Pulling in latest fhlmi and prek

.github/workflows/tests.yml

@@ -24,6 +24,8 @@ jobs:
           activate-environment: true # Activate for simple `uv sync` below
       - run: uv sync
       - uses: j178/prek-action@v1
+        with:
+          prek-version: 0.2.14 # Downpin for https://github.com/j178/prek/issues/1104
       - uses: pre-commit-ci/[email protected]
         if: always()
   lint:

.pre-commit-config.yaml

@@ -82,8 +82,9 @@ repos:
     rev: v1.18.1
     hooks:
       - id: jupytext
-        args: [--to, md, --pipe, black]
-        additional_dependencies: [black<25.11.0] # Downpin for https://github.com/mwouts/jupytext/issues/1467
+        # SEE: https://github.com/mwouts/jupytext/issues/1467
+        args: [--to, md, --pipe-fmt, ipynb, --pipe, "black {}"]
+        additional_dependencies: [black]
         files: ^docs/.*\.ipynb$
   - repo: https://github.com/jsh9/markdown-toc-creator
     rev: 0.1.3

pyproject.toml

@@ -29,7 +29,7 @@ classifiers = [
 dependencies = [
     "anyio",
     "fhaviary[llm]>=0.27",  # For partial tool concurrency
-    "fhlmi>=0.40.1",  # Pin for new llm router calling syntax
+    "fhlmi>=0.41.0",  # Pin for LiteLLMModel.get_router
     "html2text",  # TODO: evaluate moving to an opt-in dependency
     "httpx",
     "httpx-aiohttp",
@@ -64,7 +64,7 @@ dev = [
     "litellm>=1.71",  # Lower pin for aiohttp transport adoption
     "mypy>=1.8",  # Pin for mutable-override
     "paper-qa[docling,image,ldp,memory,pypdf-media,pymupdf,typing,zotero,local,qdrant,office]",
-    "prek",
+    "prek<0.2.15",  # Downpin for https://github.com/j178/prek/issues/1104
     "pydantic~=2.11",  # Pin for start of model_fields deprecation
     "pylint-pydantic",
     "pytest-asyncio",

src/paperqa/settings.py

@@ -1111,7 +1111,7 @@ def make_aviary_tool_selector(self, agent_type: str | type) -> ToolSelector | No
         ):
             return ToolSelector(
                 model_name=self.agent.agent_llm,
-                acompletion=self.get_agent_llm().router().acompletion,
+                acompletion=self.get_agent_llm().get_router().acompletion,
                 **(self.agent.agent_config or {}),
             )
         return None

src/paperqa/types.py

@@ -394,14 +394,14 @@ def add_tokens(self, result: LLMResult | Message) -> None:
                 prompt_count=result.info["usage"][0],
                 completion_count=result.info["usage"][1],
             )
+
+        prompt_count = result.prompt_count or 0
+        completion_count = result.completion_count or 0
         if result.model not in self.token_counts:
-            self.token_counts[result.model] = [
-                result.prompt_count,
-                result.completion_count,
-            ]
+            self.token_counts[result.model] = [prompt_count, completion_count]
         else:
-            self.token_counts[result.model][0] += result.prompt_count
-            self.token_counts[result.model][1] += result.completion_count
+            self.token_counts[result.model][0] += prompt_count
+            self.token_counts[result.model][1] += completion_count
 
         self.cost += result.cost
 

tests/test_paperqa.py

@@ -447,7 +447,9 @@ def accum(x) -> None:
     assert isinstance(first_id, UUID)
     assert completion.text
     assert completion.seconds_to_first_token > 0
+    assert completion.prompt_count is not None
     assert completion.prompt_count > 0
+    assert completion.completion_count is not None
     assert completion.completion_count > 0
     assert completion.model == "babbage-002"
     assert str(completion) == "".join(outputs)
@@ -461,15 +463,17 @@ def accum(x) -> None:
     assert completion.text
     assert completion.seconds_to_first_token == 0
     assert completion.seconds_to_last_token > 0
+    assert completion.prompt_count is not None
     assert completion.prompt_count > 0
+    assert completion.completion_count is not None
     assert completion.completion_count > 0
     try:
         assert completion.model == "babbage-002"
         assert completion.cost > 0
     except AssertionError:
         # Account for https://github.com/BerriAI/litellm/issues/10572
         assert any(
-            "Could not find cost for model".lower() in r.message.lower()
+            "Failed to calculate cost".lower() in r.message.lower()
             for r in caplog.records
         )
 
@@ -492,7 +496,9 @@ def accum(x) -> None:
         callbacks=[accum],
     )
     assert completion.seconds_to_first_token > 0
+    assert completion.prompt_count is not None
     assert completion.prompt_count > 0
+    assert completion.completion_count is not None
     assert completion.completion_count > 0
     assert str(completion) == "".join(outputs)
     assert isinstance(completion.text, str)