rm hardcoded gpt-4 in tests (use gpt-4o) (#475)

* rm hardcoded gpt-4 in tests (use gpt-4o)

* appease mypy

langroid/agent/base.py

@@ -94,9 +94,16 @@ def __init__(self, config: AgentConfig = AgentConfig()):
         self._indent = ""
         self.llm = LanguageModel.create(config.llm)
         self.vecdb = VectorStore.create(config.vecdb) if config.vecdb else None
+        # token_encoding_model is used to obtain the tokenizer,
+        # so we ensure that the tokenizer corresponding to the model is used.
+        if config.parsing is not None and self.config.llm is not None:
+            config.parsing.token_encoding_model = (
+                self.config.llm.chat_model or self.config.llm.completion_model
+            )
         self.parser: Optional[Parser] = (
             Parser(config.parsing) if config.parsing else None
         )
+
         self.callbacks = SimpleNamespace(
             start_llm_stream=lambda: noop_fn,
             cancel_llm_stream=noop_fn,

pyproject.toml

@@ -34,7 +34,7 @@ halo = "^0.0.31"
 typer = "^0.9.0"
 colorlog = "^6.7.0"
 openai = "^1.14.0"
-tiktoken = "^0.5.1"
+tiktoken = "^0.7.0"
 pygithub = "^1.58.1"
 pygments = "^2.15.1"
 redis = "^5.0.1"

tests/main/test_agent.py

@@ -9,7 +9,6 @@
 class CustomAgentConfig(lr.AgentConfig):
     max_tokens: int = 10000
     llm = lm.OpenAIGPTConfig(
-        chat_model=lm.OpenAIChatModel.GPT4,
         cache_config=lr.cachedb.redis_cachedb.RedisCacheConfig(fake=False),
     )
 

tests/main/test_batch.py

@@ -13,7 +13,7 @@
 from langroid.agent.task import Task
 from langroid.agent.tool_message import ToolMessage
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.utils.configuration import Settings, set_global
 from langroid.utils.constants import DONE
@@ -24,7 +24,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     vecdb: VectorStoreConfig = None
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
 

tests/main/test_chat_agent.py

@@ -2,7 +2,7 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.task import Task
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.prompts.prompts_config import PromptsConfig
 from langroid.utils.configuration import Settings, set_global
@@ -12,7 +12,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     max_tokens: int = 200
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
     prompts: PromptsConfig = PromptsConfig(

tests/main/test_chat_agent_async.py

@@ -6,7 +6,7 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.task import Task
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.utils.configuration import Settings, set_global
 from langroid.vector_store.base import VectorStoreConfig
@@ -16,7 +16,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     vecdb: VectorStoreConfig = None
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
 

tests/main/test_doc_chat_agent.py

@@ -12,7 +12,7 @@
 from langroid.agent.task import Task
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
 from langroid.embedding_models.models import OpenAIEmbeddingsConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import DocMetaData, Document, Entity
 from langroid.parsing.parser import ParsingConfig, Splitter
 from langroid.parsing.utils import generate_random_text
@@ -160,7 +160,6 @@ class _TestDocChatAgentConfig(DocChatAgentConfig):
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         stream=True,
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
 
@@ -323,7 +322,6 @@ class _MyDocChatAgentConfig(DocChatAgentConfig):
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         stream=True,
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
 

tests/main/test_doc_chat_relevance.py

@@ -6,7 +6,7 @@
 from langroid.agent.special.doc_chat_agent import DocChatAgent, DocChatAgentConfig
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
 from langroid.embedding_models.models import OpenAIEmbeddingsConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import DocMetaData, Document
 from langroid.parsing.parser import ParsingConfig, Splitter
 from langroid.utils.configuration import Settings, set_global
@@ -106,7 +106,6 @@ class _MyDocChatAgentConfig(DocChatAgentConfig):
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         stream=True,
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
 

tests/main/test_llm.py

@@ -35,11 +35,6 @@ def test_openai_gpt(test_settings: Settings, streaming, country, capital):
         type="openai",
         max_output_tokens=100,
         min_output_tokens=10,
-        chat_model=(
-            OpenAIChatModel.GPT3_5_TURBO
-            if test_settings.gpt3_5
-            else OpenAIChatModel.GPT4  # or GPT4_TURBO
-        ),
         completion_model=OpenAICompletionModel.GPT3_5_TURBO_INSTRUCT,
         cache_config=RedisCacheConfig(fake=False),
     )
@@ -108,7 +103,6 @@ def _test_context_length_error(test_settings: Settings, mode: str, max_tokens: i
     cfg = OpenAIGPTConfig(
         stream=False,
         max_output_tokens=max_tokens,
-        chat_model=OpenAIChatModel.GPT4,  # or GPT4_TURBO,
         completion_model=OpenAICompletionModel.TEXT_DA_VINCI_003,
         cache_config=RedisCacheConfig(fake=False),
     )
@@ -168,7 +162,7 @@ def check_warning(
                 except catch_errors:
                     pass
 
-    # Default is GPT4; we should not generate the warning in this case
+    # Default is GPT4o; we should not generate the warning in this case
     lr.language_models.openai_gpt.defaultOpenAIChatModel = OpenAIChatModel.GPT4_TURBO
     llm = OpenAIGPT(config=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT3_5_TURBO))
     check_warning(llm, False)

tests/main/test_llm_async.py

@@ -29,11 +29,6 @@ async def test_openai_gpt_async(test_settings: Settings, streaming, country, cap
         type="openai",
         max_output_tokens=100,
         min_output_tokens=10,
-        chat_model=(
-            OpenAIChatModel.GPT3_5_TURBO
-            if test_settings.gpt3_5
-            else OpenAIChatModel.GPT4
-        ),
         completion_model=OpenAICompletionModel.GPT3_5_TURBO_INSTRUCT,
         cache_config=RedisCacheConfig(fake=False),
     )

tests/main/test_multi_agent.py

@@ -6,7 +6,7 @@
 from langroid.agent.task import Task
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
 from langroid.language_models.base import Role
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import DocMetaData, Document, Entity
 from langroid.parsing.parser import ParsingConfig
 from langroid.prompts.prompts_config import PromptsConfig
@@ -20,7 +20,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     vecdb: VectorStoreConfig = None
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
     parsing: ParsingConfig = ParsingConfig()

tests/main/test_multi_agent_complex.py

@@ -4,7 +4,7 @@
 from langroid.agent.task import Task
 from langroid.agent.tools.recipient_tool import RecipientTool
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.parsing.parser import ParsingConfig
 from langroid.prompts.prompts_config import PromptsConfig
@@ -17,7 +17,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     vecdb: VectorStoreConfig = None
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
     parsing: ParsingConfig = ParsingConfig()

tests/main/test_multi_agent_complex_async.py

@@ -4,7 +4,7 @@
 from langroid.agent.task import Task
 from langroid.agent.tools.recipient_tool import RecipientTool
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.parsing.parser import ParsingConfig
 from langroid.prompts.prompts_config import PromptsConfig
@@ -17,7 +17,6 @@ class _TestChatAgentConfig(ChatAgentConfig):
     vecdb: VectorStoreConfig = None
     llm: OpenAIGPTConfig = OpenAIGPTConfig(
         cache_config=RedisCacheConfig(fake=False),
-        chat_model=OpenAIChatModel.GPT4,
         use_chat_for_completion=True,
     )
     parsing: ParsingConfig = ParsingConfig()

tests/main/test_openai_assistant.py

@@ -11,9 +11,9 @@
 from langroid.agent.task import Task
 from langroid.agent.tool_message import ToolMessage
 from langroid.agent.tools.recipient_tool import RecipientTool
-from langroid.language_models import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models import OpenAIGPTConfig
 from langroid.mytypes import Entity
-from langroid.utils.configuration import Settings, set_global, settings
+from langroid.utils.configuration import Settings, set_global
 from langroid.utils.constants import NO_ANSWER
 
 
@@ -111,7 +111,7 @@ def test_openai_assistant_fn_tool(test_settings: Settings, fn_api: bool):
     set_global(test_settings)
     cfg = OpenAIAssistantConfig(
         name="NabroskyBot",
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4_TURBO),
+        llm=OpenAIGPTConfig(),
         use_functions_api=fn_api,
         use_tools=not fn_api,
         system_message="""
@@ -152,7 +152,7 @@ def test_openai_assistant_fn_2_level(test_settings: Settings, fn_api: bool):
     set_global(test_settings)
     cfg = OpenAIAssistantConfig(
         name="Main",
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4_TURBO),
+        llm=OpenAIGPTConfig(),
         use_functions_api=fn_api,
         use_tools=not fn_api,
         system_message="""
@@ -168,7 +168,7 @@ def test_openai_assistant_fn_2_level(test_settings: Settings, fn_api: bool):
 
     nabrosky_cfg = OpenAIAssistantConfig(
         name="NabroskyBot",
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4_TURBO),
+        llm=OpenAIGPTConfig(),
         use_functions_api=fn_api,
         use_tools=not fn_api,
         system_message="""
@@ -237,11 +237,8 @@ def test_openai_assistant_retrieval(test_settings: Settings):
     based on retrieval from file.
     """
     set_global(test_settings)
-    # in tests/main/conftest.py we enforce gpt-4, overriding the default gpt4-turbo,
-    # but in case of OpenAIAssistant when using retrieval, we must use gpt4-turbo
-    settings.chat_model = OpenAIChatModel.GPT4_TURBO
     cfg = OpenAIAssistantConfig(
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4_TURBO),
+        llm=OpenAIGPTConfig(),
         system_message="Answer questions based on the provided document.",
     )
     agent = OpenAIAssistant(cfg)
@@ -275,7 +272,7 @@ def test_openai_asst_code_interpreter(test_settings: Settings):
     """
     set_global(test_settings)
     cfg = OpenAIAssistantConfig(
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4_TURBO),
+        llm=OpenAIGPTConfig(),
         system_message="Answer questions by running code if needed",
     )
     agent = OpenAIAssistant(cfg)
@@ -316,7 +313,7 @@ def test_openai_assistant_multi(test_settings: Settings):
         use_cached_assistant=False,
         use_cached_thread=False,
         name="Teacher",
-        llm=OpenAIGPTConfig(chat_model=OpenAIChatModel.GPT4),
+        llm=OpenAIGPTConfig(),
     )
     agent = OpenAIAssistant(cfg)
 

tests/main/test_recipient_tool.py

@@ -30,7 +30,7 @@
 from langroid.agent.task import Task
 from langroid.agent.tool_message import ToolMessage
 from langroid.agent.tools.recipient_tool import RecipientTool
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.utils.configuration import Settings, set_global
 from langroid.utils.constants import DONE
@@ -62,9 +62,7 @@ def test_agents_with_recipient_tool(
 ):
     set_global(test_settings)
     config = ChatAgentConfig(
-        llm=OpenAIGPTConfig(
-            chat_model=OpenAIChatModel.GPT4,
-        ),
+        llm=OpenAIGPTConfig(),
         use_tools=not fn_api,
         use_functions_api=fn_api,
         vecdb=None,

tests/main/test_recipient_tool_async.py

@@ -30,7 +30,7 @@
 from langroid.agent.task import Task
 from langroid.agent.tool_message import ToolMessage
 from langroid.agent.tools.recipient_tool import RecipientTool
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.mytypes import Entity
 from langroid.utils.configuration import Settings, set_global
 from langroid.utils.constants import DONE
@@ -63,9 +63,7 @@ async def test_agents_with_recipient_tool(
 ):
     set_global(test_settings)
     config = ChatAgentConfig(
-        llm=OpenAIGPTConfig(
-            chat_model=OpenAIChatModel.GPT4,
-        ),
+        llm=OpenAIGPTConfig(),
         use_tools=not fn_api,
         use_functions_api=fn_api,
         vecdb=None,

tests/main/test_stateful_tool.py

@@ -8,7 +8,7 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.task import Task
 from langroid.agent.tool_message import ToolMessage
-from langroid.language_models.openai_gpt import OpenAIChatModel, OpenAIGPTConfig
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.utils.configuration import Settings, set_global
 
 
@@ -44,9 +44,7 @@ def test_stateful_tool(test_settings: Settings, fn_api: bool):
     number_game_agent = NumberGameAgent(
         ChatAgentConfig(
             name="Gamer",
-            llm=OpenAIGPTConfig(
-                chat_model=OpenAIChatModel.GPT4,
-            ),
+            llm=OpenAIGPTConfig(),
             vecdb=None,
             use_tools=not fn_api,
             use_functions_api=fn_api,
@@ -74,9 +72,7 @@ async def test_stateful_tool_async(test_settings: Settings, fn_api: bool):
     number_game_agent = NumberGameAgent(
         ChatAgentConfig(
             name="Gamer",
-            llm=OpenAIGPTConfig(
-                chat_model=OpenAIChatModel.GPT4,
-            ),
+            llm=OpenAIGPTConfig(),
             vecdb=None,
             use_tools=not fn_api,
             use_functions_api=fn_api,

tests/main/test_stateless_tool_messages.py

@@ -6,10 +6,7 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.tool_message import ToolMessage
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import (
-    OpenAIChatModel,
-    OpenAIGPTConfig,
-)
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.parsing.parser import ParsingConfig
 from langroid.prompts.prompts_config import PromptsConfig
 from langroid.utils.configuration import Settings, set_global
@@ -45,7 +42,6 @@ def handle(self) -> str:
     vecdb=None,
     llm=OpenAIGPTConfig(
         type="openai",
-        chat_model=OpenAIChatModel.GPT4,
         cache_config=RedisCacheConfig(fake=False),
     ),
     parsing=ParsingConfig(),

tests/main/test_structured_output.py

@@ -6,18 +6,14 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.tool_message import ToolMessage
 from langroid.cachedb.redis_cachedb import RedisCacheConfig
-from langroid.language_models.openai_gpt import (
-    OpenAIChatModel,
-    OpenAIGPTConfig,
-)
+from langroid.language_models.openai_gpt import OpenAIGPTConfig
 from langroid.utils.configuration import Settings, set_global
 
 cfg = ChatAgentConfig(
     name="test-langroid",
     vecdb=None,
     llm=OpenAIGPTConfig(
         type="openai",
-        chat_model=OpenAIChatModel.GPT4,
         cache_config=RedisCacheConfig(fake=False),
     ),
 )