fix some typos (#24071)
Signed-off-by: co63oc <co63oc@users.noreply.github.com>
This commit is contained in:
@@ -134,7 +134,7 @@ class SimpleModelWithTwoGraphs(ParentModel):
|
||||
# Test will fail without set_model_tag here with error:
|
||||
# "ValueError: too many values to unpack (expected 3)"
|
||||
# This is because CompiledAttention and CompiledAttentionTwo
|
||||
# have different implmentations but the same torch.compile
|
||||
# have different implementations but the same torch.compile
|
||||
# cache dir will be used as default prefix is 'model_tag'
|
||||
with set_model_tag("attn_one"):
|
||||
self.attn_one = CompiledAttention(
|
||||
|
||||
@@ -224,7 +224,7 @@ def tg_mxfp4_moe(
|
||||
assert (w2_bias.dim() == 2 and w2_bias.shape[0] == num_experts
|
||||
and w2_bias.shape[1] == hidden_size)
|
||||
|
||||
# Swap w1 and w3 as the defenition of
|
||||
# Swap w1 and w3 as the definition of
|
||||
# swiglu is different in the trtllm-gen
|
||||
w13_weight_scale_ = w13_weight_scale.clone()
|
||||
w13_weight_ = w13_weight.clone()
|
||||
|
||||
@@ -52,7 +52,7 @@ def test_profiling(model_id: str, max_model_len: int):
|
||||
chunks_per_image = prod(mm_data["patches_per_image"])
|
||||
total_num_patches = chunks_per_image * tokens_per_patch
|
||||
num_tiles = mm_data["aspect_ratios"][0][0] * mm_data["aspect_ratios"][0][
|
||||
1] # x-y seperator tokens
|
||||
1] # x-y separator tokens
|
||||
total_tokens = total_num_patches.item() + num_tiles.item(
|
||||
) + 3 # image start, image, image end
|
||||
|
||||
|
||||
@@ -27,7 +27,7 @@ def use_v0_only(monkeypatch):
|
||||
reason="ModelOpt FP8 is not supported on this GPU type.")
|
||||
def test_modelopt_fp8_checkpoint_setup(vllm_runner):
|
||||
"""Test ModelOpt FP8 checkpoint loading and structure validation."""
|
||||
# TODO: provide a small publically available test checkpoint
|
||||
# TODO: provide a small publicly available test checkpoint
|
||||
model_path = ("/home/scratch.omniml_data_1/zhiyu/ckpts/test_ckpts/"
|
||||
"TinyLlama-1.1B-Chat-v1.0-fp8-0710")
|
||||
|
||||
|
||||
@@ -82,7 +82,7 @@ def test_beam_search_with_concurrency_limit(
|
||||
beam_width: int,
|
||||
) -> None:
|
||||
# example_prompts[1]&[3]&[7] fails due to unknown reason even without
|
||||
# concurency limit. skip them for now.
|
||||
# concurrency limit. skip them for now.
|
||||
example_prompts = (example_prompts[:8])
|
||||
concurrency_limit = 2
|
||||
assert len(example_prompts) > concurrency_limit
|
||||
|
||||
@@ -160,7 +160,7 @@ def test_local_attention_virtual_batches(test_data: LocalAttentionTestData):
|
||||
# Use torch.arange instead of torch.randint so we can assert on
|
||||
# block table tensor values. The block table will have shape
|
||||
# (num_batches, cdiv(max_seq_len, block_size)) and the values will be
|
||||
# aranged from 0 to cdiv(max_seq_len, block_size)-1
|
||||
# arranged from 0 to cdiv(max_seq_len, block_size)-1
|
||||
arange_block_indices=True,
|
||||
)
|
||||
|
||||
|
||||
@@ -33,7 +33,7 @@ def _check_path_len(path):
|
||||
|
||||
|
||||
def _list_path(path):
|
||||
"""Return the list of foldername (hashes generatd) under the path"""
|
||||
"""Return the list of foldername (hashes generated) under the path"""
|
||||
return list(path.iterdir())
|
||||
|
||||
|
||||
@@ -41,7 +41,7 @@ def run_test(tmp_path, processor, llm: LLM, question: str,
|
||||
image_urls: list[Image], expected_len: int, info: str):
|
||||
"""
|
||||
One individual test to process the prompt and output base on 1 set of input
|
||||
Then check if the length in the strorage path matches the expected length
|
||||
Then check if the length in the storage path matches the expected length
|
||||
`info` introduces details or purpose of the individual test
|
||||
"""
|
||||
print(f"***info: {info}***")
|
||||
@@ -115,7 +115,7 @@ def test_shared_storage_connector_hashes(tmp_path):
|
||||
"""
|
||||
Tests that SharedStorageConnector saves KV to the storage locations
|
||||
with proper hashes; that are unique for inputs with identical text but
|
||||
differnt images (same size), or same multiple images but different orders.
|
||||
different images (same size), or same multiple images but different orders.
|
||||
"""
|
||||
# Using tmp_path as the storage path to store KV
|
||||
print(f"KV storage path at: {str(tmp_path)}")
|
||||
@@ -171,12 +171,12 @@ def test_shared_storage_connector_hashes(tmp_path):
|
||||
img=[image_1],
|
||||
expected_len=2,
|
||||
info=("image_1 single input the 2nd time. "
|
||||
"It should not form aother new hash.")),
|
||||
"It should not form another new hash.")),
|
||||
InputCase(text=TEXT_PROMPTS[0],
|
||||
img=[image_2],
|
||||
expected_len=2,
|
||||
info=("image_2 single input the 2nd time. "
|
||||
"It should not form aother new hash.")),
|
||||
"It should not form another new hash.")),
|
||||
InputCase(text=TEXT_PROMPTS[0],
|
||||
img=[image_1, image_2],
|
||||
expected_len=3,
|
||||
@@ -189,12 +189,12 @@ def test_shared_storage_connector_hashes(tmp_path):
|
||||
img=[image_1, image_2],
|
||||
expected_len=4,
|
||||
info=("[image_1, image_2] input the 2nd time. "
|
||||
"It should not form aother new hash.")),
|
||||
"It should not form another new hash.")),
|
||||
InputCase(text=TEXT_PROMPTS[0],
|
||||
img=[image_2, image_1],
|
||||
expected_len=4,
|
||||
info=("[image_2, image_1] input the 2nd time. "
|
||||
"It should not form aother new hash.")),
|
||||
"It should not form another new hash.")),
|
||||
InputCase(text=TEXT_PROMPTS[0],
|
||||
img=[],
|
||||
expected_len=5,
|
||||
|
||||
@@ -81,7 +81,7 @@ def _run_test(kwargs: dict, logitproc_loaded: bool) -> None:
|
||||
target_token = params.extra_args[DUMMY_LOGITPROC_ARG]
|
||||
if not all(x == target_token for x in lp_toks):
|
||||
raise AssertionError(
|
||||
f"Request {bdx} generated {lp_toks}, shoud all be "
|
||||
f"Request {bdx} generated {lp_toks}, should all be "
|
||||
f"{target_token}")
|
||||
else:
|
||||
# This request does not exercise custom logitproc (or custom
|
||||
|
||||
Reference in New Issue
Block a user