Skip to content
Merged
Changes from 1 commit
Commits
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
7 changes: 5 additions & 2 deletions vllm/models/minimax_m3/nvidia/model.py
Original file line number Diff line number Diff line change
Expand Up @@ -1395,8 +1395,11 @@ def forward_input_norm(
)
else:
# MXFP8 output when qkv_proj consumes it (checked per call: LoRA
# wrappers installed after construction change the key).
mxfp8 = _consumes_mxfp8_128x4(self.self_attn.qkv_proj)
# wrappers installed after construction change the key), but not
# under sequence parallelism, which all-gathers the norm output.
mxfp8 = not self.use_sequence_parallel and _consumes_mxfp8_128x4(
self.self_attn.qkv_proj
)
if residual is None:
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states, mxfp8=mxfp8)
Expand Down
Loading