Skip to content
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

Update torchtune generation to be more flexible #1970

Merged
merged 1 commit into from
Nov 8, 2024
Merged

Update torchtune generation to be more flexible #1970

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
20 changes: 12 additions & 8 deletions torchtune/generation/_generation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ def generate_next_token(
model: TransformerDecoder,
input_pos: torch.Tensor,
x: torch.Tensor,
q: torch.Tensor,
q: Optional[torch.Tensor] = None,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Need to update the docstring to reflect the new typing, lint is failing

*,
mask: Optional[torch.Tensor] = None,
temperature: float = 1.0,
Expand All @@ -82,7 +82,7 @@ def generate_next_token(
with shape [bsz x seq_length].
x (torch.Tensor): tensor with the token IDs associated with the given prompt,
with shape [bsz x seq_length].
q (torch.Tensor): randomly sampled tensor for softmax sampling trick.
q (Optional[torch.Tensor]): randomly sampled tensor for softmax sampling trick.
See https://github.com/pytorch-labs/gpt-fast/blob/32971d3129541c5bfb4f715abc33d1c5f408d204/generate.py#L40
mask (Optional[torch.Tensor]): attention mask with shape [bsz x seq_length x seq_length],
default None.
Expand Down Expand Up @@ -302,9 +302,11 @@ def generate(
# tensors are of identical shape to the prompt
curr_masks = masks[:, :prompt_length, :prompt_length]

q = torch.empty(
(bsz, model.tok_embeddings.num_embeddings), device=prompt.device
).exponential_(1, generator=rng)
q = None
if rng is not None:
q = torch.empty(
(bsz, model.tok_embeddings.num_embeddings), device=prompt.device
).exponential_(1, generator=rng)
tokens, generated_logits = generate_next_token(
model,
input_pos=input_pos[:, :prompt_length].squeeze(),
Expand Down Expand Up @@ -360,9 +362,11 @@ def generate(
curr_input_pos = input_pos[:, : curr_pos + 1]
curr_masks = masks[:, : curr_pos + 1, : curr_pos + 1]

q = torch.empty(
(bsz, model.tok_embeddings.num_embeddings), device=prompt.device
).exponential_(1, generator=rng)
q = None
if rng is not None:
q = torch.empty(
(bsz, model.tok_embeddings.num_embeddings), device=prompt.device
).exponential_(1, generator=rng)
tokens, logits = custom_generate_next_token(
model,
input_pos=curr_input_pos,
Expand Down
Loading