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

[VLM] Move supported limits and max tokens to merged multi-modal processor (#11669)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Cyrus Leung
2025-01-01 23:44:42 +08:00
committed by GitHub
parent 73001445fb
commit a115ac46b5
16 changed files with 351 additions and 361 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
tests/multimodal/test_processing.py
View File

@@ -538,6 +538,11 @@ def _test_processing_cache_correctness(
else:
hf_overrides = {}
limit_mm_per_prompt = {
modality: 3 if supports_multi else 1
for modality, supports_multi in modalities.items()
}
model_config = ModelConfig(
model_id,
task="auto",
@@ -548,6 +553,7 @@ def _test_processing_cache_correctness(
dtype="float16",
revision=None,
hf_overrides=hf_overrides,
limit_mm_per_prompt=limit_mm_per_prompt,
)
model_cls = MULTIMODAL_REGISTRY._get_model_cls(model_config)
@@ -580,18 +586,14 @@ def _test_processing_cache_correctness(
min_wh=128,
max_wh=256),
"audio":
partial(_rand_audio, rng, min_len=256, max_len=512, sr=16000),
}
input_max_count = {
modality: 3 if supports_multi else 1
for modality, supports_multi in modalities.items()
partial(_rand_audio, rng, min_len=512, max_len=1024, sr=16000),
}
for batch_idx in range(num_batches):
mm_data = {
k:
[(input_to_hit[k] if rng.rand() < hit_rate else input_factory[k]())
for _ in range(rng.randint(input_max_count[k]))]
for _ in range(rng.randint(limit_mm_per_prompt[k]))]
for k in modalities
}