xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 09:25:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI] [Hybrid] Speed up hybrid models test by removing large models (#22563)

Browse Source 
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
This commit is contained in:
Thomas Parnell 2025-08-09 11:04:42 +02:00 committed by GitHub
parent a6022e6fbc
commit 1bf5e1f25b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
tests/models/language/generation/test_hybrid.py
View File

@ -20,7 +20,7 @@ pytestmark = pytest.mark.hybrid_model
SSM_MODELS = [ SSM_MODELS = [
"state-spaces/mamba-130m-hf", "state-spaces/mamba-130m-hf",
"tiiuae/falcon-mamba-tiny-dev", "tiiuae/falcon-mamba-tiny-dev",
"mistralai/Mamba-Codestral-7B-v0.1", "yujiepan/mamba2-codestral-v0.1-tiny-random",
] ]
HYBRID_MODELS = [ HYBRID_MODELS = [
@ -29,8 +29,6 @@ HYBRID_MODELS = [
# "pfnet/plamo-2-1b", # "pfnet/plamo-2-1b",
"Zyphra/Zamba2-1.2B-instruct", "Zyphra/Zamba2-1.2B-instruct",
"hmellor/tiny-random-BambaForCausalLM", "hmellor/tiny-random-BambaForCausalLM",
"ibm-ai-platform/Bamba-9B-v1",
"nvidia/Nemotron-H-8B-Base-8K",
"ibm-granite/granite-4.0-tiny-preview", "ibm-granite/granite-4.0-tiny-preview",
"tiiuae/Falcon-H1-0.5B-Base", "tiiuae/Falcon-H1-0.5B-Base",
] ]
@ -40,23 +38,18 @@ HF_UNSUPPORTED_MODELS = [
# Mamba2 is buggy for Codestral as it doesn't handle n_groups, so the test # Mamba2 is buggy for Codestral as it doesn't handle n_groups, so the test
# doesn't compare vLLM output with HF output. # doesn't compare vLLM output with HF output.
# See https://github.com/huggingface/transformers/pull/35943 # See https://github.com/huggingface/transformers/pull/35943
"mistralai/Mamba-Codestral-7B-v0.1", "yujiepan/mamba2-codestral-v0.1-tiny-random",
# Note: I'm not seeing the same output from vLLM V0 vs. HF transformers # transformers 4.55 is still producing garbage for this model
# for Nemotron-H-8B; currently only compare vLLM V0 vs. vLLM V1 # TODO(tdoublep): follow-up on transformers side
"nvidia/Nemotron-H-8B-Base-8K", "ibm-granite/granite-4.0-tiny-preview"
# NOTE: Currently the test fails due to HF transformers issue fixed in:
# https://github.com/huggingface/transformers/pull/39033
# We will enable vLLM test for Granite after next HF transformers release.
"ibm-granite/granite-4.0-tiny-preview",
] ]
V1_SUPPORTED_MODELS = [ V1_SUPPORTED_MODELS = [
"state-spaces/mamba-130m-hf", "state-spaces/mamba-130m-hf",
"ai21labs/Jamba-tiny-dev", "ai21labs/Jamba-tiny-dev",
"mistralai/Mamba-Codestral-7B-v0.1", "yujiepan/mamba2-codestral-v0.1-tiny-random",
"ibm-ai-platform/Bamba-9B-v1",
"Zyphra/Zamba2-1.2B-instruct", "Zyphra/Zamba2-1.2B-instruct",
"nvidia/Nemotron-H-8B-Base-8K", "hmellor/tiny-random-BambaForCausalLM",
"ibm-granite/granite-4.0-tiny-preview", "ibm-granite/granite-4.0-tiny-preview",
"tiiuae/Falcon-H1-0.5B-Base", "tiiuae/Falcon-H1-0.5B-Base",
] ]