make format happy

aria/data.py

@@ -22,9 +22,10 @@
 from typing import Dict, Iterable, List
 
 import torch
-from datasets import DatasetDict, concatenate_datasets, load_dataset
 from datasets.features import Features, Sequence, Value
 
+from datasets import DatasetDict, concatenate_datasets, load_dataset
+
 
 def apply_chat_template_and_tokenize(
     messages_batch: List[List[Dict]],

aria/inference.py

@@ -121,10 +121,12 @@ def inference(
             do_sample=True,
             temperature=0.9,
         )
-
-    for i in range(inputs['input_ids'].shape[0]):
-        prompt_len = len(inputs['input_ids'][i])
-        output_text = tokenizer.decode(output[i][prompt_len:], skip_special_tokens=True).replace("<|im_end|>", "")
+
+    for i in range(inputs["input_ids"].shape[0]):
+        prompt_len = len(inputs["input_ids"][i])
+        output_text = tokenizer.decode(
+            output[i][prompt_len:], skip_special_tokens=True
+        ).replace("<|im_end|>", "")
 
     return output_text
 

examples/nextqa/evaluation.py

@@ -1,14 +1,13 @@
 import argparse
 import json
 import os
+import random
 
 import numpy as np
-import random
 import torch
 from peft import PeftConfig, PeftModel
 from torch.utils.data import DataLoader, Dataset
 from tqdm import tqdm
-from transformers import AutoTokenizer
 
 from aria.load_video import load_video
 from aria.lora.layers import GroupedGemmLoraLayer
@@ -64,7 +63,7 @@ def load_model_and_tokenizer(args):
     processor = AriaProcessor.from_pretrained(
         args.base_model_path, tokenizer_path=args.tokenizer_path
     )
-    processor.tokenizer.padding_side="left"
+    processor.tokenizer.padding_side = "left"
     tokenizer = processor.tokenizer
 
     model = AriaForConditionalGeneration.from_pretrained(
@@ -98,8 +97,10 @@ def process_batch(model, tokenizer, inputs, original_batch, prompts):
         )
 
     for i, prompt in enumerate(prompts):
-        prompt_len = len(inputs['input_ids'][i])
-        output_text = tokenizer.decode(output[i][prompt_len:], skip_special_tokens=True).replace("<|im_end|>", "")
+        prompt_len = len(inputs["input_ids"][i])
+        output_text = tokenizer.decode(
+            output[i][prompt_len:], skip_special_tokens=True
+        ).replace("<|im_end|>", "")
         original_batch[i]["pred"] = output_text
 
     return original_batch
@@ -123,7 +124,8 @@ def collate_fn(batch, processor, tokenizer):
         messages.append(item["messages"])
 
     texts = [
-        processor.apply_chat_template(msg, add_generation_prompt=True) for msg in messages
+        processor.apply_chat_template(msg, add_generation_prompt=True)
+        for msg in messages
     ]
     inputs = processor(
         text=texts,

examples/nlvr2/evaluation.py

@@ -7,7 +7,6 @@
 from PIL import Image
 from torch.utils.data import DataLoader, Dataset
 from tqdm import tqdm
-from transformers import AutoTokenizer
 
 from aria.lora.layers import GroupedGemmLoraLayer
 from aria.model import AriaForConditionalGeneration, AriaProcessor, GroupedGEMM
@@ -64,7 +63,7 @@ def load_model_and_tokenizer(args):
     processor = AriaProcessor.from_pretrained(
         args.base_model_path, tokenizer_path=args.tokenizer_path
     )
-    processor.tokenizer.padding_side="left"
+    processor.tokenizer.padding_side = "left"
     tokenizer = processor.tokenizer
 
     model = AriaForConditionalGeneration.from_pretrained(
@@ -98,8 +97,10 @@ def process_batch(model, tokenizer, inputs, original_batch, prompts):
         )
 
     for i, prompt in enumerate(prompts):
-        prompt_len = len(inputs['input_ids'][i])
-        output_text = tokenizer.decode(output[i][prompt_len:], skip_special_tokens=True).replace("<|im_end|>", "")
+        prompt_len = len(inputs["input_ids"][i])
+        output_text = tokenizer.decode(
+            output[i][prompt_len:], skip_special_tokens=True
+        ).replace("<|im_end|>", "")
         original_batch[i]["pred"] = output_text
 
     return original_batch
@@ -115,7 +116,8 @@ def collate_fn(batch, processor, tokenizer):
         messages.append(item["messages"])
 
     texts = [
-        processor.apply_chat_template(msg, add_generation_prompt=True) for msg in messages
+        processor.apply_chat_template(msg, add_generation_prompt=True)
+        for msg in messages
     ]
     inputs = processor(
         text=texts,

examples/refcoco/evaluation.py

@@ -9,7 +9,6 @@
 from torch.utils.data import DataLoader, Dataset
 from torchvision.ops.boxes import box_area
 from tqdm import tqdm
-from transformers import AutoTokenizer
 
 from aria.lora.layers import GroupedGemmLoraLayer
 from aria.model import AriaForConditionalGeneration, AriaProcessor, GroupedGEMM
@@ -66,7 +65,7 @@ def load_model_and_tokenizer(args):
     processor = AriaProcessor.from_pretrained(
         args.base_model_path, tokenizer_path=args.tokenizer_path
     )
-    processor.tokenizer.padding_side="left"
+    processor.tokenizer.padding_side = "left"
     tokenizer = processor.tokenizer
 
     model = AriaForConditionalGeneration.from_pretrained(
@@ -100,8 +99,10 @@ def process_batch(model, tokenizer, inputs, original_batch, prompts):
         )
 
     for i, prompt in enumerate(prompts):
-        prompt_len = len(inputs['input_ids'][i])
-        output_text = tokenizer.decode(output[i][prompt_len:], skip_special_tokens=True).replace("<|im_end|>", "")
+        prompt_len = len(inputs["input_ids"][i])
+        output_text = tokenizer.decode(
+            output[i][prompt_len:], skip_special_tokens=True
+        ).replace("<|im_end|>", "")
         original_batch[i]["pred"] = output_text
 
     return original_batch
@@ -117,7 +118,8 @@ def collate_fn(batch, processor, tokenizer):
         messages.append(item["messages"])
 
     texts = [
-        processor.apply_chat_template(msg, add_generation_prompt=True) for msg in messages
+        processor.apply_chat_template(msg, add_generation_prompt=True)
+        for msg in messages
     ]
     inputs = processor(
         text=texts,

examples/refcoco/inference.py

@@ -96,12 +96,14 @@ def inference(
             do_sample=True,
             temperature=0.9,
         )
-
-    for i in range(inputs['input_ids'].shape[0]):
-        prompt_len = len(inputs['input_ids'][i])
-        output_text = tokenizer.decode(output[i][prompt_len:], skip_special_tokens=True).replace("<|im_end|>", "")
 
-    return output_text 
+    for i in range(inputs["input_ids"].shape[0]):
+        prompt_len = len(inputs["input_ids"][i])
+        output_text = tokenizer.decode(
+            output[i][prompt_len:], skip_special_tokens=True
+        ).replace("<|im_end|>", "")
+
+    return output_text
 
 
 def parse_bbox(model_output, img_wh):