Update dependencies (#8)

* updated outdated packages (major versions)

* Updated package version

* Removed jupyter notebook results that were saved

* Made anthropic utilities work for tests when no api key is present

.vscode/settings.json

@@ -28,4 +28,6 @@
     "--profile",
     "black"
   ],
+  "jupyter.debugJustMyCode": true,
+  "debugpy.debugJustMyCode": true,
 }

code_tests/unit_tests/test_ai_models/test_time_limited_models.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import os
 from unittest.mock import Mock
 
 import pytest
@@ -12,6 +13,9 @@
 from forecasting_tools.ai_models.basic_model_interfaces.time_limited_model import (
     TimeLimitedModel,
 )
+from forecasting_tools.ai_models.model_archetypes.anthropic_text_model import (
+    AnthropicTextToTextModel,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -47,6 +51,14 @@ def test_ai_model_does_not_time_out_when_run_time_less_than_timeout_time(
     if not issubclass(subclass, TimeLimitedModel):
         raise ValueError(TIME_LIMITED_ERROR_MESSAGE)
 
+    if (
+        issubclass(subclass, AnthropicTextToTextModel)
+        and os.getenv("ANTHROPIC_API_KEY") is None
+    ):
+        pytest.skip(
+            "Skipping test for AnthropicTextModel since API key is not set and is needed for token counting"
+        )
+
     AiModelMockManager.mock_ai_model_direct_call_with_predefined_mock_value(
         mocker, subclass
     )

forecasting_tools/ai_models/model_archetypes/anthropic_text_model.py

@@ -1,4 +1,5 @@
 import logging
+import os
 from abc import ABC
 
 from langchain_anthropic import ChatAnthropic
@@ -7,6 +8,7 @@
     _get_anthropic_claude_token_cost,
 )
 from langchain_core.messages import BaseMessage, HumanMessage, SystemMessage
+from pydantic import SecretStr
 
 from forecasting_tools.ai_models.ai_utils.response_types import (
     TextTokenCostResponse,
@@ -19,6 +21,12 @@
 
 
 class AnthropicTextToTextModel(TraditionalOnlineLlm, ABC):
+    API_KEY_MISSING = True if os.getenv("ANTHROPIC_API_KEY") is None else False
+    ANTHROPIC_API_KEY = SecretStr(
+        os.getenv("ANTHROPIC_API_KEY")  # type: ignore
+        if not API_KEY_MISSING
+        else "fake-api-key-so-tests-dont-fail-to-initialize"
+    )
 
     async def invoke(self, prompt: str) -> str:
         response: TextTokenCostResponse = (
@@ -46,6 +54,7 @@ async def _call_online_model_using_api(
             timeout=None,
             stop=None,
             base_url=None,
+            api_key=self.ANTHROPIC_API_KEY,
         )
         messages = self._turn_model_input_into_messages(prompt)
         answer_message = await anthropic_llm.ainvoke(messages)
@@ -91,16 +100,23 @@ def _get_mock_return_for_direct_call_to_model_using_cheap_input(
         probable_output = "Hello! How can I assist you today? Feel free to ask any questions or let me know if you need help with anything."
 
         model = cls()
-        prompt_tokens = model.input_to_tokens(cheap_input)
+        prompt_tokens = (
+            model.input_to_tokens(cheap_input)
+            if not cls.API_KEY_MISSING
+            else 13
+        )
         anthropic_llm = ChatAnthropic(
             model_name=model.MODEL_NAME,
             timeout=None,
             stop=None,
             base_url=None,
+            api_key=cls.ANTHROPIC_API_KEY,
+        )
+        completion_tokens = (
+            anthropic_llm.get_num_tokens(probable_output)
+            if not cls.API_KEY_MISSING
+            else 26
         )
-        completion_tokens = anthropic_llm.get_num_tokens(probable_output)
-        adjustment = 9  # Through manual experimentation, it was found that the number of tokens returned by the API is 9 off for the completion response
-        completion_tokens += adjustment
         total_cost = model.calculate_cost_from_tokens(
             prompt_tkns=prompt_tokens, completion_tkns=completion_tokens
         )
@@ -126,16 +142,10 @@ def input_to_tokens(self, prompt: str) -> int:
             timeout=None,
             stop=None,
             base_url=None,
+            api_key=self.ANTHROPIC_API_KEY,
         )
         messages = self._turn_model_input_into_messages(prompt)
         tokens = llm.get_num_tokens_from_messages(messages)
-        adjustment = 0
-        for message in messages:
-            if isinstance(message, HumanMessage):
-                adjustment += 5  # Through manual experimentation, it was found that the number of tokens returned by the API is 5 off for user messages
-            if isinstance(message, SystemMessage):
-                adjustment -= 2  # Through manual experimentation, it was found that the number of tokens returned by the API is 2 off for system messages
-        tokens += adjustment
         return tokens
 
     def calculate_cost_from_tokens(