We don't publish
your competitive advantage.

IFValueError: Unsupported max_frags_z when running Gemma-2 with FlashInfer on GPUs with small shared memory (e.g., RTX A6000 sm86).

THENUpgrade FlashInfer to v0.1.1 or later, which fixes the issue by adjusting the kernel to accommodate smaller shared memory sizes. Ensure your FlashInfer version is compatible with your vLLM and PyTorch versions.

IFImporting ChatVertexAI from langchain_google_vertexai fails with NameError: name 'SafetySetting' is not defined due to a forward reference issue in pydantic 2.10+.

THENDowngrade pydantic to version 2.9.0 using 'pip install pydantic==2.9.0' until the integration fixes the forward reference. Alternatively, monitor the fix in the linked GitHub issue (langchain-ai/langchain-google#610) and update the langchain-google-vertexai package when a patched version is released.

IFNameError: name 'init_empty_weights' is not defined when loading a model with transformers.

THENDowngrade transformers to version 4.50.3 or upgrade to a version that includes the fix from PR #37324. This resolves a regression introduced in PR #37306 that caused the missing import.

IFImportError for ESMForMaskedLM from transformers when using ESM protein models.

THENUpgrade the transformers library to version 4.12.0 or later, which includes the ESM model classes. Alternatively, install transformers with the ESM extras via `pip install transformers[esm]`. Run `!pip install --upgrade transformers` to ensure the latest release is used.

IFRunning Gemma-2 model with FlashInfer attention backend on RTX A6000 (sm86) GPU triggers ValueError: Unsupported max_frags_z due to small shared memory size.

THENUpgrade FlashInfer to version 0.1.1 or later, which fixes the shared memory size check for sm86 GPUs. Use `pip install flashinfer>=0.1.1` or pin the version in your requirements file.

IFImportError: No module named 'langchain.embeddings.base' when importing llama_index after installing via pip.

THENDowngrade langchain to version 0.0.292 by running `pip install langchain==0.0.292` or pin langchain to <=0.0.292 in your requirements. Also consider updating llama_index to a version that supports newer langchain releases.

IFImporting AmazonTextractPDFLoader from langchain_community raises AttributeError: module 'sqlalchemy' has no attribute 'Select'.

THENUpgrade SQLAlchemy to version 2.0.0 or later using `pip install -U 'SQLAlchemy>=1.4,<3'`. This resolves the import error caused by the langchain_community code expecting the new `sa.Select` API introduced in SQLAlchemy 2.0.

IFImporting SentenceTransformer with transformers 4.52.2 raises NameError: name 'Replicate' is not defined.

THENDowngrade transformers to version 4.52.1 as a temporary workaround. The error occurs due to a missing import in the tensor_parallel integration module. Run `pip install transformers==4.52.1` to resolve until a fixed version is released.