Release 0.55.7

.github/workflows/ci.yml

@@ -36,35 +36,24 @@ jobs:
       - name: Apt install
         run: cat Aptfile | sudo xargs apt-get install
 
-      - name: Install poetry
-        uses: snok/install-poetry@76e04a911780d5b312d89783f7b1cd627778900a # v1
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
         with:
-          version: 2.1.3
-          virtualenvs-create: true
-          virtualenvs-in-project: true
+          enable-cache: true
+
+      - name: Install Python
+        run: uv python install 3.12
 
-      - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
-        with:
-          python-version-file: "pyproject.toml"
-          cache: "poetry"
-      - name: Install poetry with pip
-        run: python -m pip install poetry
-      - name: Validate lockfile
-        run: poetry check --lock
-      - name: Set Poetry Python
-        run: poetry env use python3.12
       - name: Install dependencies
-        run: |
-          source $(poetry env info --path)/bin/activate
-          poetry install --no-interaction
+        run: uv sync --locked
 
       - name: Create test local state
         run: ./scripts/test/stub-data.sh
 
       - name: Tests
         run: |
-          poetry run ./manage.py collectstatic --noinput --clear
-          poetry run ./manage.py check --fail-level WARNING
+          uv run ./manage.py collectstatic --noinput --clear
+          uv run ./manage.py check --fail-level WARNING
           export MEDIA_ROOT="$(mktemp -d)"
           ./scripts/test/python_tests.sh
         env:

AGENTS.md

@@ -44,17 +44,17 @@ Run inside Docker containers with `docker compose`:
 
 ```bash
 # Run all tests (parallel)
-docker compose run --rm web poetry run pytest -n logical
+docker compose run --rm web uv run pytest -n logical
 
 # Run specific test file
-docker compose run --rm web poetry run pytest learning_resources/models_test.py
+docker compose run --rm web uv run pytest learning_resources/models_test.py
 
 # Run specific test
-docker compose run --rm web poetry run pytest learning_resources/models_test.py::test_name -v
+docker compose run --rm web uv run pytest learning_resources/models_test.py::test_name -v
 
 # Lint and format with ruff
-docker compose run --rm web poetry run ruff format .
-docker compose run --rm web poetry run ruff check . --fix
+docker compose run --rm web uv run ruff format .
+docker compose run --rm web uv run ruff check . --fix
 
 # Run Django management commands
 docker compose run --rm web python manage.py <command>

Dockerfile

@@ -25,25 +25,17 @@ RUN mkdir /src && \
     adduser --disabled-password --gecos "" mitodl && \
     mkdir /var/media && chown -R mitodl:mitodl /var/media
 
-FROM system AS poetry
+FROM system AS uv
 
-## Set some poetry and python config
+## Set some python config
 ENV  \
   PYTHONUNBUFFERED=1 \
   PYTHONDONTWRITEBYTECODE=1 \
-  PIP_DISABLE_PIP_VERSION_CHECK=on \
-  POETRY_NO_INTERACTION=1 \
-  POETRY_VERSION=2.1.3 \
-  POETRY_VIRTUALENVS_CREATE=true \
-  POETRY_CACHE_DIR='/tmp/cache/poetry' \
-  POETRY_HOME='/home/mitodl/.local' \
-  VIRTUAL_ENV='/opt/venv'
-ENV PATH="$VIRTUAL_ENV/bin:$POETRY_HOME/bin:$PATH"
-
-# Install poetry
-RUN pip install --no-cache-dir "poetry==$POETRY_VERSION"
-
+  UV_PROJECT_ENVIRONMENT='/opt/venv'
+ENV PATH="/opt/venv/bin:$PATH"
 
+# Install uv
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /usr/local/bin/
 
 # Install Chromium (commented out lines illustrate the syntax for getting specific chromium versions)
 RUN echo "deb http://deb.debian.org/debian/ sid main" >> /etc/apt/sources.list \
@@ -61,22 +53,14 @@ RUN apt-get update -qqy \
   && apt-get -qqy install chromium-driver \
   && rm -rf /var/lib/apt/lists/* /var/cache/apt/*
 COPY pyproject.toml /src
-COPY poetry.lock /src
-RUN chown -R mitodl:mitodl /src && \
-    mkdir ${VIRTUAL_ENV} && chown -R mitodl:mitodl ${VIRTUAL_ENV}
+COPY uv.lock /src
+RUN mkdir -p /opt/venv && chown -R mitodl:mitodl /src /opt/venv
 
-## Install poetry itself, and pre-create a venv with predictable name
 USER mitodl
-RUN curl -sSL https://install.python-poetry.org \
-  | \
-  POETRY_VERSION=${POETRY_VERSION} \
-  POETRY_HOME=${POETRY_HOME} \
-  python3 -q
 WORKDIR /src
-RUN python3 -m venv $VIRTUAL_ENV
-RUN poetry install && rm -rf /tmp/cache
+RUN uv sync --frozen --no-install-project
 
-FROM poetry AS code
+FROM uv AS code
 
 # Add project
 USER root

Dockerfile-nb

@@ -4,7 +4,7 @@ USER root
 
 WORKDIR /tmp
 
-RUN pip install --force-reinstall jupyter
+RUN uv pip install --force-reinstall jupyter
 
 USER mitodl
 WORKDIR /src

RELEASE.rst

@@ -1,6 +1,13 @@
 Release Notes
 =============
 
+Version 0.55.7
+--------------
+
+- fix: Add back langchain-litellm dependency
+- chore: migrate from poetry/pip to uv for dependency management (#2975)
+- Fix contextwindow exceeded error when summarizing large transcripts (#2960)
+
 Version 0.55.6 (Released February 26, 2026)
 --------------
 

fixtures/common.py

@@ -42,6 +42,8 @@ def warnings_as_errors():
         warnings.filterwarnings("ignore", category=InsecureRequestWarning)
         warnings.filterwarnings("ignore", category=PytestMockWarning)
         warnings.filterwarnings("ignore", category=ResourceWarning)
+        # PyJWT 2.11+ warns when HMAC key is < 32 bytes (tests use short keys)
+        warnings.filterwarnings("ignore", module="jwt.*", category=UserWarning)
         # Ignore deprecation warnings in third party libraries
         warnings.filterwarnings(
             "ignore",

learning_resources/conftest.py

@@ -158,8 +158,12 @@ def setup_s3_ocw(settings):
 
 
 @pytest.fixture
-def summarizer_configuration():
+def summarizer_configuration(mocker):
     """Create a summarizer configuration"""
+    mocker.patch(
+        "learning_resources.content_summarizer.truncate_to_tokens", autospec=True
+    )
+    mocker.patch("learning_resources.content_summarizer.get_max_tokens", autospec=True)
     return ContentSummarizerConfigurationFactory.create()
 
 

learning_resources/content_summarizer.py

@@ -1,10 +1,12 @@
 import logging
 from typing import Annotated
 
+import litellm
 from django.conf import settings
 from django.db import transaction
 from django.db.models import Q
-from langchain_community.chat_models import ChatLiteLLM
+from langchain_litellm import ChatLiteLLM
+from litellm import get_max_tokens
 from typing_extensions import TypedDict
 
 from learning_resources.exceptions import (
@@ -15,9 +17,13 @@
     ContentFile,
     ContentSummarizerConfiguration,
 )
+from learning_resources.utils import truncate_to_tokens
 
 logger = logging.getLogger(__name__)
 
+# drop unsupported model params
+litellm.drop_params = True
+
 
 class Flashcard(TypedDict):
     """Flashcard structure model"""
@@ -232,10 +238,17 @@ def _generate_summary(self, content: str, llm_model: str) -> str:
             - str: Generated summary
         """
         try:
-            llm = self._get_llm(model=llm_model, temperature=0.3, max_tokens=1000)
-            response = llm.invoke(
-                f"Summarize the key points from this video. Transcript:{content}"
+            max_output_tokens = 1000
+            max_input_tokens = get_max_tokens(llm_model)
+            summarizer_message = truncate_to_tokens(
+                f"Summarize the key points from this video. Transcript:{content}",
+                max_input_tokens - max_output_tokens,
+                llm_model,
+            )
+            llm = self._get_llm(
+                model=llm_model, temperature=0.3, max_tokens=max_output_tokens
             )
+            response = llm.invoke(summarizer_message)
             logger.debug("Generating Summary using model: %s", llm)
             generated_summary = response.content
             logger.debug("Generated summary: %s", generated_summary)
@@ -263,13 +276,23 @@ def _generate_flashcards(
             - list[dict[str, str]]: List of flashcards
         """
         try:
-            llm = self._get_llm(model=llm_model, temperature=0.3, max_tokens=2048)
+            max_output_tokens = 2048
+            max_input_tokens = get_max_tokens(llm_model)
+            llm = self._get_llm(
+                model=llm_model, temperature=1, max_tokens=max_output_tokens
+            )
             logger.debug("Generating flashcards using model: %s", llm)
             structured_llm = llm.with_structured_output(FlashcardsResponse)
 
-            response = structured_llm.invoke(
-                settings.CONTENT_SUMMARIZER_FLASHCARD_PROMPT.format(content=content)
+            flashcard_prompt = settings.CONTENT_SUMMARIZER_FLASHCARD_PROMPT.format(
+                content=content
+            )
+            flashcard_prompt = truncate_to_tokens(
+                flashcard_prompt,
+                max_input_tokens - max_output_tokens,
+                llm_model,
             )
+            response = structured_llm.invoke(flashcard_prompt)
             if response:
                 generated_flashcards = response.get("flashcards", [])
                 logger.debug("Generated flashcards: %s", generated_flashcards)

learning_resources/utils.py

@@ -9,6 +9,7 @@
 import html2text
 import rapidjson
 import requests
+import tiktoken
 import yaml
 from botocore.exceptions import ClientError
 from django.conf import settings
@@ -671,3 +672,14 @@ def json_to_markdown(obj, indent=0):
     else:
         markdown += f"{indent_str}{obj}\n\n"
     return markdown
+
+
+def truncate_to_tokens(text: str, max_tokens: int, model: str = "gpt-4o") -> str:
+    """
+    Truncate text to a maximum number of tokens for a given model.
+    """
+    encoding = tiktoken.encoding_for_model(model)
+    tokens = encoding.encode(text)
+    if len(tokens) <= max_tokens:
+        return text
+    return encoding.decode(tokens[:max_tokens])

learning_resources/utils_test.py

@@ -3,11 +3,13 @@
 """
 
 import json
+import random
 from pathlib import Path
 
 import markdown
 import pytest
 import yaml
+from faker import Faker
 
 from data_fixtures import utils as data_utils
 from learning_resources import utils
@@ -37,6 +39,7 @@
 from learning_resources.utils import (
     add_parent_topics_to_learning_resource,
     transfer_list_resources,
+    truncate_to_tokens,
 )
 
 pytestmark = pytest.mark.django_db
@@ -619,3 +622,21 @@ def test_json_to_markdown(courses_data):
     # strip double newlines for compact fixture
     content = rendered_markdown.replace("\n\n", "\n")
     assert markdown.markdown(expected_markdown_response) == markdown.markdown(content)
+
+
+def test_truncate_to_tokens_util(mocker):
+    class MockEncoding:
+        def encode(self, text):
+            return list(text)
+
+        def decode(self, tokens):
+            return "".join(tokens)
+
+    mock_encoding = MockEncoding()
+    mocker.patch("tiktoken.encoding_for_model", return_value=mock_encoding)
+
+    model = "gpt-4o"
+    content = Faker().text() * 100
+    max_tokens = random.randint(1, 50)  # noqa: S311
+    truncated = truncate_to_tokens(content, max_tokens, model)
+    assert len(mock_encoding.encode(truncated)) <= max_tokens

main/settings.py

@@ -34,7 +34,7 @@
 from main.settings_pluggy import *  # noqa: F403
 from openapi.settings_spectacular import open_spectacular_settings
 
-VERSION = "0.55.6"
+VERSION = "0.55.7"
 
 log = logging.getLogger()