SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix flaky tests (#34069)

Browse Source 
* fix mllama only

* allow image token index
This commit is contained in:
Raushan Turganbay
2024-10-11 15:41:46 +02:00
committed by GitHub
parent 3a24ba82ad
commit fd70464fa7
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/transformers/models/mllama/modeling_mllama.py
View File

@@ -2214,7 +2214,7 @@ class MllamaForConditionalGeneration(MllamaPreTrainedModel, GenerationMixin):
# If we're in pre-fill or cacheless decoding step, then we need pixel_values and aspect ratios # If we're in pre-fill or cacheless decoding step, then we need pixel_values and aspect ratios
# to compute image hidden states, otherwise they are cached within each cross attn layer # to compute image hidden states, otherwise they are cached within each cross attn layer
if (input_ids == self.config.image_token_index).any(): if cache_position[0] == 0:
model_inputs["pixel_values"] = pixel_values model_inputs["pixel_values"] = pixel_values
model_inputs["aspect_ratio_ids"] = aspect_ratio_ids model_inputs["aspect_ratio_ids"] = aspect_ratio_ids
model_inputs["aspect_ratio_mask"] = aspect_ratio_mask model_inputs["aspect_ratio_mask"] = aspect_ratio_mask

1
utils/check_config_attributes.py
View File

@@ -243,6 +243,7 @@ def check_attribute_being_used(config_class, attributes, default_value, source_s
"pad_index", "pad_index",
"unk_index", "unk_index",
"mask_index", "mask_index",
"image_token_index", # for VLMs
"image_size", "image_size",
"use_cache", "use_cache",
"out_features", "out_features",