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

[Model] Refactor Ultravox to use merged input processor (#11198)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
Isotr0py
2024-12-16 18:09:53 +08:00
committed by GitHub
parent bddbbcb132
commit d927dbcd88
7 changed files with 121 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/decoder_only/audio_language/test_ultravox.py
View File

@@ -16,7 +16,7 @@ MODEL_NAME = "fixie-ai/ultravox-v0_3"
AudioTuple = Tuple[np.ndarray, int]
VLLM_PLACEHOLDER = "<|reserved_special_token_0|>"
VLLM_PLACEHOLDER = "<|audio|>"
HF_PLACEHOLDER = "<|audio|>"
CHUNKED_PREFILL_KWARGS = {
@@ -46,7 +46,8 @@ def audio(request):
def server(request, audio_assets):
args = [
"--dtype=bfloat16", "--max-model-len=4096", "--enforce-eager",
f"--limit-mm-per-prompt=audio={len(audio_assets)}"
f"--limit-mm-per-prompt=audio={len(audio_assets)}",
"--trust-remote-code"
] + [
f"--{key.replace('_','-')}={value}"
for key, value in request.param.items()