name} could not be associated with any of the supported models. Please check "
+        f"https://spacy.io/api/large-language-models#models-hf to ensure the specified model name is correct."
+    )

spacy_llm/models/langchain/model.py

@@ -99,6 +99,7 @@ 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/rest/anthropic/registry.py

@@ -7,6 +7,43 @@
 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/cohere/registry.py

@@ -7,6 +7,43 @@
 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(),
@@ -56,7 +93,7 @@ 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"]): Model  to use.
+    name (Literal["command", "command-light", "command-light-nightly", "command-nightly"]): Name of 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/openai/registry.py

@@ -8,6 +8,47 @@
 
 _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

spacy_llm/models/rest/palm/registry.py

@@ -7,6 +7,48 @@
 from .model import Endpoints, PaLM
 
 
+@registry.llm_models("spacy.Google.v1")
+def google_v1(
+    name: str,
+    config: Dict[Any, Any] = SimpleFrozenDict(temperature=0),
+    strict: bool = PaLM.DEFAULT_STRICT,
+    max_tries: int = PaLM.DEFAULT_MAX_TRIES,
+    interval: float = PaLM.DEFAULT_INTERVAL,
+    max_request_time: float = PaLM.DEFAULT_MAX_REQUEST_TIME,
+    context_length: Optional[int] = None,
+    endpoint: Optional[str] = None,
+) -> Callable[[Iterable[Iterable[str]]], Iterable[Iterable[str]]]:
+    """Returns Google model instance using REST to prompt API.
+    name (str): Name of 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
+        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.
+    endpoint (Optional[str]): Endpoint to use. Defaults to standard endpoint.
+    RETURNS (PaLM): PaLM model instance.
+    """
+    default_endpoint = (
+        Endpoints.TEXT.value if name in {"text-bison-001"} else Endpoints.MSG.value
+    )
+    return PaLM(
+        name=name,
+        endpoint=endpoint or default_endpoint,
+        config=config,
+        strict=strict,
+        max_tries=max_tries,
+        interval=interval,
+        max_request_time=max_request_time,
+        context_length=None,
+    )
+
+
 @registry.llm_models("spacy.PaLM.v2")
 def palm_bison_v2(
     config: Dict[Any, Any] = SimpleFrozenDict(temperature=0),
@@ -18,7 +60,7 @@ def palm_bison_v2(
     context_length: Optional[int] = None,
 ) -> Callable[[Iterable[Iterable[str]]], Iterable[Iterable[str]]]:
     """Returns Google instance for PaLM Bison model using REST to prompt API.
-    name (Literal["chat-bison-001", "text-bison-001"]): Model  to use.
+    name (Literal["chat-bison-001", "text-bison-001"]): Name of 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
@@ -57,7 +99,7 @@ def palm_bison(
     endpoint: Optional[str] = None,
 ) -> PaLM:
     """Returns Google instance for PaLM Bison model using REST to prompt API.
-    name (Literal["chat-bison-001", "text-bison-001"]): Model  to use.
+    name (Literal["chat-bison-001", "text-bison-001"]): Name of 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/pipeline/llm.py

@@ -24,7 +24,7 @@
 logger.addHandler(logging.NullHandler())
 
 DEFAULT_MODEL_CONFIG = {
-    "@llm_models": "spacy.GPT-3-5.v2",
+    "@llm_models": "spacy.GPT-3-5.v3",
     "strict": True,
 }
 DEFAULT_CACHE_CONFIG = {
@@ -238,6 +238,7 @@ def _process_docs(self, docs: List[Doc]) -> List[Doc]:
                 else self._task.generate_prompts(noncached_doc_batch),
                 n_iters + 1,
             )
+
             responses_iters = tee(
                 self._model(
                     # Ensure that model receives Iterable[Iterable[Any]]. If task doesn't shard, its prompt is wrapped
@@ -251,7 +252,7 @@ def _process_docs(self, docs: List[Doc]) -> List[Doc]:
             )
 
             for prompt_data, response, doc in zip(
-                prompts_iters[1], responses_iters[0], noncached_doc_batch
+                prompts_iters[1], list(responses_iters[0]), noncached_doc_batch
             ):
                 logger.debug(
                     "Generated prompt for doc: %s\n%s",
@@ -266,7 +267,7 @@ def _process_docs(self, docs: List[Doc]) -> List[Doc]:
                         elem[1] if support_sharding else noncached_doc_batch[i]
                         for i, elem in enumerate(prompts_iters[2])
                     ),
-                    responses_iters[1],
+                    list(responses_iters[1]),
                 )
             )
 

spacy_llm/tests/models/test_cohere.py

@@ -84,7 +84,7 @@ def test_cohere_api_response_when_error():
 def test_cohere_error_unsupported_model():
     """Ensure graceful handling of error when model is not supported"""
     incorrect_model = "x-gpt-3.5-turbo"
-    with pytest.raises(ValueError, match="model not found"):
+    with pytest.raises(ValueError, match="Request to Cohere API failed"):
         Cohere(
             name=incorrect_model,
             config={},