Skip to content

fix(slime): pin numpy < 2 so Megatron init does not abort the train worker #4

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
littlemex merged 2 commits into from
Jul 3, 2026
Merged

fix(slime): pin numpy < 2 so Megatron init does not abort the train worker #4

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9d833c8
fix(slime): pin numpy < 2 so Megatron init does not abort the train w…
littlemex Jul 3, 2026
3c97897
docs(slime): correct the numpy<2 pin comment and add a removal trigger
littlemex Jul 3, 2026
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: 20 additions & 0 deletions 3.test_cases/pytorch/slime/requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,26 @@
# SPDX-License-Identifier: MIT-0

# Reinforcement-learning Python dependencies for the SLIME image.

# Megatron-LM asserts numpy 1.x at init (slime .../megatron_utils/initialize.py:
# `assert np.__version__.startswith("1.")`, per NVIDIA/Megatron-LM#1563), but
# sglang[all] (installed just before this file in slime.Dockerfile) pulls in
# numpy 2.x transitively. Pin numpy < 2 so this requirements install downgrades
# it back to 1.x; without it the train worker aborts during Megatron init with
# "Megatron does not support numpy 2.x". Upstream SLIME's own docker/Dockerfile
# does the same (`pip install "numpy<2"`), so this mirrors the sanctioned pin.
#
# Why 1.x survives to runtime: the only pip steps after this one in the
# Dockerfile are the slime and sgl-router installs (--no-deps) and
# ring_flash_attn==0.1.8 (which declares no dependencies), so none of them
# reintroduce numpy 2.x. If a later step is added that pulls numpy WITHOUT
# --no-deps, re-pin numpy after it.
#
# TODO(numpy<2): remove this pin once MEGATRON_LM_VERSION is bumped to a commit
# that no longer asserts numpy 1.x (drops the numpy 2.x assert); until then the
# pin is required.
numpy<2

nltk==3.9.4
awscli==1.45.26
pynvml==13.0.1
Expand Down