[Bugfix] Honor OPT token embedding width for prompt embeddings (#61736)

Signed-off-by: maithilijoshi20 <maithilij2003@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
maithilijoshi20
2026-10-16 04:41:50 +00:00
committed by GitHub
co-authored by Cyrus Leung
parent b466fbacd2
commit 92dc3dd393
9 changed files with 60 additions and 9 deletions
@@ -75,6 +75,9 @@ AITER_MODEL_LIST = [
"facebook/opt-125m", # opt
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param(
"facebook/opt-350m", # opt with projected token embeddings
),
pytest.param(
"microsoft/phi-2", # phi
marks=[pytest.mark.core_model, pytest.mark.slow_test],
+2 -1
View File
@@ -462,7 +462,8 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"OlmoHybridForCausalLM": _HfExamplesInfo("allenai/Olmo-Hybrid-7B"),
"OlmoeForCausalLM": _HfExamplesInfo("allenai/OLMoE-1B-7B-0924-Instruct"),
"OPTForCausalLM": _HfExamplesInfo(
"facebook/opt-125m", {"1b": "facebook/opt-iml-max-1.3b"}
"facebook/opt-125m",
{"350m": "facebook/opt-350m", "1b": "facebook/opt-iml-max-1.3b"},
),
"OrionForCausalLM": _HfExamplesInfo(
"OrionStarAI/Orion-14B-Chat", trust_remote_code=True