add clip model for encoder

.gitignore

@@ -139,4 +139,7 @@ dmypy.json
 
 *.ini
 
-**/multicache_serving.py
+**/multicache_serving.py
+**/modelcache_serving.py
+
+**/model/

model/clip_zh/__init__.py

@@ -0,0 +1,12 @@
+# -*- coding: utf-8 -*-
+"""
+   Alipay.com Inc.
+   Copyright (c) 2004-2023 All Rights Reserved.
+   ------------------------------------------------------
+   File Name : __init__.py.py
+   Author : fuhui.phe
+   Create Time : 2024/5/7 14:05
+   Description : description what the main function of this file
+   Change Activity: 
+        version0 : 2024/5/7 14:05 by fuhui.phe  init
+"""

modelcache/adapter/adapter_query.py

@@ -30,17 +30,20 @@ def adapt_query(cache_data_convert, *args, **kwargs):
             report_func=chat_cache.report.embedding,
         )(pre_embedding_data)
 
+        # print('embedding_data: {}'.format(embedding_data))
+
     if cache_enable:
         cache_data_list = time_cal(
             chat_cache.data_manager.search,
-            func_name="milvus_search",
+            func_name="vector_search",
             report_func=chat_cache.report.search,
         )(
             embedding_data,
             extra_param=context.get("search_func", None),
             top_k=kwargs.pop("top_k", -1),
             model=model
         )
+        print('cache_data_list: {}'.format(cache_data_list))
         cache_answers = []
         cache_questions = []
         cache_ids = []
@@ -78,8 +81,8 @@ def adapt_query(cache_data_convert, *args, **kwargs):
             return
 
         for cache_data in cache_data_list:
+            print('cache_data: {}'.format(cache_data))
             primary_id = cache_data[1]
-            start_time = time.time()
             ret = chat_cache.data_manager.get_scalar_data(
                 cache_data, extra_param=context.get("get_scalar_data", None)
             )

modelcache/core.py

@@ -4,7 +4,7 @@
 from modelcache.processor.post import first
 from modelcache.similarity_evaluation import ExactMatchEvaluation
 from modelcache.similarity_evaluation import SimilarityEvaluation
-from modelcache.embedding.string import to_embeddings as string_embedding
+from modelcache.embedding.string_text import to_embeddings as string_embedding
 from modelcache.report import Report
 from modelcache.config import Config
 from modelcache.utils.cache_func import cache_all

modelcache/embedding/clip.py

@@ -0,0 +1,90 @@
+# -*- coding: utf-8 -*-
+import os
+import torch
+from modelcache.embedding.base import BaseEmbedding
+from modelscope.utils.constant import Tasks
+from modelscope.pipelines import pipeline
+from modelscope.preprocessors.image import load_image
+
+
+# def mean_pooling(model_output, attention_mask):
+#     token_embeddings = model_output[0]  # First element of model_output contains all token embeddings
+#     input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+#     return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+
+
+class ClipAudio(BaseEmbedding):
+    def __init__(self, model: str = "sentence-transformers/all-MiniLM-L6-v2"):
+        # current_dir = os.path.dirname(os.path.abspath(__file__))
+        # parent_dir = os.path.dirname(current_dir)
+        # model_dir = os.path.dirname(parent_dir)
+        # model = os.path.join(model_dir, 'model/text2vec-base-chinese/')
+
+        self.clip_pipeline = pipeline(task=Tasks.multi_modal_embedding,
+                                      model='damo/multi-modal_clip-vit-base-patch16_zh', model_revision='v1.0.1')
+
+        self.__dimension = 1024
+
+    def to_embeddings(self, data_dict, **_):
+        text_list = data_dict['text']
+        image_data = data_dict['image']
+
+        img_data = None
+        txt_data = None
+
+        if image_data:
+            input_img = load_image(image_data)
+            # 2D Tensor, [图片数, 特征维度]
+            img_embedding = self.clip_pipeline.forward({'img': input_img})['img_embedding'].tolist()[0] if input_img else []
+            print('img_embedding: {}'.format(img_embedding))
+        else:
+            raise ValueError('image_data is None, please check!')
+
+        if text_list and len(text_list) > 0:
+            # 2D Tensor, [文本数, 特征维度]
+            text_embedding = self.clip_pipeline.forward({'text': text_list})['text_embedding'].tolist()[0] if text_list else []
+            print('text_embedding: {}'.format(text_embedding))
+        else:
+            raise ValueError('text_list is None, please check!')
+
+        return {'image_embedding': img_embedding, 'text_embeddings': text_embedding}
+
+        # return {'image_embedding': img_feats, 'text_embeddings': txt_feats}
+        # input_texts = ["杰尼龟", "妙蛙种子", "小火龙", "皮卡丘"]
+        # input_img = load_image(
+        #     'https://clip-cn-beijing.oss-cn-beijing.aliyuncs.com/pokemon.jpeg')
+
+        # img_embedding = self.clip_pipeline.forward({'img': input_img})['img_embedding']  # 2D Tensor, [图片数, 特征维度]
+        # print('img_embedding: {}'.format(img_embedding))
+        # text_embedding = self.clip_pipeline.forward({'text': input_texts})['text_embedding']  # 2D Tensor, [文本数, 特征维度]
+
+
+        # return embedding_array
+
+    def post_proc(self, token_embeddings, inputs):
+        attention_mask = inputs["attention_mask"]
+        input_mask_expanded = (
+            attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+        )
+        sentence_embs = torch.sum(
+            token_embeddings * input_mask_expanded, 1
+        ) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+        return sentence_embs
+
+    @property
+    def dimension(self):
+        """Embedding dimension.
+
+        :return: embedding dimension
+        """
+        return self.__dimension
+
+
+# if __name__ == '__main__':
+#     clip_vec = ClipAudio()
+#     text_list = ['hello', '你好']
+#     text = ['###'.join(text_list)]
+#     image = 'https://clip-cn-beijing.oss-cn-beijing.aliyuncs.com/pokemon.jpeg'
+#     data_dict = {'text': text, 'image': image}
+#     resp = clip_vec.to_embeddings(data_dict)
+#     print('resp: {}'.format(resp))

modelcache/embedding/clip_demo.py

@@ -0,0 +1,49 @@
+# -*- coding: utf-8 -*-
+"""
+   Alipay.com Inc.
+   Copyright (c) 2004-2023 All Rights Reserved.
+   ------------------------------------------------------
+   File Name : clip_demo.py
+   Author : fuhui.phe
+   Create Time : 2024/5/7 11:58
+   Description : description what the main function of this file
+   Change Activity: 
+        version0 : 2024/5/7 11:58 by fuhui.phe  init
+"""
+import torch
+from modelscope.utils.constant import Tasks
+from modelscope.pipelines import pipeline
+from modelscope.preprocessors.image import load_image
+
+
+pipeline = pipeline(task=Tasks.multi_modal_embedding,
+                    model='damo/multi-modal_clip-vit-base-patch16_zh', model_revision='v1.0.1')
+
+# pipeline = pipeline(task=Tasks.multi_modal_embedding,
+#     model='/Users/penghongen/PycharmProjects/CodeFuse-ModelCache/model/clip_zh', model_revision='v1.0.1')
+
+# pipeline = pipeline(task=Tasks.multi_modal_embedding, model='/Users/penghongen/PycharmProjects/CodeFuse-ModelCache/model/clip_zh')
+
+
+input_img = load_image('https://clip-cn-beijing.oss-cn-beijing.aliyuncs.com/pokemon.jpeg') # 支持皮卡丘示例图片路径/本地图片 返回PIL.Image
+
+
+input_texts = ["杰尼龟", "妙蛙种子", "小火龙", "皮卡丘"]
+
+# 支持一张图片(PIL.Image)或多张图片(List[PIL.Image])输入，输出归一化特征向量
+img_embedding = pipeline.forward({'img': input_img})['img_embedding'] # 2D Tensor, [图片数, 特征维度]
+print('img_embedding: {}'.format(img_embedding))
+
+# 支持一条文本(str)或多条文本(List[str])输入，输出归一化特征向量
+text_embedding = pipeline.forward({'text': input_texts})['text_embedding'] # 2D Tensor, [文本数, 特征维度]
+
+# 计算图文相似度
+with torch.no_grad():
+    # 计算内积得到logit，考虑模型temperature
+    logits_per_image = (img_embedding / pipeline.model.temperature) @ text_embedding.t()
+    # 根据logit计算概率分布
+    probs = logits_per_image.softmax(dim=-1).cpu().numpy()
+
+print("图文匹配概率:", probs)
+
+

requirements.txt

@@ -8,7 +8,7 @@ pymilvus==2.3.1
 PyMySQL==1.1.0
 Requests==2.31.0
 torch==2.1.0
-transformers==4.34.1
+transformers==4.38.2
 faiss-cpu==1.7.4
 redis==5.0.1
-
+modelscope==1.14.0