Skip to content

Expose doc_breakpoint #147

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

Merged
daviswer merged 15 commits into from
May 29, 2025
Merged

Expose doc_breakpoint #147

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
6a46c16
Add fim
daviswer Jan 10, 2025
673e6e5
Blacking
daviswer Jan 10, 2025
d3cc468
Corrected fim/clm combo
daviswer Jan 10, 2025
d5da300
reblacking
daviswer Jan 10, 2025
1981fa9
Rereblacking
daviswer Jan 10, 2025
9365cb2
Add multicol support
daviswer May 23, 2025
1892149
Add everything else
daviswer May 23, 2025
426dd16
Some cleanup (no continue)
daviswer May 23, 2025
15f4d7e
Blacking
daviswer May 23, 2025
a58eecd
Reblacking
daviswer May 23, 2025
4c26fb4
isorting
daviswer May 23, 2025
0f667be
Make doc cutoff an arg
daviswer May 27, 2025
c9f2bdb
Re-insert missing countfile search lines
daviswer May 27, 2025
6d751e5
Expose doc breakpoint in cfg
daviswer May 29, 2025
7521297
Merge branch 'mamba-tiktoken' into data-fixes
daviswer May 29, 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
1 change: 1 addition & 0 deletions fms_fsdp/config/training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,7 @@ class train_config:
logical_shards: int = 1024
num_workers: int = 1
doc_cutoff: int = 1_000_000
doc_breakpoint: int = 65_536

# fsdp policies
sharding_strategy: str = "hsdp"
Expand Down
4 changes: 3 additions & 1 deletion fms_fsdp/utils/dataloader_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
import torch
from math import ceil

from fms_fsdp.utils.dataset_utils import (
ArrowHandler,
Expand Down Expand Up @@ -94,7 +95,7 @@ def get_data_loader(cfg, rank, world_size):
)
else:
filehandler = _handler_map[cfg.file_type](cols)

# Base reader layer
data = StreamingDocDataset(
cfg.data_path,
Expand All @@ -105,6 +106,7 @@ def get_data_loader(cfg, rank, world_size):
bos_token=cfg.bos_token,
strip_tokens=set(droplist),
min_length=3,
max_consecutive_chunks=ceil(cfg.doc_breakpoint/1024),
seed=cfg.seed,
)
# Add rescaling/resharding
Expand Down
Loading