biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI] change spell checker from codespell to typos (#18711)

Browse Source 
Signed-off-by: Andy Xie <andy.xning@gmail.com>
This commit is contained in:
Ning Xie
2025-06-12 10:57:10 +08:00
committed by GitHub
parent 42f52cc95b
commit 2f1c19b245
57 changed files with 335 additions and 163 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/kernels/attention/test_cache.py
View File

@@ -72,8 +72,8 @@ def test_copy_blocks(
# destination blocks.
assert 2 * num_mappings <= num_blocks
src_blocks = random.sample(range(num_blocks), num_mappings)
remainig_blocks = list(set(range(num_blocks)) - set(src_blocks))
dst_blocks = random.sample(remainig_blocks, 2 * num_mappings)
remaining_blocks = list(set(range(num_blocks)) - set(src_blocks))
dst_blocks = random.sample(remaining_blocks, 2 * num_mappings)
block_mapping: list[tuple[int, int]] = []
for i in range(num_mappings):
src = src_blocks[i]
@@ -189,12 +189,12 @@ def test_reshape_and_cache(
# Run the reference implementation.
reshaped_key = key.reshape(num_tokens, *key_cache[0, :, :, 0, :].shape)
block_indicies = torch.div(slot_mapping, block_size, rounding_mode="floor")
block_indicies_lst = block_indicies.cpu().tolist()
block_indices = torch.div(slot_mapping, block_size, rounding_mode="floor")
block_indices_lst = block_indices.cpu().tolist()
block_offsets = slot_mapping % block_size
block_offsets_lst = block_offsets.cpu().tolist()
for i in range(num_tokens):
block_idx = block_indicies_lst[i]
block_idx = block_indices_lst[i]
block_offset = block_offsets_lst[i]
cloned_key_cache[block_idx, :, :, block_offset, :] = reshaped_key[i]
cloned_value_cache[block_idx, :, :, block_offset] = value[i]
@@ -322,12 +322,12 @@ def test_reshape_and_cache_flash(
kv_dtype=kv_cache_dtype)
# Run the reference implementation.
block_indicies = torch.div(slot_mapping, block_size, rounding_mode="floor")
block_indicies_lst = block_indicies.cpu().tolist()
block_indices = torch.div(slot_mapping, block_size, rounding_mode="floor")
block_indices_lst = block_indices.cpu().tolist()
block_offsets = slot_mapping % block_size
block_offsets_lst = block_offsets.cpu().tolist()
for i in range(num_tokens):
block_idx = block_indicies_lst[i]
block_idx = block_indices_lst[i]
block_offset = block_offsets_lst[i]
if kv_cache_layout == "NHD":
cloned_key_cache[block_idx, block_offset, :, :] = key[i]

2
tests/kernels/attention/test_encoder_decoder_attn.py
View File

@@ -46,7 +46,7 @@ CUDA_DEVICE = "cuda:0"
MAX_DEC_SEQ_LENS = [128]
MAX_ENC_SEQ_LENS = [128]
# Narrow teest-cases for unsupported-scenario
# Narrow test-cases for unsupported-scenario
# tests
HEAD_SIZES_FOR_UNSUPP = [HEAD_SIZES[0]]

8
tests/kernels/core/test_rotary_embedding.py
View File

@@ -39,10 +39,10 @@ def rotary_embedding_opcheck(rot,
@pytest.mark.parametrize("head_size", [32, 108])
@pytest.mark.parametrize("seq_len", [11, 1024])
@pytest.mark.parametrize("use_key", [True, False])
@pytest.mark.parametrize("head_stride_is_contingous", [True, False])
@pytest.mark.parametrize("head_stride_is_contiguous", [True, False])
def test_rotary_embedding_opcheck(dist_init, device, max_position,
is_neox_style, rotary_dim, head_size,
seq_len, use_key, head_stride_is_contingous):
seq_len, use_key, head_stride_is_contiguous):
batch_size = 1
base = 10000
num_heads = 7
@@ -52,7 +52,7 @@ def test_rotary_embedding_opcheck(dist_init, device, max_position,
positions = torch.randint(0,
max_position, (batch_size, seq_len),
device=device)
head_stride = head_size + (64 if head_stride_is_contingous else 0)
head_stride = head_size + (64 if head_stride_is_contiguous else 0)
query = torch.randn(batch_size,
seq_len,
@@ -72,7 +72,7 @@ def test_rotary_embedding_opcheck(dist_init, device, max_position,
# if we have a contiguous head stride, test the alternate
# [..., num_heads * head_dim] shape/layout
if head_stride_is_contingous:
if head_stride_is_contiguous:
rotary_embedding_opcheck(
rot, positions, query.flatten(start_dim=-2),
key.flatten(start_dim=-2) if use_key else None)

27
tests/kernels/mamba/test_mamba_ssm_ssd.py
View File

@@ -107,15 +107,15 @@ def generate_random_inputs(batch_size,
return A, dt, X, B, C
def generate_continous_batched_examples(example_lens_by_batch,
num_examples,
full_length,
last_taken,
exhausted,
n_heads,
d_head,
itype,
device='cuda'):
def generate_continuous_batched_examples(example_lens_by_batch,
num_examples,
full_length,
last_taken,
exhausted,
n_heads,
d_head,
itype,
device='cuda'):
# this function generates a random examples of certain length
# and then cut according to "example_lens_by_batch" and feed
@@ -269,11 +269,10 @@ def test_mamba_chunk_scan_cont_batch(d_head, n_heads, seq_len_chunk_size_cases,
exhausted: dict = {} # map: eg -> boolean indicating example is exhausted
states = None
for Y_min, cu_seqlens, seq_idx, (A, dt, X, B,
C) in generate_continous_batched_examples(
cases, num_examples, seqlen,
last_taken, exhausted, n_heads,
d_head, itype):
for Y_min, cu_seqlens, seq_idx, (
A, dt, X, B, C) in generate_continuous_batched_examples(
cases, num_examples, seqlen, last_taken, exhausted, n_heads,
d_head, itype):
chunk_indices, chunk_offsets = \
_query_start_loc_to_chunk_indices_offsets(