Skip to content
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

[Model][Bugfix] Support TP for PixtralHF ViT #10405

Merged
merged 1 commit into from
Nov 18, 2024
Merged

[Model][Bugfix] Support TP for PixtralHF ViT #10405

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 6 additions & 2 deletions vllm/model_executor/models/pixtral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@

from vllm.attention import AttentionMetadata
from vllm.config import ModelConfig, VllmConfig
from vllm.distributed import divide, get_tensor_model_parallel_world_size
from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, DummyData,
InputContext, token_inputs)
from vllm.model_executor.layers.activation import get_act_and_mul_fn
Expand Down Expand Up @@ -830,17 +831,20 @@ def __init__(

self.config = config
assert not config.hidden_size % config.num_attention_heads
self.n_heads = config.num_attention_heads
self.total_num_heads = config.num_attention_heads
tp_size = get_tensor_model_parallel_world_size()
self.n_heads = divide(config.num_attention_heads, tp_size)
self.head_dim = config.hidden_size // config.num_attention_heads

self.qkv_proj = QKVParallelLinear(
hidden_size=config.hidden_size,
head_size=self.head_dim,
total_num_heads=self.n_heads,
total_num_heads=self.total_num_heads,
bias=False,
quant_config=quant_config,
prefix=f"{prefix}.qkv_proj",
)
assert self.total_num_heads * self.head_dim == config.hidden_size
self.o_proj = RowParallelLinear(
input_size=config.hidden_size,
output_size=config.hidden_size,
Expand Down