Sign up to save your podcasts
Or
Share “You can remove GPT2’s LayerNorm by fine-tuning for an hour” by StefanHex
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
“You can remove GPT2’s LayerNorm by fine-tuning for an hour” by StefanHex
This work was produced at Apollo Research, based on initial research done at MATS.
LayerNorm is annoying for mechanstic interpretability research (“[...] reason #78 for why interpretability researchers hate LayerNorm” – Anthropic, 2023).
Here's a Hugging Face link to a GPT2-small model without any LayerNorm.
The final model is only slightly worse than a GPT2 with LayerNorm[1]:
DatasetOriginal GPT2Fine-tuned GPT2 with LayerNormFine-tuned GPT without LayerNormOpenWebText (ce_loss)3.0952.9893.014 (+0.025)ThePile (ce_loss)2.8562.8802.926 (+0.046)HellaSwag (accuracy)29.56%29.82%29.54%I fine-tuned GPT2-small on OpenWebText while slowly removing its LayerNorm layers, waiting for the loss to go back down after reach removal:
Introduction
LayerNorm (LN) is a component in Transformer models that normalizes embedding vectors to have constant length; specifically it divides the embeddings by their standard deviation taken over the hidden dimension. It was originally introduced to stabilize and speed up training of models (as a replacement for batch normalization). It is active during training and inference.
_mathrm{LN}(x) = frac{x - [...] ---
Outline:
(01:11) Introduction
(02:45) Motivation
(03:33) Method
(09:15) Implementation
(10:40) Results
(13:59) Residual stream norms
(14:32) Discussion
(14:35) Faithfulness to the original model
(15:45) Does the noLN model generalize worse?
(16:13) Appendix
(16:16) Representing the no-LayerNorm model in GPT2LMHeadModel
(18:08) Which order to remove LayerNorms in
(19:28) Which kinds of LayerNorms to remove first
(20:29) Which layer to remove LayerNorms in first
(21:13) Data-reuse and seeds
(21:35) Infohazards
(21:58) Acknowledgements
The original text contained 4 footnotes which were omitted from this narration.
The original text contained 5 images which were described by AI.
---
First published:
August 8th, 2024
Source:
https://www.lesswrong.com/posts/THzcKKQd4oWkg4dSP/you-can-remove-gpt2-s-layernorm-by-fine-tuning-for-an-hour
---
Narrated by TYPE III AUDIO.
---
Images from the article:
LayerNorm is annoying for mechanstic interpretability research (“[...] reason #78 for why interpretability researchers hate LayerNorm” – Anthropic, 2023).
Here's a Hugging Face link to a GPT2-small model without any LayerNorm.
The final model is only slightly worse than a GPT2 with LayerNorm[1]:
DatasetOriginal GPT2Fine-tuned GPT2 with LayerNormFine-tuned GPT without LayerNormOpenWebText (ce_loss)3.0952.9893.014 (+0.025)ThePile (ce_loss)2.8562.8802.926 (+0.046)HellaSwag (accuracy)29.56%29.82%29.54%I fine-tuned GPT2-small on OpenWebText while slowly removing its LayerNorm layers, waiting for the loss to go back down after reach removal:
Introduction
LayerNorm (LN) is a component in Transformer models that normalizes embedding vectors to have constant length; specifically it divides the embeddings by their standard deviation taken over the hidden dimension. It was originally introduced to stabilize and speed up training of models (as a replacement for batch normalization). It is active during training and inference.
_mathrm{LN}(x) = frac{x - [...] ---
Outline:
(01:11) Introduction
(02:45) Motivation
(03:33) Method
(09:15) Implementation
(10:40) Results
(13:59) Residual stream norms
(14:32) Discussion
(14:35) Faithfulness to the original model
(15:45) Does the noLN model generalize worse?
(16:13) Appendix
(16:16) Representing the no-LayerNorm model in GPT2LMHeadModel
(18:08) Which order to remove LayerNorms in
(19:28) Which kinds of LayerNorms to remove first
(20:29) Which layer to remove LayerNorms in first
(21:13) Data-reuse and seeds
(21:35) Infohazards
(21:58) Acknowledgements
The original text contained 4 footnotes which were omitted from this narration.
The original text contained 5 images which were described by AI.
---
First published:
August 8th, 2024
Source:
https://www.lesswrong.com/posts/THzcKKQd4oWkg4dSP/you-can-remove-gpt2-s-layernorm-by-fine-tuning-for-an-hour
---
Narrated by TYPE III AUDIO.
---
Images from the article:
View all episodes
By LessWrong
4.8
1212 ratings
This work was produced at Apollo Research, based on initial research done at MATS.
LayerNorm is annoying for mechanstic interpretability research (“[...] reason #78 for why interpretability researchers hate LayerNorm” – Anthropic, 2023).
Here's a Hugging Face link to a GPT2-small model without any LayerNorm.
The final model is only slightly worse than a GPT2 with LayerNorm[1]:
DatasetOriginal GPT2Fine-tuned GPT2 with LayerNormFine-tuned GPT without LayerNormOpenWebText (ce_loss)3.0952.9893.014 (+0.025)ThePile (ce_loss)2.8562.8802.926 (+0.046)HellaSwag (accuracy)29.56%29.82%29.54%I fine-tuned GPT2-small on OpenWebText while slowly removing its LayerNorm layers, waiting for the loss to go back down after reach removal:
Introduction
LayerNorm (LN) is a component in Transformer models that normalizes embedding vectors to have constant length; specifically it divides the embeddings by their standard deviation taken over the hidden dimension. It was originally introduced to stabilize and speed up training of models (as a replacement for batch normalization). It is active during training and inference.
_mathrm{LN}(x) = frac{x - [...] ---
Outline:
(01:11) Introduction
(02:45) Motivation
(03:33) Method
(09:15) Implementation
(10:40) Results
(13:59) Residual stream norms
(14:32) Discussion
(14:35) Faithfulness to the original model
(15:45) Does the noLN model generalize worse?
(16:13) Appendix
(16:16) Representing the no-LayerNorm model in GPT2LMHeadModel
(18:08) Which order to remove LayerNorms in
(19:28) Which kinds of LayerNorms to remove first
(20:29) Which layer to remove LayerNorms in first
(21:13) Data-reuse and seeds
(21:35) Infohazards
(21:58) Acknowledgements
The original text contained 4 footnotes which were omitted from this narration.
The original text contained 5 images which were described by AI.
---
First published:
August 8th, 2024
Source:
https://www.lesswrong.com/posts/THzcKKQd4oWkg4dSP/you-can-remove-gpt2-s-layernorm-by-fine-tuning-for-an-hour
---
Narrated by TYPE III AUDIO.
---
Images from the article:
LayerNorm is annoying for mechanstic interpretability research (“[...] reason #78 for why interpretability researchers hate LayerNorm” – Anthropic, 2023).
Here's a Hugging Face link to a GPT2-small model without any LayerNorm.
The final model is only slightly worse than a GPT2 with LayerNorm[1]:
DatasetOriginal GPT2Fine-tuned GPT2 with LayerNormFine-tuned GPT without LayerNormOpenWebText (ce_loss)3.0952.9893.014 (+0.025)ThePile (ce_loss)2.8562.8802.926 (+0.046)HellaSwag (accuracy)29.56%29.82%29.54%I fine-tuned GPT2-small on OpenWebText while slowly removing its LayerNorm layers, waiting for the loss to go back down after reach removal:
Introduction
LayerNorm (LN) is a component in Transformer models that normalizes embedding vectors to have constant length; specifically it divides the embeddings by their standard deviation taken over the hidden dimension. It was originally introduced to stabilize and speed up training of models (as a replacement for batch normalization). It is active during training and inference.
_mathrm{LN}(x) = frac{x - [...] ---
Outline:
(01:11) Introduction
(02:45) Motivation
(03:33) Method
(09:15) Implementation
(10:40) Results
(13:59) Residual stream norms
(14:32) Discussion
(14:35) Faithfulness to the original model
(15:45) Does the noLN model generalize worse?
(16:13) Appendix
(16:16) Representing the no-LayerNorm model in GPT2LMHeadModel
(18:08) Which order to remove LayerNorms in
(19:28) Which kinds of LayerNorms to remove first
(20:29) Which layer to remove LayerNorms in first
(21:13) Data-reuse and seeds
(21:35) Infohazards
(21:58) Acknowledgements
The original text contained 4 footnotes which were omitted from this narration.
The original text contained 5 images which were described by AI.
---
First published:
August 8th, 2024
Source:
https://www.lesswrong.com/posts/THzcKKQd4oWkg4dSP/you-can-remove-gpt2-s-layernorm-by-fine-tuning-for-an-hour
---
Narrated by TYPE III AUDIO.
---
Images from the article:
More shows like LessWrong (Curated & Popular)
View all
Making Sense with Sam Harris
26,375 Listeners
Conversations with Tyler
2,424 Listeners
Robert Wright's Nonzero
590 Listeners
Future of Life Institute Podcast
107 Listeners
The Good Fight
904 Listeners
ManifoldOne
92 Listeners
The Prof G Pod with Scott Galloway
5,493 Listeners
Machine Learning Street Talk (MLST)
90 Listeners
Dwarkesh Podcast
75 Listeners
Hard Fork
5,469 Listeners
Clearer Thinking with Spencer Greenberg
130 Listeners
Complex Systems with Patrick McKenzie (patio11)
133 Listeners
The Marginal Revolution Podcast
93 Listeners
Statecraft
35 Listeners
The Last Invention
704 Listeners