Skip to content

Fix mutable default arguments and remove duplicate line in VLM models #130

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
jashshah999 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix mutable default arguments and remove duplicate line in VLM models #130

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 3 additions & 1 deletion cosmos_predict2/_src/predict2/text_encoders/reason1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,10 +231,12 @@ def _forward(
MODIFICATIONS: adding the hidden states to the output.
"""

def forward(self, tokens, data_batch={}, start_pos: int = 0) -> torch.Tensor:
def forward(self, tokens, data_batch=None, start_pos: int = 0) -> torch.Tensor:
"""
The training step of the model, including the loss computation.
"""
if data_batch is None:
data_batch = {}
assert "pixel_values" not in data_batch, "pixel_values should not be in data_batch, use images instead"
pixel_values = data_batch.get("images", None)
image_grid_thw = data_batch.get("image_grid_thw", None)
Expand Down
2 changes: 1 addition & 1 deletion cosmos_predict2/_src/reason1/models/vlm_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -414,7 +414,7 @@ def training_step(
def build_model(self, model_config):
raise NotImplementedError

def forward(self, tokens, data_batch={}, start_pos: int = 0) -> torch.Tensor:
def forward(self, tokens, data_batch=None, start_pos: int = 0) -> torch.Tensor:
"""
The forward pass of the model.
Returns:
Expand Down
4 changes: 3 additions & 1 deletion cosmos_predict2/_src/reason1/models/vlm_qwen.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,10 +370,12 @@ def _forward(
logits = DTensor.from_local(logits, device_mesh=self.cp_mesh, placements=[Shard(1)]).full_tensor()
return logits

def forward(self, tokens, data_batch={}, start_pos: int = 0) -> torch.Tensor:
def forward(self, tokens, data_batch=None, start_pos: int = 0) -> torch.Tensor:
"""
The training step of the model, including the loss computation.
"""
if data_batch is None:
data_batch = {}
assert "pixel_values" not in data_batch, "pixel_values should not be in data_batch, use images instead"
pixel_values = data_batch.get("images", None)
image_grid_thw = data_batch.get("image_grid_thw", None)
Expand Down
6 changes: 3 additions & 3 deletions cosmos_predict2/_src/reason1/models/vlm_qwen_omni.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -268,19 +268,19 @@ def _forward(
MODIFICATIONS: adding the hidden states to the output.
"""

def forward(self, tokens, data_batch={}, start_pos: int = 0) -> torch.Tensor:
def forward(self, tokens, data_batch=None, start_pos: int = 0) -> torch.Tensor:
"""
The training step of the model, including the loss computation.
"""
if data_batch is None:
data_batch = {}
assert "pixel_values" not in data_batch, "pixel_values should not be in data_batch, use images instead"
pixel_values = data_batch.get("images", None)
image_grid_thw = data_batch.get("image_grid_thw", None)
pixel_values_videos = data_batch.get("videos", None)
video_grid_thw = data_batch.get("video_grid_thw", None)
attention_mask = data_batch.get("padding_mask", None)

attention_mask = data_batch.get("padding_mask", None)

if image_grid_thw is not None:
assert len(image_grid_thw) == 1, "Only batch=1 is supported for now, due to `get_rope_index`"
image_grid_thw = image_grid_thw[0] # 1, N_img, 3 -> N_img, 3
Expand Down