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 15, 02:15 UTC
Support for LlamaForSequenceClassification (#20807)
Signed-off-by: thechaos16 <thechaos16@gmail.com>
Minkyu Kim committed
Jul 13, 2025 at 16:09 UTC
bd4c1e6fdbec56594079764bcb74c7e2a81ce525
3 files changed
+7
-1
tests/models/registry.py
+1
@@ -330,6 +330,7 @@ _CROSS_ENCODER_EXAMPLE_MODELS = {
330
hf_overrides={"architectures": ["GemmaForSequenceClassification"], # noqa: E501
331
"classifier_from_token": ["Yes"], # noqa: E501
332
"method": "no_post_processing"}), # noqa: E501
333
+ "LlamaForSequenceClassification": _HfExamplesInfo("Skywork/Skywork-Reward-V2-Llama-3.2-1B"), # noqa: E501
334
"ModernBertForSequenceClassification": _HfExamplesInfo("Alibaba-NLP/gte-reranker-modernbert-base", v0_only=True), # noqa: E501
335
"RobertaForSequenceClassification": _HfExamplesInfo("cross-encoder/quora-roberta-base", v0_only=True), # noqa: E501
336
"XLMRobertaForSequenceClassification": _HfExamplesInfo("BAAI/bge-reranker-v2-m3", v0_only=True), # noqa: E501
vllm/model_executor/models/llama.py
+4
@@ -49,6 +49,7 @@ from vllm.model_executor.model_loader.weight_utils import (
49
from vllm.model_executor.sampling_metadata import SamplingMetadata
50
from vllm.sequence import IntermediateTensors
51
52
+from .adapters import as_seq_cls_model
53
from .interfaces import SupportsLoRA, SupportsPP
54
from .utils import (AutoWeightsLoader, PPMissingLayer, extract_layer_index,
55
is_pp_missing_parameter,
@@ -645,3 +646,6 @@ class LlamaForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
646
name = name.replace(item, mapping[item])
647
648
return name, loaded_weight
649
+
650
+
651
+LlamaForSequenceClassification = as_seq_cls_model(LlamaForCausalLM)
vllm/model_executor/models/registry.py
+2
-1
@@ -183,7 +183,8 @@ _CROSS_ENCODER_MODELS = {
183
"GemmaForSequenceClassification": ("gemma", "GemmaForSequenceClassification"), # noqa: E501
184
"Qwen2ForSequenceClassification": ("qwen2", "Qwen2ForSequenceClassification"), # noqa: E501
185
"Qwen3ForSequenceClassification": ("qwen3", "Qwen3ForSequenceClassification"), # noqa: E501
186
- "JinaVLForRanking": ("jina_vl", "JinaVLForSequenceClassification"), # noqa: E501
186
+ "LlamaForSequenceClassification": ("llama", "LlamaForSequenceClassification"), # noqa: E501
187
+ "JinaVLForRanking": ("jina_vl", "JinaVLForSequenceClassification"), # noqa: E501,
188
}
189
190
_MULTIMODAL_MODELS = {