Andrei Rusu

Sorry for intervening in your discussion, but I have a question regarding the workaround posted here for the OP issue: Is there any way one could bypass the non-deterministic issue...

I use edge attributes. If add_self_loops is enabled, a NotImplementedError arises (e.g. check the GATConv forward implementation if edge_dim is set). In more recent versions, one cannot actually omit setting...

OK. So, in ``gat_conv.py`` in the ``forward()`` method you have the following logic which returns an error for SparseTensors if ``add_self_loops`` is True and ``edge_dim`` is set: https://github.com/pyg-team/pytorch_geometric/blob/aa99b50c61743f3e33ef8a39ab6223c12f29786b/torch_geometric/nn/conv/gat_conv.py#L219-L239 At the...

Well, from my understanding ``GINEConv`` requires the ``edge_attr`` to have a dimensionality of ``in_channels``. ``GATConv``, on the other hand, requires ``edge_attr`` to have a dimensionality of ``heads * out_channels``. I...

No, the issue is not related to ``SystemMessages``. Instead, I believe that the integration does not support starting the chat history with an ``AIMessage``.

Yes @hinthornw , that was precisely the use case. In particular, I was trying to adapt to Gemini the streaming agent example from [here](https://github.com/langchain-ai/streamlit-agent/blob/main/streamlit_agent/basic_streaming.py), which passes to OpenAI the entire...

Functionally unrelated, but from the same block of code that seems to have been forgotten when cleaning this up for the release: The height of the IFrame is for some...