update

2024.01.27_chat.guanjihuan.com/使用 Transformer 软件包调用开源模型 ChatGLM/ChatGLM.py

@@ -0,0 +1,105 @@
+"""
+This code is supported by the website: https://www.guanjihuan.com
+The newest version of this code is on the web page: https://www.guanjihuan.com/archives/38502
+"""
+
+import streamlit as st
+st.set_page_config(
+    page_title="Chat",
+    layout='wide'
+)
+
+choose_load_method = 1  # 选择加载模型的方式
+
+if choose_load_method == 0:
+    # 默认加载（需要13G显存）
+    @st.cache_resource
+    def load_model_chatglm3():
+        from transformers import AutoModel, AutoTokenizer
+        tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm3-6b-32k", trust_remote_code=True)
+        model = AutoModel.from_pretrained("THUDM/chatglm3-6b-32k",trust_remote_code=True).half().cuda()
+        model = model.eval()
+        return  model, tokenizer
+    model_chatglm3, tokenizer_chatglm3 = load_model_chatglm3()
+
+elif choose_load_method == 1:
+    # 量化加载（需要6G显存）
+    @st.cache_resource
+    def load_model_chatglm3():
+        from transformers import AutoTokenizer, BitsAndBytesConfig, AutoModelForCausalLM
+        tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm3-6b-32k", trust_remote_code=True)
+        nf4_config = BitsAndBytesConfig(
+            load_in_4bit=True,
+            bnb_4bit_quant_type="nf4",
+        )
+        model = AutoModelForCausalLM.from_pretrained("THUDM/chatglm3-6b-32k", trust_remote_code=True, quantization_config=nf4_config)
+        model = model.eval()
+        return  model, tokenizer
+    model_chatglm3, tokenizer_chatglm3 = load_model_chatglm3()
+
+elif choose_load_method == 2:
+    # 在CPU上加载（需要25G内存，对话速度会比较慢，不推荐）
+    @st.cache_resource
+    def load_model_chatglm3():
+        from transformers import AutoModel, AutoTokenizer
+        tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm3-6b-32k", trust_remote_code=True)
+        model = AutoModel.from_pretrained("THUDM/chatglm3-6b-32k",trust_remote_code=True).float()
+        model = model.eval()
+        return  model, tokenizer
+    model_chatglm3, tokenizer_chatglm3 = load_model_chatglm3()
+
+with st.sidebar:
+    with st.expander('参数', expanded=True):
+        max_length = 409600
+        top_p = st.slider('top_p', 0.01, 1.0, step=0.01, value=0.8, key='top_p_session')
+        temperature = st.slider('temperature', 0.51, 1.0, step=0.01, value=0.8, key='temperature_session') 
+        def reset_parameter():
+            st.session_state['top_p_session'] = 0.8
+            st.session_state['temperature_session'] = 0.8
+        reset_parameter_button = st.button('重置', on_click=reset_parameter)
+
+prompt = st.chat_input("在这里输入您的命令")
+
+def chat_response_chatglm3(prompt):
+    history, past_key_values = st.session_state.history_ChatGLM3, st.session_state.past_key_values_ChatGLM3
+    for response, history, past_key_values in model_chatglm3.stream_chat(tokenizer_chatglm3, prompt, history,
+                                                                past_key_values=past_key_values,
+                                                                max_length=max_length, top_p=top_p,
+                                                                temperature=temperature,
+                                                                return_past_key_values=True):
+        message_placeholder_chatglm3.markdown(response)
+        if stop_button:
+            break
+    st.session_state.ai_response.append({"role": "robot", "content": response, "avatar": "assistant"})
+    st.session_state.history_ChatGLM3 = history
+    st.session_state.past_key_values_ChatGLM3 = past_key_values
+    return response
+
+def clear_all():
+    st.session_state.history_ChatGLM3 = []
+    st.session_state.past_key_values_ChatGLM3 = None
+    st.session_state.ai_response = []
+
+if 'history_ChatGLM3' not in st.session_state:
+    st.session_state.history_ChatGLM3 = []
+if 'past_key_values_ChatGLM3' not in st.session_state:
+    st.session_state.past_key_values_ChatGLM3 = None
+if 'ai_response' not in st.session_state:
+    st.session_state.ai_response = []
+
+for ai_response in st.session_state.ai_response:
+    with st.chat_message(ai_response["role"], avatar=ai_response.get("avatar")):
+        st.markdown(ai_response["content"])
+
+prompt_placeholder = st.chat_message("user", avatar='user')
+with st.chat_message("robot", avatar="assistant"):
+    message_placeholder_chatglm3 = st.empty()
+
+if prompt:
+    prompt_placeholder.markdown(prompt)
+    st.session_state.ai_response.append({"role": "user", "content": prompt, "avatar": 'user'})
+    stop = st.empty()
+    stop_button = stop.button('停止', key='break_response')
+    chat_response_chatglm3(prompt)
+    stop.empty()
+button_clear = st.button("清空", on_click=clear_all, key='clear')

2024.01.27_chat.guanjihuan.com/使用 Transformer 软件包调用开源模型 ChatGLM/requirements.txt

@@ -0,0 +1,33 @@
+# basic requirements
+
+protobuf>=4.25.2
+transformers>=4.36.2
+tokenizers>=0.15.0
+cpm_kernels>=1.0.11
+torch>=2.1.0
+gradio>=4.14.0
+sentencepiece>=0.1.99
+sentence_transformers>=2.2.2
+accelerate>=0.26.1
+streamlit>=1.30.0
+fastapi>=0.109.0
+loguru~=0.7.2
+mdtex2html>=1.2.0
+latex2mathml>=3.77.0
+
+# for openai demo
+
+openai>=1.7.2
+zhipuai>=2.0.0
+
+pydantic>=2.5.3
+sse-starlette>=1.8.2
+uvicorn>=0.25.0
+timm>=0.9.12
+tiktoken>=0.5.2
+
+# for langchain demo
+
+langchain>=0.1.0
+langchainhub>=0.1.14
+arxiv>=2.1.0