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

Enforce chunk_size as multiple of 256 #243

Merged
merged 1 commit into from
Nov 18, 2024
Merged

Enforce chunk_size as multiple of 256 #243

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
15 changes: 13 additions & 2 deletions backends/exllamav2/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -321,9 +321,20 @@ async def create(cls, model_directory: pathlib.Path, quiet=False, **kwargs):
if num_experts_override:
self.config.num_experts_per_token = kwargs.get("num_experts_per_token")

# Make sure chunk size is >= 16 and <= max seq length
# Make sure chunk size is >= 256, keep near or below max seq len
user_chunk_size = unwrap(kwargs.get("chunk_size"), 2048)
chunk_size = sorted((16, user_chunk_size, self.config.max_seq_len))[1]
chunk_size = sorted((256, user_chunk_size, self.config.max_seq_len))[1]
chunk_remainder = chunk_size % 256
if chunk_remainder != 0:
rounded_chunk_size = int(256 * ((chunk_size - chunk_remainder) / 256 + 1))

logger.warning(
f"The given chunk size ({chunk_size}) is "
"not a multiple of 256.\n"
"Overriding chunk_size with an overestimated value of "
f"{rounded_chunk_size} tokens."
)
chunk_size = rounded_chunk_size
self.config.max_input_len = chunk_size
self.config.max_attention_size = chunk_size**2

Expand Down
Loading