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

[Misc] Update transformers version limits of multi-modal tests (#16381)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-04-10 14:03:33 +08:00
committed by GitHub
parent a9bd832fc5
commit 3d4c87758e
4 changed files with 20 additions and 270 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
tests/models/decoder_only/vision_language/test_models.py
View File

@@ -425,23 +425,20 @@ VLM_TEST_SETTINGS = {
max_num_seqs=2,
patch_hf_runner=model_utils.molmo_patch_hf_runner,
),
# Tests for phi3v currently live in another file because of a bug in
# transformers. Once this issue is fixed, we can enable them here instead.
# https://github.com/huggingface/transformers/issues/34307
# "phi3v": VLMTestInfo(
# models=["microsoft/Phi-3.5-vision-instruct"],
# test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
# prompt_formatter=lambda img_prompt: f"<|user|>\n{img_prompt}<|end|>\n<|assistant|>\n", # noqa: E501
# img_idx_to_prompt=lambda idx: f"<|image_{idx}|>\n",
# max_model_len=4096,
# max_num_seqs=2,
# task="generate",
# # use eager mode for hf runner since phi3v didn't work with flash_attn
# hf_model_kwargs={"_attn_implementation": "eager"},
# use_tokenizer_eos=True,
# vllm_output_post_proc=model_utils.phi3v_vllm_to_hf_output,
# num_logprobs=10,
# ),
"phi3v": VLMTestInfo(
models=["microsoft/Phi-3.5-vision-instruct"],
test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
prompt_formatter=lambda img_prompt: f"<|user|>\n{img_prompt}<|end|>\n<|assistant|>\n", # noqa: E501
img_idx_to_prompt=lambda idx: f"<|image_{idx}|>\n",
max_model_len=4096,
max_num_seqs=2,
task="generate",
# use eager mode for hf runner since phi3v didn't work with flash_attn
hf_model_kwargs={"_attn_implementation": "eager"},
use_tokenizer_eos=True,
vllm_output_post_proc=model_utils.phi3v_vllm_to_hf_output,
num_logprobs=10,
),
"pixtral_hf": VLMTestInfo(
models=["nm-testing/pixtral-12b-FP8-dynamic"],
test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),