Skip to content

[RL] Fix GPT-OSS 20B dimension mismatch error in vLLM adapter by resolving intermediate_size fallback #4175

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

Open
susanbao wants to merge 2 commits into
base: main
Choose a base branch
from
+6 −1
Open

[RL] Fix GPT-OSS 20B dimension mismatch error in vLLM adapter by resolving intermediate_size fallback #4175

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
9be8f96
[RL] Fix GPT-OSS 20B dimension mismatch error in vLLM adapter by reso…
susanbao Jun 15, 2026
3c07d86
Map sigmoid activation to swigluoai for fused MoE in vLLM TPU path
susanbao Jun 17, 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
5 changes: 4 additions & 1 deletion src/maxtext/integration/vllm/maxtext_vllm_adapter/adapter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,7 +112,10 @@ def generate_maxtext_config(vllm_config: VllmConfig) -> pyconfig.HyperParameters
if hasattr(vllm_config.model_config.hf_config, "text_config")
else vllm_config.model_config.hf_config
)
hidden_size = getattr(hf_config, "moe_intermediate_size", None)
hidden_size = (
getattr(hf_config, "moe_intermediate_size", None)
or getattr(hf_config, "intermediate_size", None)
)
num_lanes = pltpu.get_tpu_info().num_lanes
num_kv_heads = hf_config.num_key_value_heads

Expand Down
2 changes: 2 additions & 0 deletions src/maxtext/layers/moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2448,6 +2448,8 @@ def fused_moe_matmul(

# Map MaxText config fields to fused_moe_func args
activation = self.config.mlp_activations[0] # e.g. "silu"
if activation == "sigmoid":
activation = "swigluoai"
scoring_fn = self.config.routed_score_func if self.config.routed_score_func else "softmax"

# Check if the model architecture intrinsically renormalizes weights
Expand Down
Loading