Skip to content

fix: FP8 TP fixes #176

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
ani300 merged 2 commits into from
Aug 6, 2025
Merged

fix: FP8 TP fixes #176

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3b4ef87
Fix TP>1 for FP8 (and other qunatizations too), add missing import
ani300 Aug 6, 2025
3a006dc
Merge branch 'main' into fp8-tp-fixes
ani300 Aug 6, 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_mo/aiu_addons/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ def _infer_quantization_config(quant_config: dict) -> dict | None:
# First, import required FP8 linear classes from fms-mo
# Local
import fms_mo.aiu_addons.fp8.fp8_adapter # pylint: disable=unused-import
import fms_mo.aiu_addons.fp8.fp8_attn # pylint: disable=unused-import
import fms_mo.aiu_addons.fp8.fp8_linear # pylint: disable=unused-import

# This is used by get_linear to decide whether a linear layer
Expand Down
6 changes: 3 additions & 3 deletions fms_mo/aiu_addons/fp8/fp8_linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -321,12 +321,12 @@ def shard_fp8_linear(
sharding | param | shard | dim |
----------+----------------+-------+-----|
colwise | weight | Y | 0 |
| weight_scale | N | - |
| weight_scale | Y/N | 0/- |
| input_scale | N | - |
| bias | Y | 0 |
----------+----------------+-------+-----|
rowwise | weight | Y | 1 |
| weight_scale | Y/N | 0/- |
| weight_scale | N | - |
| input_scale | Y/N | 0/- |
| bias | 0 | - |
"""
Expand All @@ -339,7 +339,7 @@ def shard_fp8_linear(
]
# Scales are per-row or per-tensor
# Only sharding needed when row parallel and per-row
shard_scales = weight_strategy != "tensor" and module_info.sharding_dim == 1
shard_scales = weight_strategy != "tensor" and module_info.sharding_dim == 0
params: dict[str, LinearParameterShardingInfo] = {
"weight": LinearParameterShardingInfo(
module_info.sharding_dim, ShardType.SHARD
Expand Down
Loading