solve env conflict

agent/wulewule_agent.py

@@ -4,7 +4,6 @@ import requests
 from typing import List, Dict, Any, Optional, Iterator
 from PIL import Image
 import re
-import torch
 
 from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
 # from llama_index.core.postprocessor import LLMRerank
@@ -300,4 +299,3 @@ if __name__ == "__main__":
             # 使用st.audio函数播放音频
             st.audio("audio.mp3")
             st.write(f"语音内容为: {audio_text}")
-        torch.cuda.empty_cache()

app.py

@@ -33,6 +33,51 @@ def load_simple_rag(config, used_lmdeploy=False):
     wulewule_rag = WuleRAG(data_source_dir, db_persist_directory, base_mode, embeddings_model, reranker_model, rag_prompt_template)
     return wulewule_rag
 
+
+@st.cache_resource
+def load_wulewule_agent(config):
+    from agent.wulewule_agent import MultiModalAssistant, Settings
+    use_remote = config["use_remote"]
+    SiliconFlow_api = config["SiliconFlow_api"]
+    data_source_dir = config["data_source_dir"]
+    if len(SiliconFlow_api)<51 and os.environ.get('SiliconFlow_api', ""):
+        SiliconFlow_api = os.environ.get('SiliconFlow_api')
+
+    print(f"======= loading llm =======")
+    if use_remote:
+        from llama_index.llms.siliconflow import SiliconFlow
+        from llama_index.embeddings.siliconflow import SiliconFlowEmbedding
+        api_base_url =  "https://api.siliconflow.cn/v1/chat/completions"
+        # model = "Qwen/Qwen2.5-72B-Instruct"
+        # model = "deepseek-ai/DeepSeek-V2.5"
+        remote_llm = config["remote_llm"]
+        remote_embeddings_model = config["remote_embeddings_model"]
+        llm = SiliconFlow( model=remote_llm, base_url=api_base_url, api_key=SiliconFlow_api,  max_tokens=4096)
+        embed_model = SiliconFlowEmbedding(  model=remote_embeddings_model, api_key=SiliconFlow_api)
+    else:
+        from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+        from llama_index.llms.huggingface import HuggingFaceLLM
+        local_llm = config["llm_model"]
+        local_embeddings_model = config["agent_embeddings_model"]
+        llm = HuggingFaceLLM(
+            model_name=local_llm,
+            tokenizer_name=local_llm,
+            model_kwargs={"trust_remote_code":True},
+            tokenizer_kwargs={"trust_remote_code":True},
+            # context_window=4096,
+            # max_new_tokens=4096,
+        )
+        embed_model = HuggingFaceEmbedding(
+            model_name=local_embeddings_model
+        )
+    # settings
+    Settings.llm = llm
+    Settings.embed_model = embed_model
+    wulewule_assistant = MultiModalAssistant(data_source_dir, llm, SiliconFlow_api)
+    print(f"======= finished loading ! =======")
+    return wulewule_assistant
+
+
 GlobalHydra.instance().clear()
 @hydra.main(version_base=None, config_path="./configs", config_name="model_cfg")
 def main(cfg):

requirements.txt

@@ -1,29 +1,37 @@
-BCEmbedding
+# BCEmbedding
 transformers==4.45.0
 streamlit==1.36.0
-gradio==5.0.2
-sentencepiece==0.2.0
-accelerate==0.30.1
+# gradio==5.0.2
+# sentencepiece==0.2.0
+# accelerate==0.30.1
 transformers_stream_generator==0.0.5
-sentence-transformers==3.0.1
-peft==0.11.1
-xtuner==0.1.23 
+# sentence-transformers==3.0.1
+# peft==0.11.1
+# xtuner==0.1.23 
 openxlab
-tiktoken
-einops
-oss2
+# tiktoken
+# einops
+# oss2
 requests
-langchain==0.2.10
-langchain_community==0.2.9
-langchain_core
-langchain-huggingface==0.0.3
-langchain_text_splitters==0.2.2
-chromadb==0.5.0
-loguru==0.7.2
+# langchain==0.2.10
+# langchain_community==0.2.9
+# langchain_core
+# langchain-huggingface==0.0.3
+# langchain_text_splitters==0.2.2
+# chromadb==0.5.0
+# loguru==0.7.2
 modelscope==1.18.0
-numpy==1.26.4
-pandas==2.2.2
-timm==1.0.8 
+# numpy==1.26.4
+# pandas==2.2.2
+# timm==1.0.8 
 openai==1.40.3 
-lmdeploy[all]==0.5.3
-hydra-core==1.3.2
+# lmdeploy[all]==0.5.3
+hydra-core==1.3.2
+## agent used
+llama-index
+llama-index-core
+llama-index-llms-huggingface
+llama-index-embeddings-huggingface
+llama-index-llms-siliconflow
+llama-index-embeddings-siliconflow
+# huggingface-hub==0.27.0