Upload README.md

README.md

@@ -7,13 +7,7 @@ license: apache-2.0
 在运行脚本之前，首先安装如下必要的依赖。
 
 ```shell
-pip install --upgrade pip
-pip install torch transformers==4.40.0
-pip install sentencepiece protobuf
-pip install accelerate pillow
-pip install ninja
-pip install packaging
-pip install flash-attn --no-build-isolation
+pip install torch transformers==4.40.0 accelerate pillow sentencepiece
 ```
 
 ```python
@@ -24,6 +18,26 @@ from PIL import Image
 import warnings
 import numpy as np
 
+# set device
+device = 'cuda'  # or cpu
+torch.set_default_device(device)
+
+# create model
+model = AutoModelForCausalLM.from_pretrained(
+    'NaughtyDog97/FormalEnhencedGPS-9B',
+    torch_dtype=torch.float16, # float32 for cpu
+    device_map='auto',
+    trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained(
+    'NaughtyDog97/FormalEnhencedGPS-9B',
+    use_fast=False,
+    trust_remote_code=True)
+
+# text prompt
+img_path = 'sample/4927.png'
+qs = 'As shown in the diagram, AE/AB=1/4, M is the midpoint of segment AC, BE is parallel to CP, EA is parallel to CP. Find the ratio of the length of line BC to the length of line CD.'
+prompt = f'Using the provided geometric image and question, first predict the construction_cdl and image_cdl. Then, give a detailed step-by-step solution.\nThe question is:\n{qs}'
+text = f'<|im_start|>user\n<image>\n{prompt}<|im_end|>\n<|im_start|>assistant\n'
 
 def tokenizer_image_token(prompt, tokenizer, image_token_index, return_tensors=None):
     prompt_chunks = [tokenizer(chunk).input_ids for chunk in prompt.split('<image>')]
@@ -45,27 +59,7 @@ def tokenizer_image_token(prompt, tokenizer, image_token_index, return_tensors=N
             return torch.tensor(input_ids, dtype=torch.long)
         raise ValueError(f'Unsupported tensor type: {return_tensors}')
     return input_ids
-
-# set device
-device = 'cuda'  # or cpu
-torch.set_default_device(device)
-
-# create model
-model = AutoModelForCausalLM.from_pretrained(
-    'NaughtyDog97/FormalEnhencedGPS-9B',
-    torch_dtype=torch.float16, # float32 for cpu
-    device_map='auto',
-    trust_remote_code=True)
-tokenizer = AutoTokenizer.from_pretrained(
-    'NaughtyDog97/FormalEnhencedGPS-9B',
-    use_fast=False,
-    trust_remote_code=True)
-
-# text prompt
-img_path = 'sample/4927.png'
-qs = 'As shown in the diagram, AE/AB=1/4, M is the midpoint of segment AC, BE is parallel to CP, EA is parallel to CP. Find the ratio of the length of line BC to the length of line CD.'
-prompt = f'Using the provided geometric image and question, first predict the construction_cdl and image_cdl. Then, give a detailed step-by-step solution.\nThe question is:\n{qs}'
-text = f'<|im_start|>user\n<image>\n{prompt}<|im_end|>\n<|im_start|>assistant\n'
+    
 input_ids = tokenizer_image_token(text, tokenizer, -200, return_tensors='pt').unsqueeze(0).cuda()
 
 # image, sample images can be found in images folder