[CI] Try introducing isort. (#3495)

This commit is contained in:
SangBin Cho
2024-03-25 07:59:47 -07:00
committed by GitHub
parent e67c295b0c
commit 01bfb22b41
144 changed files with 512 additions and 465 deletions
+1
View File
@@ -1,4 +1,5 @@
import pytest
from vllm.utils import create_kv_caches_with_random
+1 -1
View File
@@ -2,10 +2,10 @@ from typing import Type
import pytest
import torch
from allclose_default import get_default_atol, get_default_rtol
from vllm.model_executor.layers.activation import (FastGELU, GeluAndMul,
NewGELU, SiluAndMul)
from allclose_default import get_default_atol, get_default_rtol
DTYPES = [torch.half, torch.bfloat16, torch.float]
NUM_TOKENS = [7, 83, 2048] # Arbitrary values for testing
+3 -4
View File
@@ -3,13 +3,12 @@ from typing import List, Optional, Tuple
import pytest
import torch
from allclose_default import get_default_atol, get_default_rtol
from xformers import ops as xops
from xformers.ops.fmha.attn_bias import BlockDiagonalCausalMask
from vllm._C import ops, cache_ops
from vllm.utils import get_max_shared_memory_bytes
from vllm.utils import is_hip
from allclose_default import get_default_atol, get_default_rtol
from vllm._C import cache_ops, ops
from vllm.utils import get_max_shared_memory_bytes, is_hip
FLOAT32_BYTES = torch.finfo(torch.float).bits // 8
# This will change depending on the compute capability.
+1 -2
View File
@@ -1,10 +1,9 @@
import random
from typing import Tuple
import pytest
import torch
from typing import Tuple
from vllm._C import cache_ops
COPYING_DIRECTION = [('cuda', 'cpu'), ('cuda', 'cuda'), ('cpu', 'cuda')]
+1 -1
View File
@@ -7,8 +7,8 @@ import torch
from transformers import MixtralConfig
from transformers.models.mixtral.modeling_mixtral import MixtralSparseMoeBlock
from vllm.model_executor.layers.fused_moe import fused_moe
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.fused_moe import fused_moe
from vllm.model_executor.models.mixtral import MixtralMoE
+2 -1
View File
@@ -1,9 +1,10 @@
from itertools import accumulate
from typing import List, Optional
import pytest
import torch
from allclose_default import get_default_atol, get_default_rtol
from itertools import accumulate
from vllm.model_executor.layers.rotary_embedding import get_rope
IS_NEOX_STYLE = [True, False]
+3 -2
View File
@@ -1,12 +1,13 @@
import random
import pytest
import time
import pytest
import torch
from vllm.attention.ops.prefix_prefill import context_attention_fwd
from xformers import ops as xops
from xformers.ops.fmha.attn_bias import BlockDiagonalCausalFromBottomRightMask
from vllm.attention.ops.prefix_prefill import context_attention_fwd
NUM_HEADS = [64]
NUM_QUERIES_PER_KV = [1, 8, 64]
HEAD_SIZES = [128]
+3 -2
View File
@@ -1,7 +1,8 @@
import torch
import pytest
import random
import pytest
import torch
from vllm.model_executor.layers.ops.rand import seeded_uniform
from vllm.model_executor.utils import set_random_seed
+4 -4
View File
@@ -1,15 +1,15 @@
import gc
import torch
import pytest
import torch
import triton
import triton.language as tl
from vllm.model_executor.layers.ops.sample import (
_uniform_to_exponential, sample, get_num_triton_sampler_splits,
MAX_TRITON_N_COLS)
from vllm.model_executor.utils import set_random_seed
MAX_TRITON_N_COLS, _uniform_to_exponential, get_num_triton_sampler_splits,
sample)
from vllm.model_executor.sampling_metadata import SamplingTensors
from vllm.model_executor.utils import set_random_seed
SINGLE_SPLIT_VOCAB_SIZE = 32000 # llama/mistral/mixtral vocab size
MULTI_SPLIT_VOCAB_SIZE = MAX_TRITON_N_COLS + 100