Skip to content

Cherry-picks to enable Llama4 Maverick #882

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
wpyszka merged 4 commits into from
Jan 28, 2026
Merged

Cherry-picks to enable Llama4 Maverick #882

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4f7f615
Interleaved sliding window fix (#805)
rsmyrek Jan 21, 2026
d4ef895
[GAUDISW-245665] fix diverge from vllm in multiModalBudget (#837)
linoybu Jan 21, 2026
511fc13
Fix Llama4 shape mismatch for 32k+ context window (#842) (#855)
afierka-intel Jan 26, 2026
70d8e72
Flatten positions only when QK norm is enabled
rsmyrek Jan 22, 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 vllm_gaudi/ops/hpu_fused_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,7 +160,10 @@ def forward_oot(
permuted_weights=True,
activation=layer.activation,
)
return output.view(*(output.size(0), *input_shape[1:]))
if layer.dp_size > 1:
return output.view(*(output.size(0), *input_shape[1:]))
else:
return output.view(*input_shape)


def reduce_output(self, states: torch.Tensor) -> torch.Tensor:
Expand Down
10 changes: 5 additions & 5 deletions vllm_gaudi/v1/worker/hpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,7 +370,8 @@ def patch_llama4_get_attn_scale(model):
orig = attn._get_attn_scale

def _get_attn_scale_for_hpu(self, positions, _orig=orig):
positions = positions.flatten()
if self.qk_norm is not None:
positions = positions.flatten()
return _orig(positions)

attn._get_attn_scale = types.MethodType(_get_attn_scale_for_hpu, attn)
Expand Down Expand Up @@ -4620,8 +4621,7 @@ def warmup_multimodal_graphs(self, buckets):
phase = 'Graph/Multimodal'
from vllm.v1.worker.utils import MultiModalBudget
self.mm_budget = MultiModalBudget(
self.model_config,
self.scheduler_config,
self.vllm_config,
self.mm_registry,
) if self.supports_mm_inputs else None

Expand Down Expand Up @@ -5668,12 +5668,12 @@ def process_metadata(self, attn_metadata: HPUAttentionMetadataV1, batch_size: in
"""
if attn_metadata.is_prompt:
attn_metadata = self._set_attn_bias(attn_metadata, batch_size, seq_len, device, dtype)
if self.interleaved_sliding_window:
if self.interleaved_sliding_window and self.sliding_window is not None:
attn_metadata = self._set_attn_bias_for_sliding_window(attn_metadata, batch_size, seq_len,
self.sliding_window, device, dtype)
else:
attn_metadata = self._set_block_mapping(attn_metadata, batch_size, device, dtype)
if self.interleaved_sliding_window:
if self.interleaved_sliding_window and self.sliding_window is not None:
attn_metadata = self._set_block_mapping(attn_metadata, batch_size, device, dtype, True)
return attn_metadata

Expand Down
Loading