Skip to content

Migrate Gemma-3-1b-it to TT-Transformers Library #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
MohammedTaherMcW wants to merge 4 commits into
base: mcw/gemma_3_1b/pr_1_experimental
Choose a base branch
from
Open

Migrate Gemma-3-1b-it to TT-Transformers Library #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0e17632
Migrate Gemma-3-1b-it to TT-Transformers Library
MohammedTaherMcW Jul 29, 2025
8af16ce
Rebase RotarySetup
MohammedTaherMcW Aug 5, 2025
6e516f4
Removed Experimental Gemma-3-1b-it
MohammedTaherMcW Aug 5, 2025
777cfbe
Add Decoder Wrapper
MohammedTaherMcW Aug 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions models/common/rmsnorm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ def __init__(
torch_weight,
device=device,
dtype=weight_dtype,
layout=ttnn.ROW_MAJOR_LAYOUT,
layout=ttnn.TILE_LAYOUT,
memory_config=weight_memory_config,
cache_file_name=cache_name,
mesh_mapper=ttnn.ReplicateTensorToMesh(device) if is_mesh_device else None,
Expand All @@ -93,7 +93,7 @@ def __init__(
torch_weight,
device=device,
dtype=weight_dtype,
layout=ttnn.ROW_MAJOR_LAYOUT,
layout=ttnn.TILE_LAYOUT,
memory_config=weight_memory_config,
cache_file_name=cache_name,
mesh_mapper=ttnn.ShardTensor2dMesh(device, dims=(None, 2), mesh_shape=list(device.shape))
Expand Down Expand Up @@ -125,6 +125,11 @@ def forward(self, x: ttnn.Tensor, mode, in_sharded=False, out_sharded=False) ->
else:
assert not out_sharded, "Non-sharded version of RMSNorm cannot output a sharded tensor"

if x.shape[-1] % weight.shape[-1] == 0:
# Reshape weight only if x's last dimension is divisible by weight's last dimension,
# to avoid padding errors in RMSNorm when dimensions are not aligned
weight = ttnn.reshape(weight, [1, 1, 1, -1])

x = norm(
x,
epsilon=self.eps,
Expand Down
277 changes: 0 additions & 277 deletions models/experimental/gemma3_1b/tests/test_attention.py
View file
Open in desktop

This file was deleted.

Loading