New ingestion pipeline #279
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dharaneeshvrd
force-pushed
the
separate-summarization
branch
3 times, most recently
from
9b6771b to
fa1171e
Compare
manju956
reviewed
Feb 3, 2026
dharaneeshvrd
force-pushed
the
separate-summarization
branch
2 times, most recently
from
f82f6c3 to
06100e0
Compare
Member
Author
|
@manju956 Addressed your comments, ptal. |
yussufsh
previously approved these changes
Feb 5, 2026
Member
yussufsh
left a comment
yussufsh
left a comment
There was a problem hiding this comment.
Logic looks good to me.
This solves 2 problems:
- Handling bigger pdfs to control the memory usage
- Start processing steps as soon as the conversion is done for a pdf.
dharaneeshvrd
force-pushed
the
separate-summarization
branch
from
1075f6a to
06100e0
Compare
Niharika0306
reviewed
Feb 6, 2026
dharaneeshvrd
dismissed stale reviews from manju956 and manalilatkar
via
e8455e9
dharaneeshvrd
force-pushed
the
separate-summarization
branch
from
06100e0 to
e8455e9
Compare
Niharika0306
reviewed
Feb 6, 2026
Niharika0306
reviewed
Feb 6, 2026
dharaneeshvrd
force-pushed
the
separate-summarization
branch
5 times, most recently
from
d426e50 to
b130a0f
Compare
Instead of doing ingestion stage by stage via different methods
- New ingestion pipeline would look like below
- Generate current conversion status against the files available in cache to do only missing parts of ingestion pipeline. e.g. If conversion & processing is done, only chunking will be attempted.
- Split the files to process in light and heavy(pdf pages > 500) batches. This is required to process heavy files with less concurrency to avoid OOM kill. Light files = 4 workers, Heavy files = 2 workers
- Replace Process Pool with Thread Pool for processing & chunking since it involves only network call & counting tokens which does not require process which is required for CPU heavy tasks like conversion
- Start the ingestion pipeline
- Start conversion for pdfs which needs conversion
- As soon as a pdf’s conversion is completed, trigger its text & table processing
- Wait for all conversion to get completed
- As soon as a pdf’s processing is completed, trigger chunking
- Wait for all processing to get completed
- Wait for all chunking to get completed
- Improved stats by printing ingestion pipeline’s stage wise timings for all the pdfs
Signed-off-by: Dharaneeshwaran Ravichandran <dharaneeshwaran.ravichandran@ibm.com>
dharaneeshvrd
force-pushed
the
separate-summarization
branch
from
b130a0f to
f052251
Compare
Niharika0306
approved these changes
Feb 9, 2026
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Instead of doing ingestion stage by stage via different methods
To Address: AISERVICES-326 & other improvements