Skip to content

Commit f2bd246

Browse files
authored
[VLM] Fix paligemma, fuyu and persimmon with transformers 4.45 : use config.text_config.vocab_size (#8707)
1 parent a79e522 commit f2bd246

File tree

3 files changed

+9
-8
lines changed

3 files changed

+9
-8
lines changed

vllm/model_executor/models/fuyu.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -229,7 +229,7 @@ def __init__(self,
229229
self.multimodal_config = multimodal_config
230230

231231
self.padding_idx = config.pad_token_id
232-
self.vocab_size = config.vocab_size
232+
self.vocab_size = config.text_config.vocab_size
233233
self.image_token_id = _IMAGE_TOKEN_ID
234234
self.image_feature_size = config.patch_size**2 * config.num_channels
235235

vllm/model_executor/models/paligemma.py

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -152,7 +152,8 @@ def __init__(self,
152152
self.unpadded_vocab_size = config.text_config.vocab_size
153153
logit_scale = getattr(config, "logit_scale", 1.0)
154154
self.logits_processor = LogitsProcessor(self.unpadded_vocab_size,
155-
config.vocab_size, logit_scale)
155+
config.text_config.vocab_size,
156+
logit_scale)
156157
self.sampler = Sampler()
157158

158159
def _validate_pixel_values(self, data: torch.Tensor) -> torch.Tensor:

vllm/model_executor/models/persimmon.py

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -213,10 +213,10 @@ def __init__(self,
213213
cache_config: Optional[CacheConfig] = None,
214214
quant_config: Optional[QuantizationConfig] = None):
215215
super().__init__()
216-
self.vocab_size = config.vocab_size
216+
self.vocab_size = config.text_config.vocab_size
217217

218-
self.embed_tokens = VocabParallelEmbedding(config.vocab_size,
219-
config.hidden_size)
218+
self.embed_tokens = VocabParallelEmbedding(
219+
config.text_config.vocab_size, config.hidden_size)
220220
self.layers = nn.ModuleList([
221221
PersimmonDecoderLayer(config,
222222
cache_config=cache_config,
@@ -257,14 +257,14 @@ def __init__(self,
257257
quant_config: Optional[QuantizationConfig] = None):
258258
super().__init__()
259259
self.config = config
260-
self.vocab_size = config.vocab_size
260+
self.vocab_size = config.text_config.vocab_size
261261
self.model = PersimmonModel(config,
262262
cache_config=cache_config,
263263
quant_config=quant_config)
264-
self.lm_head = ParallelLMHead(config.vocab_size,
264+
self.lm_head = ParallelLMHead(config.text_config.vocab_size,
265265
config.hidden_size,
266266
bias=False)
267-
self.logits_processor = LogitsProcessor(config.vocab_size)
267+
self.logits_processor = LogitsProcessor(config.text_config.vocab_size)
268268
self.sampler = Sampler()
269269

270270
def forward(

0 commit comments

Comments
 (0)