Skip to content

Integration tpu#40

Open
Vishal-S-P wants to merge 230 commits intomainfrom
integration_tpu
Open

Integration tpu#40
Vishal-S-P wants to merge 230 commits intomainfrom
integration_tpu

Conversation

@Vishal-S-P
Copy link
Copy Markdown
Collaborator

@Vishal-S-P Vishal-S-P commented Jun 28, 2023

Verified code for training on v3-8 TPU

Vishal-S-P added 30 commits April 5, 2023 20:05
@Vishal-S-P
updated readme
ba9076f
@Vishal-S-P
updated readme
bca22af
@Vishal-S-P
The dataloader output None
3b47bb2
@Vishal-S-P
testing modeling folder
d30d4c4
@Vishal-S-P
maskformer arch test successful.
5062ec8
@Vishal-S-P
started debugging loss function
c3aba68
@Vishal-S-P
linear sum assignment does not work for rectangular cost matrix
3f6e030
@Vishal-S-P
fixing cross entropy loss calcualtion
79cf603
@Vishal-S-P
Made cost matrix square and compatible with matcher
2915e03
@Vishal-S-P
verified the scipy and tf version of matcher gives same indices as ou…
266592f 
…tput
@Vishal-S-P
updated NestedTensor
77f7d62
@Vishal-S-P
latest
8dbc212
@Vishal-S-P
fixed large CE loss issue dice and focal loss do not match with Pytor…
59b066a 
…ch implementation yet
@Vishal-S-P
fixed th e issue with the assignments between pytorch and tensorflow …
2ea7cda 
…matcher
@Vishal-S-P
focal loss matches with the pytorch code
f8ce6c6
@Vishal-S-P
fixed all losses working on aux losses
8e5d782
@Vishal-S-P
dataloader not working as expected
cc7bcde
@Vishal-S-P
working on dataloader
bfefafe
@Vishal-S-P
dataloader working issue
30cd1a4
@Vishal-S-P
binary masks done
42e34d6
@Vishal-S-P
almost close to training
365b400
@Vishal-S-P
first version of the working training code
f67ee4f
@Vishal-S-P
new branch
7774725
@Vishal-S-P
update
52e675b
@Vishal-S-P
update for TPU
88f0fd2
@Vishal-S-P
updated path
180b0fd
@Vishal-S-P
updated shapes
94097ec
@Vishal-S-P
upated loss loop
563ca07
@Vishal-S-P
updated code
2594f76
@Vishal-S-P
tpu update
540a6fe
Vishal-S-P added 30 commits June 18, 2023 05:31
@Vishal-S-P
working tpu v2-256 europe code
12430d2
@Vishal-S-P
working tpu v2-256 US code
6ce00d6
@Vishal-S-P
working tpu v2-256 US code
cc4e496
@Vishal-S-P
working tpu v2-256 US code
f182bfa
@Vishal-S-P
working tpu v2-256 US code
cf0a8ff
@Vishal-S-P
working tpu v2-256 US code
03c9272
@Vishal-S-P
added clip or pad to fixed size
3352782
@Vishal-S-P
updated batch size to 64, number of steps per epoch to 100
2c4448a
@Vishal-S-P
update
0bd746c
@Vishal-S-P
undo the batch size to 8
a30d2c2
@Vishal-S-P
added a line for prifiler
4ef5286
@Vishal-S-P
for tf-debug-1
fa7fded
@Vishal-S-P
for tf-debug-1
38d3293
@Vishal-S-P
for tf-debug-1
2d1e4c8
@Vishal-S-P
code failed to run on tf-train-1
c3bb923
@Vishal-S-P
eu tpu v3-8
1554325
@Vishal-S-P
added data cache and added profiler code
ab1a956
@Vishal-S-P
removed profile server and added profile capture programatically
c7462d6
@Vishal-S-P
updated code to run on faster VM
ddf957a
@Vishal-S-P
updated code to run on faster VM
a801143
@Vishal-S-P
updated code to run on faster VM
e7a8c5b
@Vishal-S-P
updated code to run on faster VM
be65563
@Vishal-S-P
added bfloat16 to the parser which used float32 before
3a8bd0c
@Vishal-S-P
updated batch size to 64 and made dtype bfloat16
fed7003
@Vishal-S-P
updated batch size to 8 and made dtype bfloat16
1d6865a
@Vishal-S-P
added mixed precision training bfloat16
1e1d8e7
@Vishal-S-P
update
cb2148c
@Vishal-S-P
updated code
3958af2
@Vishal-S-P
Merge branch 'integration_tpu' of https://github.com/PurdueDualityLab…
c5dd0a1 
…/tf-maskformer into integration_tpu
@Vishal-S-P
update
9f0a817
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@Vishal-S-P @Wenxin-Jiang