[RAG] Integrate "Retrieval-Augmented Generation" with Haystack by lalitpagaria · Pull Request #484 · deepset-ai/haystack

lalitpagaria · 2020-10-13T22:03:26Z

To deliver #443

Task List -

Not part of this PR but could be address in other PRs -

Take care of TODOs in RAGenerator by upgrading transformers >= 3.4.0
Add API for generator
Add generator info on website
Support generator in finder
Support RAG-S ([RAG] How to extract generated strings from RetrievAugLMMarginOutput huggingface/transformers#7905)
Evaluate RAG (https://github.com/huggingface/transformers/blob/master/examples/rag/eval_rag.py)
Fine tune support (https://github.com/huggingface/transformers/blob/master/examples/rag/finetune.py)

tholor

Nice! Left a few early comments ...

haystack/finder.py

haystack/generator/rag.py

haystack/retriever/dense.py

tutorials/Tutorial7_generator.py

tholor

Good progress! Left a few comments ...

haystack/generator/transformers.py

lalitpagaria · 2020-10-22T10:27:00Z

@tholor PR is almost ready except -

TODOs: These need transformers 3.4.0 (Bump transformers version to 3.4.0 FARM#598)
Info on website

So can you please review it

lalitpagaria · 2020-10-22T11:19:22Z

Also tests are failing with 137 error code, is it possible to increase memory of containers?

Error: Process completed with exit code 137.

tholor

Looking great. Left minor comments about names, typos etc

haystack/generator/transformers.py

tutorials/Tutorial7_RAG_Generator.ipynb

…ectly and need to convert embedding to tensor way

…rator implementation

…xt_input_ids

…im memory by terminating containers

…n CI and pytest splits

tholor

Looking good! Awesome job @lalitpagaria 👍

tholor reviewed Oct 14, 2020

View reviewed changes

tholor reviewed Oct 19, 2020

View reviewed changes

haystack/generator/transformers.py Outdated Show resolved Hide resolved

haystack/generator/transformers.py Outdated Show resolved Hide resolved

haystack/generator/transformers.py Outdated Show resolved Hide resolved

haystack/generator/transformers.py Outdated Show resolved Hide resolved

tholor mentioned this pull request Oct 20, 2020

Make visible an interface to the Inverse Cloze Task (ORQA Fine-tuning) #475

Closed

tholor reviewed Oct 23, 2020

View reviewed changes

lalitpagaria changed the title ~~WIP: [RAG] Adding generator implementation~~ [RAG] Adding generator implementation Oct 23, 2020

lalitpagaria changed the title ~~[RAG] Adding generator implementation~~ [RAG] Integrate "Retrieval-Augmented Generation" with Haystack Oct 23, 2020

tholor mentioned this pull request Oct 23, 2020

Add Generative QA Models like RAG #443

Closed

lalitpagaria added 21 commits October 30, 2020 17:31

Adding dummy generator implementation

54b42ed

Adding tutorial to try the model

09c3a1f

Committing current non working code

356e87e

Committing current update where we need to call generate function dir…

dbf289b

…ectly and need to convert embedding to tensor way

Addressing review comments.

460b212

Refactoring finder, and implementing rag_generator class.

0ad84fb

Refined the implementation of RAGGenerator and now it is in clean shape

9465fc5

Renaming RAGGenerator to RAGenerator

470d03f

Reverting change from finder.py and addressing review comments

d2d7ff1

Remove support for RagSequenceForGeneration

3866318

Utilizing embed_passage function from DensePassageRetriever

beed242

Adding sample test data to verify generator output

b1a7e63

Updating testing script

c83ca05

Updating testing script

fa0b03b

Fixing bug related to top_k

506d029

Updating latest farm dependency

c4ec2a3

Comment out farm dependency

a01baee

Reverting changes from TransformersReader

f378ba3

Adding transformers dataset to compare transformers and haystack gene…

2561c5e

…rator implementation

Using generator_encoder instead of question_encoder to generate conte…

78ad744

…xt_input_ids

Adding workaround to install FARM dependency from master branch

5bca5ab

lalitpagaria added 27 commits October 30, 2020 17:32

Refactoring tests to reduce memory footprint

6f95b47

Split generator tests in separate ci step and before running it recla…

35f2c79

…im memory by terminating containers

Moving tika dependent test to separate dir

68b4f43

Remove unwanted code

97252cb

Brining reader under session scope

c288562

Farm is now session object hence restoring changes from default value

dab76d7

Updating assert for pdf converter

c2758b6

Dummy commit to trigger CI flow

151caa9

REducing memory footprint required for generator tests

721ba50

Fixing mypy issues

f769373

Marking test with tika and elasticsearch markers. Reverting changes i…

874b51b

…n CI and pytest splits

reducing changes

eb72cc0

Fixing CI

b89bc76

changing elastic search ci

bdea1f8

Fixing test error

371c76f

Disabling return of embedding

3fed37d

Marking generator test as well

00c2591

Refactoring tutorials

4ea0b6f

Increasing ES memory to 750M

fa10872

Trying another fix for ES CI

26efb7e

Reverting CI changes

d8bb748

Splitting tests in CI

704bbb8

Generator and non-generator markers split

0dd1cf1

Adding pytest.ini to add markers and enable strict-markers option

2ea04eb

Reducing elastic search container memory

e7643c7

Simplifying generator test by using documents with embedding directly

a99e2ec

Bump up farm to 0.5.0

dadbd3f

tholor approved these changes Oct 30, 2020

View reviewed changes

tholor merged commit f134430 into deepset-ai:master Oct 30, 2020

lalitpagaria deleted the implement_RAG branch October 30, 2020 17:06

Conversation

lalitpagaria commented Oct 13, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tholor left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tholor left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lalitpagaria commented Oct 22, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lalitpagaria commented Oct 22, 2020

Uh oh!

tholor left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tholor left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

lalitpagaria commented Oct 13, 2020 •

edited

Loading

lalitpagaria commented Oct 22, 2020 •

edited

Loading