A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI] [Hybrid] Speed up hybrid models test by removing large models (#22563)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Aug 9, 2025 at 11:04 UTC
1bf5e1f25b92423f5739ea7cbb9266f61af12b0b
1 file changed
+7
-14
tests/models/language/generation/test_hybrid.py
+7
-14
@@ -20,7 +20,7 @@ pytestmark = pytest.mark.hybrid_model
20
SSM_MODELS = [
21
"state-spaces/mamba-130m-hf",
22
"tiiuae/falcon-mamba-tiny-dev",
23
- "mistralai/Mamba-Codestral-7B-v0.1",
23
+ "yujiepan/mamba2-codestral-v0.1-tiny-random",
24
]
25
26
HYBRID_MODELS = [
@@ -29,8 +29,6 @@ HYBRID_MODELS = [
29
# "pfnet/plamo-2-1b",
30
"Zyphra/Zamba2-1.2B-instruct",
31
"hmellor/tiny-random-BambaForCausalLM",
32
- "ibm-ai-platform/Bamba-9B-v1",
33
- "nvidia/Nemotron-H-8B-Base-8K",
32
"ibm-granite/granite-4.0-tiny-preview",
33
"tiiuae/Falcon-H1-0.5B-Base",
34
]
@@ -40,23 +38,18 @@ HF_UNSUPPORTED_MODELS = [
38
# Mamba2 is buggy for Codestral as it doesn't handle n_groups, so the test
39
# doesn't compare vLLM output with HF output.
40
# See https://github.com/huggingface/transformers/pull/35943
43
- "mistralai/Mamba-Codestral-7B-v0.1",
44
- # Note: I'm not seeing the same output from vLLM V0 vs. HF transformers
45
- # for Nemotron-H-8B; currently only compare vLLM V0 vs. vLLM V1
46
- "nvidia/Nemotron-H-8B-Base-8K",
47
- # NOTE: Currently the test fails due to HF transformers issue fixed in:
48
- # https://github.com/huggingface/transformers/pull/39033
49
- # We will enable vLLM test for Granite after next HF transformers release.
50
- "ibm-granite/granite-4.0-tiny-preview",
41
+ "yujiepan/mamba2-codestral-v0.1-tiny-random",
42
+ # transformers 4.55 is still producing garbage for this model
43
+ # TODO(tdoublep): follow-up on transformers side
44
+ "ibm-granite/granite-4.0-tiny-preview"
45
]
46
47
V1_SUPPORTED_MODELS = [
48
"state-spaces/mamba-130m-hf",
49
"ai21labs/Jamba-tiny-dev",
56
- "mistralai/Mamba-Codestral-7B-v0.1",
57
- "ibm-ai-platform/Bamba-9B-v1",
50
+ "yujiepan/mamba2-codestral-v0.1-tiny-random",
51
"Zyphra/Zamba2-1.2B-instruct",
59
- "nvidia/Nemotron-H-8B-Base-8K",
52
+ "hmellor/tiny-random-BambaForCausalLM",
53
"ibm-granite/granite-4.0-tiny-preview",
54
"tiiuae/Falcon-H1-0.5B-Base",
55
]