Skip to content

q115 research - poojith#5

Open
poojithdevan4D wants to merge 1 commit intoaloshdenny:mainfrom
poojithdevan4D:pooji_resnet
Open

q115 research - poojith#5
poojithdevan4D wants to merge 1 commit intoaloshdenny:mainfrom
poojithdevan4D:pooji_resnet

Conversation

@poojithdevan4D
Copy link
Copy Markdown

No description provided.

@aloshdenny aloshdenny changed the title Please enter the commit message for your changes. Lines starting q115 research - poojith Apr 20, 2026
@poojithdevan4D
Copy link
Copy Markdown
Author

@poojithdevan4D
Copy link
Copy Markdown
Author

poojithdevan4D commented Apr 22, 2026

Implemented as discussed, tried totally making it sf16, that is before moving to another layer, the accumulator.....
Some more changes too

@poojithdevan4D poojithdevan4D deleted the pooji_resnet branch April 22, 2026 20:10
@poojithdevan4D poojithdevan4D restored the pooji_resnet branch April 23, 2026 07:18
@poojithdevan4D
Copy link
Copy Markdown
Author

Final ResNet-20 Q1.15 Audit complete. We have achieved bit-exact parity between the training and the hardware-constrained inference pipeline. The model maintains 78.63% accuracy under strict Q1.15 integer-only constraints (including frequent quantization and folded BatchNorm). Attached is the formal Audit Report and the Golden Model script for hardware verification.

Q115_Audit_Report.pdf
q115_audit.py
q115_qat_training_v2.py

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant