Pinned
Hello world! 🐛
After a year of restless development, I'm finally happy to announce Inseq, a new tool to democratize post-hoc interpretability of sequence generation models 🐛 github.com/inseq-team/ins… #nlproc #xai
Some highlights 👇 1/
Inseq
@InseqLib
Open-Source Interpretability for Generative Language Models 🔎 🐛
- Thanks to the new treescope integration, @InseqLib now supports interactive visualizations for multidimensional attributions (show_granular), token highlights (show_tokens) and improved viz for attribute_context CLI! 🚀 Install main, will appear in v0.7 x.com/_ddjohnson/sta…By popular demand, the Treescope pretty-printer from the Penzai neural net library can now be installed separately, and supports both JAX and PyTorch! And that's not all: Penzai itself now has less boilerplate and includes more pretrained Transformer models!
- The 🐑 PECoRe / 🌴 MIRAGE demo on @huggingface Spaces is powered by our new attribute-context CLI command released in v0.6, and allows to export the code to reproduce your results locally with 🐛 Inseq. Check it out ➡️ hf.co/spaces/gsarti/…⚠️ Citations from prompting or NLI seem plausible, but may not faithfully reflect LLM reasoning. 🏝️ MIRAGE detects context dependence in generations via model internals, producing granular and faithful RAG citations. 🚀 Demo: huggingface.co/spaces/gsarti/… Fun collab w/ @Jirui_Qi,
- Today @InseqLib hit 300 ⭐️ on Github! A huge thank you to all our awesome users ❤️ Onwards to the next 300! 🤺
- @InseqLib v0.6 is out now on PyPI! 🔥 New CLI command for context attribution (@gsarti_), new perturbation-based methods by @hmohebbi75 & @casszzx and optimizations incl. multi-gpu support! ⚡️ Huge shoutout to our contributors! ❤️ Release notes ⬇️
![[points to Pyvene]
Inseq: You there, what is your profession?
Pyvene: I do trainable inference-time interventions... sir.
Inseq: [points to another library] And you, TransformerLens, what is your profession?
TransformerLens: I provide a unified hookable architecture to support mechinterp analyses on Transformer models, sir.
Inseq: I see...
[turns to a third soldier]
Inseq: You?
SAELens: I enable the training of sparse autoencoders to mitigate superposition in latent features.
Inseq: [turns back shouting] INSEQ USERS! What is OUR profession?
Inseq Users: HA-OOH! HA-OOH! HA-OOH!
Inseq Users: We do feature attribution of generative LMs! 😇](https://pbs.twimg.com/media/GMlicq3X0AM_NNG?format=webp&name=medium)
