Weight sync refactor + move sparse nccl engine (#44353)
Signed-off-by: hao-aaron <ahao@anyscale.com> Signed-off-by: haoaaron <ahao@anyscale.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Sonnet 4.6
parent
f651a8a9a4
commit
77a9c5ae28
@@ -7,7 +7,6 @@ from unittest.mock import Mock
|
||||
import numpy as np
|
||||
import pytest
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
|
||||
import vllm.v1.worker.gpu_model_runner as gpu_model_runner_module
|
||||
from vllm.config import (
|
||||
@@ -23,7 +22,6 @@ from vllm.distributed.parallel_state import (
|
||||
init_distributed_environment,
|
||||
initialize_model_parallel,
|
||||
)
|
||||
from vllm.distributed.weight_transfer.base import SparseWeightPatch
|
||||
from vllm.lora.layers import LoRAMappingType
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.model_executor.layers.attention import Attention
|
||||
@@ -864,73 +862,6 @@ def test_sample_passes_reordered_draft_probs_to_rejection_sampler():
|
||||
assert torch.equal(passed_draft_probs, expected_draft_probs)
|
||||
|
||||
|
||||
def test_apply_sparse_weight_patches_updates_only_selected_entries():
|
||||
class DummyModel(nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.weight = nn.Parameter(torch.zeros(6, dtype=torch.float32))
|
||||
|
||||
runner = object.__new__(GPUModelRunner)
|
||||
runner.model = DummyModel()
|
||||
|
||||
runner.apply_sparse_weight_patches(
|
||||
[
|
||||
SparseWeightPatch(
|
||||
name="weight",
|
||||
indices=torch.tensor([1, 4], dtype=torch.int32),
|
||||
values=torch.tensor([3.5, -2.0], dtype=torch.float32),
|
||||
)
|
||||
]
|
||||
)
|
||||
|
||||
expected = torch.tensor([0.0, 3.5, 0.0, 0.0, -2.0, 0.0], dtype=torch.float32)
|
||||
assert torch.equal(runner.get_model().weight.data, expected)
|
||||
|
||||
|
||||
def test_apply_sparse_weight_patches_rejects_mismatched_lengths():
|
||||
class DummyModel(nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.weight = nn.Parameter(torch.zeros(4, dtype=torch.float32))
|
||||
|
||||
runner = object.__new__(GPUModelRunner)
|
||||
runner.model = DummyModel()
|
||||
|
||||
with pytest.raises(ValueError, match="matching lengths"):
|
||||
runner.apply_sparse_weight_patches(
|
||||
[
|
||||
SparseWeightPatch(
|
||||
name="weight",
|
||||
indices=torch.tensor([1, 2], dtype=torch.int32),
|
||||
values=torch.tensor([1.0], dtype=torch.float32),
|
||||
)
|
||||
]
|
||||
)
|
||||
|
||||
|
||||
def test_apply_sparse_weight_patches_rejects_non_contiguous_param():
|
||||
class DummyModel(nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.weight = nn.Parameter(
|
||||
torch.arange(12, dtype=torch.float32).view(3, 4).t()
|
||||
)
|
||||
|
||||
runner = object.__new__(GPUModelRunner)
|
||||
runner.model = DummyModel()
|
||||
|
||||
with pytest.raises(NotImplementedError, match="contiguous params"):
|
||||
runner.apply_sparse_weight_patches(
|
||||
[
|
||||
SparseWeightPatch(
|
||||
name="weight",
|
||||
indices=torch.tensor([1], dtype=torch.int32),
|
||||
values=torch.tensor([1.0], dtype=torch.float32),
|
||||
)
|
||||
]
|
||||
)
|
||||
|
||||
|
||||
def test_init_kv_cache_with_kv_sharing_invalid_target_layer_order(default_vllm_config):
|
||||
torch.set_default_dtype(torch.float16)
|
||||
layer_0 = "model.layers.0.self_attn.attn"
|
||||
|
||||
Reference in New Issue
Block a user