20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[VLM] Fix paligemma, fuyu and persimmon with transformers 4.45 : use config.text_config.vocab_size (#8707)

Browse Source
This commit is contained in:
Jani Monoses 2024-09-23 17:43:09 +03:00 committed by GitHub
parent a79e522984
commit f2bd246c17
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 9 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/fuyu.py
View File

@ -229,7 +229,7 @@ class FuyuForCausalLM(nn.Module, SupportsMultiModal):
self.multimodal_config = multimodal_config
self.padding_idx = config.pad_token_id
self.vocab_size = config.vocab_size
self.vocab_size = config.text_config.vocab_size
self.image_token_id = _IMAGE_TOKEN_ID
self.image_feature_size = config.patch_size**2 * config.num_channels

3
vllm/model_executor/models/paligemma.py
View File

@ -152,7 +152,8 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsMultiModal):
self.unpadded_vocab_size = config.text_config.vocab_size
logit_scale = getattr(config, "logit_scale", 1.0)
self.logits_processor = LogitsProcessor(self.unpadded_vocab_size,
config.vocab_size, logit_scale)
config.text_config.vocab_size,
logit_scale)
self.sampler = Sampler()
def _validate_pixel_values(self, data: torch.Tensor) -> torch.Tensor:

12
vllm/model_executor/models/persimmon.py
View File

@ -213,10 +213,10 @@ class PersimmonModel(nn.Module):
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None):
super().__init__()
self.vocab_size = config.vocab_size
self.vocab_size = config.text_config.vocab_size
self.embed_tokens = VocabParallelEmbedding(config.vocab_size,
config.hidden_size)
self.embed_tokens = VocabParallelEmbedding(
config.text_config.vocab_size, config.hidden_size)
self.layers = nn.ModuleList([
PersimmonDecoderLayer(config,
cache_config=cache_config,
@ -257,14 +257,14 @@ class PersimmonForCausalLM(nn.Module):
quant_config: Optional[QuantizationConfig] = None):
super().__init__()
self.config = config
self.vocab_size = config.vocab_size
self.vocab_size = config.text_config.vocab_size
self.model = PersimmonModel(config,
cache_config=cache_config,
quant_config=quant_config)
self.lm_head = ParallelLMHead(config.vocab_size,
self.lm_head = ParallelLMHead(config.text_config.vocab_size,
config.hidden_size,
bias=False)
self.logits_processor = LogitsProcessor(config.vocab_size)
self.logits_processor = LogitsProcessor(config.text_config.vocab_size)
self.sampler = Sampler()
def forward(