[HN Gopher] Attention Viz
       ___________________________________________________________________
        
       Attention Viz
        
       Author : djoldman
       Score  : 53 points
       Date   : 2023-05-13 11:01 UTC (12 hours ago)
        
 (HTM) web link (attentionviz.com)
 (TXT) w3m dump (attentionviz.com)
        
       | GaggiX wrote:
       | Did the page load for someone?
       | 
       | In the meanwhile I have found the paper page interesting:
       | https://catherinesyeh.github.io/attn-docs/
        
         | ape4 wrote:
         | There are probably thousands of us trying to load it then
         | bailing part way thru.
        
           | swyx wrote:
           | maybe its open source and some of us can rehost it?
        
         | banga wrote:
         | The UI tries connecting to
         | http://18.219.70.154:8500/getMatrixData/vit-nat for the data,
         | but that target appears to be unavailable.
        
         | swyx wrote:
         | it just loaded for me after 1 failed attempt
        
       | mrwnmonm wrote:
       | What does attention visualization mean?
        
       | kordlessagain wrote:
       | I've been working on a project I've called DocGPT:
       | https://github.com/featurebasedb/DocGPT. This is a work in
       | progress and I'll be putting together a release next week, once
       | I've got the instructions for running it done.
       | 
       | It was my assumption that GPT-4 could be used for keyterm and
       | question extraction from the fragments of a document and perhaps
       | used to build a semantic graph that could be leveraged for rapid
       | prompt assembly, augmenting the vector searches that are normally
       | done. The idea here is to set "attention flags" on the document
       | during discussion.
       | 
       | In this example, I indexed the arxiv PDF for the paper:
       | https://arxiv.org/abs/2305.03210.
       | 
       | Here are a few example keyterms it extracted, along with the
       | count of fragments containing the terms:
       | intermediate artifacts of self-attention 1       attention
       | patterns 3       key vectors 5       gpt-2 7       bert 15
       | 
       | Here's are some questions it asked and answered, using keyterms
       | in each fragment to identify relevant document fragments for
       | assembly (in addition to using nearest neighbor vectors):
       | 
       | system> How can attention head comparison and model pruning be
       | used to identify anomalies in transformer behaviour?
       | 
       | bot> Attention head comparison and model pruning can be used to
       | identify irregularities and potential behavioural issues with
       | transformers through attention pattern exploration, which can
       | provide insights into why large language models fail at reasoning
       | tasks or math. Comparing heads could allow to find parts of the
       | model that are actually useful, and tracking attention flow
       | throughout the model could help with hypothesis testing and
       | visualizing induction head pairs.
       | 
       | system> How does Attention Viz help to visualize multiple
       | datasets from different NLP tasks?
       | 
       | bot> AttentionViz can identify consistent shapes and patterns
       | across different NLP tasks and can also identify anomalies and
       | unexpected behavior, as well as facilitate attention pattern
       | comparisons.
       | 
       | system> What are the mean correlations between query-key
       | distances and dot products for BERT and GPT on Wiki-Auto data?
       | 
       | bot> -0.938 for BERT and -0.792 for GPT
       | 
       | system> What technologies are used in the Attention Viz
       | prototype?
       | 
       | bot> The AttentionViz prototype uses a Python/Flask backend, a
       | Vue/Typescript frontend, a DeepLabv3 segmentation model and
       | Deck.gl to visualize the corresponding query-key joint
       | embeddings.
        
         | qingcharles wrote:
         | This sounds like exactly what I need. I tried ChatPDF last
         | night and paid the Plus fee, but had terrible results, sadly.
         | 
         | I have about 9,000,000 PDFs I need summaries of and I can't
         | find a good AI solution as yet...
        
           | kordlessagain wrote:
           | This sound impressive and a good challenge. Let me know if I
           | can help. We'd need to do some map reducing on it.
        
       | Nzen wrote:
       | tl;dr a 'parse tree' for how their (transformer based) LLM
       | weights the import of the tokens in a given prompt. The authors
       | noticed patterns in the visualization that they felt imitated
       | relationships that we would regard as related.
       | 
       | That's as much as I can summarize, given my ignorance.
        
         | kordlessagain wrote:
         | Use the force!
         | 
         | user-aRxb[attentionviz.pdf]> summarize the document's contents
         | 
         | bot> Querying GPT...
         | 
         | bot> This paper describes AttentionViz, a tool designed to
         | provide a global view of Transformer Attention that can be used
         | to identify sentence view, clusters of search results, semantic
         | behavior, and fine-grained attention patterns; experts have
         | suggested additional interaction modes and global search
         | patterns to be used to quickly scan for and compare attention
         | trends.
        
       | Lio wrote:
       | The site is not responding for me. I'm guessing the page is
       | temporarily hugged to death.
       | 
       | In the meantime, I originally thought this might be linked to Viz
       | magazine[1][2]. Spoilers, it's not.
       | 
       | Warning: Viz deals with a very specific sort of British humour.
       | 
       | 1. https://viz.co.uk/
       | 
       | 2. https://en.wikipedia.org/wiki/Viz_(comics)
        
       ___________________________________________________________________
       (page generated 2023-05-13 23:01 UTC)