xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-07 10:09:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI/Build] Cleanup VLM tests (#6107)

Browse Source
This commit is contained in:
Cyrus Leung 2024-07-04 09:58:18 +08:00 committed by GitHub
parent 0ed646b7aa
commit 3dd507083f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 5 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
tests/models/test_llava_next.py
View File

@ -1,4 +1,3 @@
import re
from typing import List, Optional, Tuple from typing import List, Optional, Tuple
import pytest import pytest
@ -36,7 +35,6 @@ def vllm_to_hf_output(vllm_output: Tuple[List[int], str,
output_ids, output_str, out_logprobs = vllm_output output_ids, output_str, out_logprobs = vllm_output
tokenizer = AutoTokenizer.from_pretrained(model) tokenizer = AutoTokenizer.from_pretrained(model)
image_token_str = tokenizer.decode(IMAGE_TOKEN_ID)
eos_token_id = tokenizer.eos_token_id eos_token_id = tokenizer.eos_token_id
hf_output_ids = [ hf_output_ids = [
@ -44,9 +42,8 @@ def vllm_to_hf_output(vllm_output: Tuple[List[int], str,
if token_id != IMAGE_TOKEN_ID or output_ids[idx - 1] != IMAGE_TOKEN_ID if token_id != IMAGE_TOKEN_ID or output_ids[idx - 1] != IMAGE_TOKEN_ID
] ]
hf_output_str = re.sub(fr"({image_token_str})+", "", output_str) assert output_str[0] == " "
assert hf_output_str[0] == " " hf_output_str = output_str[1:]
hf_output_str = hf_output_str[1:]
if hf_output_ids[-1] == eos_token_id: if hf_output_ids[-1] == eos_token_id:
hf_output_str = hf_output_str + tokenizer.decode(eos_token_id) hf_output_str = hf_output_str + tokenizer.decode(eos_token_id)

3
tests/models/test_phi3v.py
View File

@ -35,8 +35,7 @@ def vllm_to_hf_output(vllm_output: Tuple[List[int], str,
assert output_str_without_image[0] == " " assert output_str_without_image[0] == " "
output_str_without_image = output_str_without_image[1:] output_str_without_image = output_str_without_image[1:]
hf_output_str = output_str_without_image.replace("<|user|>", "") \ hf_output_str = output_str_without_image + "<|end|><|endoftext|>"
.replace("<|end|>\n<|assistant|>", " ")
tokenizer = AutoTokenizer.from_pretrained(model) tokenizer = AutoTokenizer.from_pretrained(model)
hf_output_ids = tokenizer.encode(output_str_without_image) hf_output_ids = tokenizer.encode(output_str_without_image)

1
tests/models/utils.py
View File

@ -77,6 +77,7 @@ def check_logprobs_close(
# Each predicted token must be in top N logprobs of the other # Each predicted token must be in top N logprobs of the other
fail_msg = ( fail_msg = (
f"Test{prompt_idx}:" f"Test{prompt_idx}:"
f"\nMatched tokens:\t{output_ids_0[:idx]}"
f"\n{name_0}:\t{output_str_0!r}\t{logprobs_elem_0}" f"\n{name_0}:\t{output_str_0!r}\t{logprobs_elem_0}"
f"\n{name_1}:\t{output_str_1!r}\t{logprobs_elem_1}") f"\n{name_1}:\t{output_str_1!r}\t{logprobs_elem_1}")

2
vllm/multimodal/image.py
View File

@ -115,7 +115,7 @@ class ImagePlugin(MultiModalPlugin):
if isinstance(data, Image.Image): if isinstance(data, Image.Image):
image_processor = self._get_hf_image_processor(model_config) image_processor = self._get_hf_image_processor(model_config)
if image_processor is None: if image_processor is None:
raise RuntimeError("No HuggingFace processor is available" raise RuntimeError("No HuggingFace processor is available "
"to process the image object") "to process the image object")
try: try:
batch_data = image_processor \ batch_data = image_processor \