Removed features that required python 3.11 (#230)

* Removed features that required python 3.11

* Added unit tests

paperqa/llms.py

@@ -3,17 +3,7 @@
 import re
 from abc import ABC, abstractmethod
 from inspect import signature
-from typing import (
-    Any,
-    AsyncGenerator,
-    Callable,
-    Coroutine,
-    Sequence,
-    Type,
-    cast,
-    get_args,
-    get_type_hints,
-)
+from typing import Any, AsyncGenerator, Callable, Coroutine, Sequence, Type, cast
 
 import numpy as np
 from openai import AsyncOpenAI
@@ -23,37 +13,60 @@
 from .types import Doc, Embeddable, LLMResult, Text
 from .utils import batch_iter, flatten, gather_with_concurrency, is_coroutine_callable
 
+# only works for python 3.11
+# def guess_model_type(model_name: str) -> str:
+#     """Guess the model type from the model name for OpenAI models"""
+#     import openai
 
-def guess_model_type(model_name: str) -> str:
-    """Guess the model type from the model name for OpenAI models"""
-    import openai
-
-    model_type = get_type_hints(
-        openai.types.chat.completion_create_params.CompletionCreateParamsBase
-    )["model"]
-    model_union = get_args(get_args(model_type)[1])
-    model_arr = list(model_union)
-    if model_name in model_arr:
-        return "chat"
-    return "completion"
+#     model_type = get_type_hints(
+#         openai.types.chat.completion_create_params.CompletionCreateParamsBase
+#     )["model"]
+#     model_union = get_args(get_args(model_type)[1])
+#     model_arr = list(model_union)
+#     if model_name in model_arr:
+#         return "chat"
+#     return "completion"
+
+# def is_openai_model(model_name):
+#     import openai
 
+#     model_type = get_type_hints(
+#         openai.types.chat.completion_create_params.CompletionCreateParamsBase
+#     )["model"]
+#     model_union = get_args(get_args(model_type)[1])
+#     model_arr = list(model_union)
 
-def is_openai_model(model_name):
-    import openai
+#     complete_model_types = get_type_hints(
+#         openai.types.completion_create_params.CompletionCreateParamsBase
+#     )["model"]
+#     complete_model_union = get_args(get_args(complete_model_types)[1])
+#     complete_model_arr = list(complete_model_union)
 
-    model_type = get_type_hints(
-        openai.types.chat.completion_create_params.CompletionCreateParamsBase
-    )["model"]
-    model_union = get_args(get_args(model_type)[1])
-    model_arr = list(model_union)
+#     return model_name in model_arr or model_name in complete_model_arr
+
+
+def guess_model_type(model_name: str) -> str:
+    if model_name.startswith("babbage"):
+        return "completion"
+    if model_name.startswith("davinci"):
+        return "completion"
+    if "instruct" in model_name:
+        return "completion"
+    if model_name.startswith("gpt-4"):
+        if "base" in model_name:
+            return "completion"
+        return "chat"
+    if model_name.startswith("gpt-3.5"):
+        return "chat"
+    return "completion"
 
-    complete_model_types = get_type_hints(
-        openai.types.completion_create_params.CompletionCreateParamsBase
-    )["model"]
-    complete_model_union = get_args(get_args(complete_model_types)[1])
-    complete_model_arr = list(complete_model_union)
 
-    return model_name in model_arr or model_name in complete_model_arr
+def is_openai_model(model_name) -> bool:
+    return (
+        model_name.startswith("gpt-")
+        or model_name.startswith("babbage")
+        or model_name.startswith("davinci")
+    )
 
 
 def process_llm_config(llm_config: dict) -> dict:

paperqa/version.py

@@ -1 +1 @@
-__version__ = "4.0.0-pre.3"
+__version__ = "4.0.0-pre.4"

setup.py

@@ -25,6 +25,7 @@
         "html2text",
         "tiktoken>=0.4.0",
     ],
+    python_requires=">=3.11",
     test_suite="tests",
     long_description=long_description,
     long_description_content_type="text/markdown",

tests/test_paperqa.py

@@ -25,6 +25,8 @@
     OpenAIEmbeddingModel,
     OpenAILLMModel,
     get_score,
+    guess_model_type,
+    is_openai_model,
 )
 from paperqa.readers import read_doc
 from paperqa.utils import (
@@ -37,6 +39,24 @@
 )
 
 
+def test_is_openai_model():
+    assert is_openai_model("gpt-3.5-turbo")
+    assert is_openai_model("babbage-002")
+    assert is_openai_model("gpt-4-1106-preview")
+    assert is_openai_model("davinci-002")
+    assert not is_openai_model("llama")
+    assert not is_openai_model("labgpt")
+    assert not is_openai_model("mixtral-7B")
+
+
+def test_guess_model_type():
+    assert guess_model_type("gpt-3.5-turbo") == "chat"
+    assert guess_model_type("babbage-002") == "completion"
+    assert guess_model_type("gpt-4-1106-preview") == "chat"
+    assert guess_model_type("gpt-3.5-turbo-instruct") == "completion"
+    assert guess_model_type("davinci-002") == "completion"
+
+
 def test_iter_citations():
     text = (
         "Yes, COVID-19 vaccines are effective. Various studies have documented the "