Update README.md
Browse files
README.md
CHANGED
@@ -69,6 +69,8 @@ This model, named `Evolutionary Multi-Modal Model`, is a multimodal transformer
|
|
69 |
|
70 |
### Direct Use
|
71 |
```python
|
|
|
|
|
72 |
git clone https://huggingface.co/zeroMN/SHMT.git
|
73 |
```
|
74 |
### Downstream Use
|
@@ -90,5 +92,74 @@ Users (both direct and downstream) should be made aware of the following risks,
|
|
90 |
|
91 |
## How to Get Started with the Model
|
92 |
```python
|
93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
```
|
|
|
69 |
|
70 |
### Direct Use
|
71 |
```python
|
72 |
+
git lfs install
|
73 |
+
|
74 |
git clone https://huggingface.co/zeroMN/SHMT.git
|
75 |
```
|
76 |
### Downstream Use
|
|
|
92 |
|
93 |
## How to Get Started with the Model
|
94 |
```python
|
95 |
+
import os
|
96 |
+
import torch
|
97 |
+
import torch.nn as nn
|
98 |
+
import numpy as np
|
99 |
+
import random
|
100 |
+
from transformers import (
|
101 |
+
BartForConditionalGeneration,
|
102 |
+
AutoModelForCausalLM,
|
103 |
+
BertModel,
|
104 |
+
Wav2Vec2Model,
|
105 |
+
CLIPModel,
|
106 |
+
AutoTokenizer
|
107 |
+
)
|
108 |
+
|
109 |
+
class MultiModalModel(nn.Module):
|
110 |
+
def __init__(self):
|
111 |
+
super(MultiModalModel, self).__init__()
|
112 |
+
# 初始化子模型
|
113 |
+
self.text_generator = BartForConditionalGeneration.from_pretrained('facebook/bart-base')
|
114 |
+
self.code_generator = AutoModelForCausalLM.from_pretrained('gpt2')
|
115 |
+
self.nlp_encoder = BertModel.from_pretrained('bert-base-uncased')
|
116 |
+
self.speech_encoder = Wav2Vec2Model.from_pretrained('facebook/wav2vec2-base-960h')
|
117 |
+
self.vision_encoder = CLIPModel.from_pretrained('openai/clip-vit-base-patch32')
|
118 |
+
|
119 |
+
# 初始化分词器和处理器
|
120 |
+
self.text_tokenizer = AutoTokenizer.from_pretrained('facebook/bart-base')
|
121 |
+
self.code_tokenizer = AutoTokenizer.from_pretrained('gpt2')
|
122 |
+
self.nlp_tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
|
123 |
+
self.speech_processor = AutoTokenizer.from_pretrained('facebook/wav2vec2-base-960h')
|
124 |
+
self.vision_processor = AutoTokenizer.from_pretrained('openai/clip-vit-base-patch32')
|
125 |
+
|
126 |
+
def forward(self, task, inputs):
|
127 |
+
if task == 'text_generation':
|
128 |
+
attention_mask = inputs.get('attention_mask')
|
129 |
+
outputs = self.text_generator.generate(
|
130 |
+
inputs['input_ids'],
|
131 |
+
max_new_tokens=100,
|
132 |
+
pad_token_id=self.text_tokenizer.eos_token_id,
|
133 |
+
attention_mask=attention_mask,
|
134 |
+
top_p=0.9,
|
135 |
+
top_k=50,
|
136 |
+
temperature=0.8,
|
137 |
+
do_sample=True
|
138 |
+
)
|
139 |
+
return self.text_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
140 |
+
elif task == 'code_generation':
|
141 |
+
attention_mask = inputs.get('attention_mask')
|
142 |
+
outputs = self.code_generator.generate(
|
143 |
+
inputs['input_ids'],
|
144 |
+
max_new_tokens=50,
|
145 |
+
pad_token_id=self.code_tokenizer.eos_token_id,
|
146 |
+
attention_mask=attention_mask,
|
147 |
+
top_p=0.95,
|
148 |
+
top_k=50,
|
149 |
+
temperature=1.2,
|
150 |
+
do_sample=True
|
151 |
+
)
|
152 |
+
return self.code_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
153 |
+
# 添加其他任务的逻辑...
|
154 |
+
|
155 |
+
# 计算模型参数数量的函数
|
156 |
+
def count_parameters(model):
|
157 |
+
return sum(p.numel() for p in model.parameters() if p.requires_grad)
|
158 |
+
|
159 |
+
# 初始化模型
|
160 |
+
model = MultiModalModel()
|
161 |
+
|
162 |
+
# 计算并打印模型参数数量
|
163 |
+
total_params = count_parameters(model)
|
164 |
+
print(f"模型总参数数量: {total_params}")
|
165 |
```
|