Skip to content

[model] fix: correct GLM-4.5V inference parallelism for 46-layer model #2322

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
yaoyu-33 merged 2 commits into from
Feb 11, 2026
Merged

[model] fix: correct GLM-4.5V inference parallelism for 46-layer model #2322

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
ccb2456
fix: update glm-4.5v inference parallelism and generation temp workar…
yaoyu-33 Feb 11, 2026
8e0a34b
Merge branch 'main' into yuya/fix-glm45v-inference-parallelism
yaoyu-33 Feb 11, 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
4 changes: 4 additions & 0 deletions examples/conversion/hf_to_megatron_generate_text.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,6 +167,10 @@ def main(args) -> None:
model_provider.initialize_model_parallel(seed=0)
model = model_provider.provide_distributed_model(wrap_with_ddp=False)

# TEMP FIX for inference failure when mtp_num_layers is not None
for m in model:
m.config.mtp_num_layers = None

model = [m.cuda() for m in model]
for m in model:
m.eval()
Expand Down
4 changes: 4 additions & 0 deletions examples/conversion/hf_to_megatron_generate_vlm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -253,6 +253,10 @@ def main(args) -> None:
model_provider.initialize_model_parallel(seed=0)
model = model_provider.provide_distributed_model(wrap_with_ddp=False)

# TEMP FIX for inference failure when mtp_num_layers is not None
for m in model:
m.config.mtp_num_layers = None

model = [m.cuda() for m in model]
for m in model:
m.eval()
Expand Down
6 changes: 3 additions & 3 deletions examples/models/vlm/glm_45v/inference.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@
WORKSPACE=${WORKSPACE:-/workspace}

# GLM-4.5V is a large MoE model (106B parameters)
# Using TP=1, PP=4, EP=2 for inference (8 GPUs minimum)
# Using TP=1, PP=2, EP=4 for inference (8 GPUs minimum)

# Inference with Hugging Face checkpoints
uv run python -m torch.distributed.run --nproc_per_node=8 examples/conversion/hf_to_megatron_generate_vlm.py \
Expand All @@ -26,8 +26,8 @@ uv run python -m torch.distributed.run --nproc_per_node=8 examples/conversion/hf
--prompt "Describe this image." \
--max_new_tokens 50 \
--tp 1 \
--pp 4 \
--ep 2 \
--pp 2 \
--ep 4 \
--trust_remote_code

# Inference with imported Megatron checkpoints
Expand Down
Loading