[Model Refactoring] Move deepseek_v4_ops to models/deepseek_v4 [3/N] (#43073)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
This commit is contained in:
@@ -13,8 +13,8 @@ from __future__ import annotations
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.models.deepseek_v4.common.ops import fused_q_kv_rmsnorm
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops import fused_q_kv_rmsnorm
|
||||
|
||||
pytestmark = pytest.mark.skipif(
|
||||
not current_platform.is_cuda_alike(),
|
||||
|
||||
@@ -17,11 +17,11 @@ import pytest
|
||||
import torch
|
||||
|
||||
from vllm import _custom_ops as ops
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops import (
|
||||
from vllm.models.deepseek_v4.common.ops import (
|
||||
dequantize_and_gather_k_cache,
|
||||
quantize_and_insert_k_cache,
|
||||
)
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops.fused_compress_quant_cache import (
|
||||
from vllm.models.deepseek_v4.common.ops.fused_compress_quant_cache import (
|
||||
_fused_kv_compress_norm_rope_insert_indexer_attn,
|
||||
_fused_kv_compress_norm_rope_insert_indexer_mxfp4_attn,
|
||||
)
|
||||
|
||||
@@ -19,7 +19,7 @@ The kernel is imported via
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops import (
|
||||
from vllm.models.deepseek_v4.common.ops import (
|
||||
dequantize_and_gather_k_cache,
|
||||
quantize_and_insert_k_cache,
|
||||
)
|
||||
|
||||
@@ -23,10 +23,8 @@ from vllm import _custom_ops as ops
|
||||
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
|
||||
per_token_group_quant_fp8,
|
||||
)
|
||||
from vllm.models.deepseek_v4.common.ops import fused_indexer_q_rope_quant
|
||||
from vllm.utils.import_utils import has_cutedsl
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops.fused_indexer_q import (
|
||||
fused_indexer_q_rope_quant,
|
||||
)
|
||||
|
||||
HEAD_DIM = 128
|
||||
ROPE_DIM = 64
|
||||
@@ -156,7 +154,7 @@ def test_fused_indexer_q_rope_quant_matches_unfused(
|
||||
# by patching the dispatcher's has_cutedsl() binding to return False.
|
||||
cutedsl_patch = (
|
||||
mock.patch(
|
||||
"vllm.v1.attention.ops.deepseek_v4_ops.fused_indexer_q.has_cutedsl",
|
||||
"vllm.models.deepseek_v4.common.ops.fused_indexer_q.has_cutedsl",
|
||||
return_value=False,
|
||||
)
|
||||
if not use_cutedsl
|
||||
|
||||
@@ -21,7 +21,7 @@ Usage:
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.v1.attention.ops.deepseek_v4_ops import fused_inv_rope_fp8_quant
|
||||
from vllm.models.deepseek_v4.common.ops import fused_inv_rope_fp8_quant
|
||||
|
||||
# -- Default dimensions matching DeepSeek V3/V4 --------------------------
|
||||
HEAD_DIM = 512
|
||||
|
||||
Reference in New Issue
Block a user