Updated Resnet50 model for Blackhole, with Batch = 32 #17985
+160
−84
Conversation
This file contains 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
mywoodstock
force-pushed
the
asarje/rn50-bh-largebatch-20250218
branch
2 times, most recently
from
72712db to
90fb639
Compare
mywoodstock
changed the title
mywoodstock
requested review from
tt-aho,
a team,
ayerofieiev-tt,
dmakoviichuk-tt,
rfurko-tt,
cfjchu,
TT-BrianLiu,
razorback3,
dongjin-na,
bbradelTT,
omilyutin-tt,
yugaoTT,
vsureshTT,
edwinleeTT,
ntarafdar,
sjameelTT,
jaykru-tt,
yugi957,
jvegaTT,
llongTT and
nardoTT
as code owners
bbradelTT
reviewed
Feb 27, 2025
| @@ -202,7 +202,7 @@ void MAIN { | |||
| #ifdef ARCH_BLACKHOLE | |||
| // FIXME: This is a temporary workaround to avoid hangs on blackhole. | |||
| // https://github.com/tenstorrent/tt-metal/issues/16439 | |||
| for (uint32_t i = 0; i < 10; i++) { | |||
| for (uint32_t i = 0; i < 100; i++) { | |||
There was a problem hiding this comment.
Do we really need to add an order of magnitude to the delay?
Someone said this was a 30% slowdown that is now a 300% slowdown.
There was a problem hiding this comment.
oops, this snuck in --> didn't mean to push this in
| @@ -914,6 +966,19 @@ def run(self, input_tensor, device, ops_parallel_config, conv_op_cache={}) -> tt | |||
| ), | |||
| } | |||
| ) | |||
| # ## 128 | |||
| # core_range_set = ttnn.CoreRangeSet( | |||
There was a problem hiding this comment.
Please remove commented out code here and in other parts of the PR.
| and layer_module | ||
| and (layer_module == "layer1_module2" or layer_module == "layer1_module3") | ||
| ): | ||
| conv_kwargs_2["conv_config"].act_block_h_override = 0 |
There was a problem hiding this comment.
At this point it may be better to have a function go through all three cases, and then just do something like
if f(batch_size, layer_module):
conv_kwargs_2["conv_config"].act_block_h_override = 0
yugaoTT
reviewed
Feb 27, 2025
mywoodstock
force-pushed
the
asarje/rn50-bh-largebatch-20250218
branch
from
af8258a to
cad363d
Compare
mywoodstock
force-pushed
the
asarje/rn50-bh-largebatch-20250218
branch
from
0648d6a to
752a333
Compare
yugaoTT
approved these changes
Feb 27, 2025
bbradelTT
approved these changes
Feb 27, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Ticket
#17393
#18341
Problem description
This PR enables larger batch sizes 20 and 32 for Resnet50 on Blackhole.
What's changed
Some updates to the model itself to allow batch 32.
More updates to the fold op to allow non-rectangular grids.
Checklist