[Model] Systematic support for fp32 head, pooling models part (#23810)
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
@@ -306,7 +306,9 @@ class ModernBertForSequenceClassification(nn.Module, SupportsCrossEncoding):
|
||||
self.config = config
|
||||
self.model = ModernBertModel(vllm_config=vllm_config,
|
||||
prefix=maybe_prefix(prefix, "modernbert"))
|
||||
self.classifier = nn.Linear(config.hidden_size, config.num_labels)
|
||||
self.classifier = nn.Linear(config.hidden_size,
|
||||
config.num_labels,
|
||||
dtype=vllm_config.model_config.head_dtype)
|
||||
self.pooling = ModernBertPooler(config)
|
||||
|
||||
pooler_config = vllm_config.model_config.pooler_config
|
||||
|
||||
Reference in New Issue
Block a user