-
### Proposal
Please add support for finetuned-variants of pretrained models, like Pythia < 1B and gpt2-small.
### Motivation
I'm frustrated when I find a very appropriate dataset and circuit…
-
### 🐛 Describe the bug
The current `DTensor` sharding propagation caching policy for `aten.scaled_dot_product_efficient_attention` (default) can result in silently incorrect gradients or trigger a…
-
### What is your request?
The Max Engine can interface with GPU code, and supports access to both the SRAM and HBM needed for the latest LLM GPU architectures.
### What is your motivation for this c…
-
How do I log plt -> plotly plots without converting them to images with pytorch lightning? cf https://docs.wandb.ai/guides/track/log/plots#matplotlib-and-plotly-plots
-
https://github.com/JasonGross/guarantees-based-mechanistic-interpretability/blob/d87b9c746bcdb3e1c81d8f5b2f80dd987e6ca7aa/.gitignore#L97-L102
Maybe we should remove
https://github.com/JasonGross/gua…
-
I'm starting a new issue rather than continue the already lengthy #16, though much of what we discussed there informs this topic.
At long last, I have some updates on the first stage of our expans…
-
Click the edit wheel on the right of "About" [here](https://github.com/neelnanda-io/TransformerLens/) and add:
https://neelnanda-io.github.io/TransformerLens/
Maybe add a one-liner description a…
-
-
For arxiv papers, specifically.
By the way, Vik, if you'd like a paper with challenging tables for testing marker, here's one:
https://arxiv.org/abs/2311.15131
Here's what I get from a conversi…
-
Every few months I try to use vscode for jupyter because I would really love to just use vscode for everything. Every few months, I am disappointed and switch back to the web version.
There are two…