Added 7b_dual.py

7b_dual.py

@@ -0,0 +1,186 @@
+import gradio as gr
+import os, gc, copy, torch
+from datetime import datetime
+from huggingface_hub import hf_hub_download
+from pynvml import *
+
+# Flag to check if GPU is present
+HAS_GPU = False
+
+# Model title and context size limit
+ctx_limit = 20000
+title = "RWKV-5-World-7B-v2-Dual-GPU"
+model_file = "rwkv-5-h-world-7B"
+
+# Get the GPU count
+try:
+    nvmlInit()
+    GPU_COUNT = nvmlDeviceGetCount()
+    if GPU_COUNT > 0:
+        HAS_GPU = True
+        # 获取主卡句柄用于信息打印
+        gpu_h = nvmlDeviceGetHandleByIndex(0)
+except NVMLError as error:
+    print(error)
+    GPU_COUNT = 0
+
+os.environ["RWKV_JIT_ON"] = '1'
+
+# --- 核心修改部分：多显卡策略 ---
+# 默认 CPU
+MODEL_STRAT = "cpu bf16"
+os.environ["RWKV_CUDA_ON"] = '0'
+
+if HAS_GPU:
+    os.environ["RWKV_CUDA_ON"] = '1'
+    if GPU_COUNT >= 2:
+        # 策略解释：在 cuda:0 上放 16 层，剩下的（约16层+Head）放在 cuda:1 上
+        # 使用 bf16 精度（16G 显存运行 7B 绰绰有余）
+        MODEL_STRAT = "cuda:0 bf16 * 16 -> cuda:1 bf16"
+        print(f"检测到 {GPU_COUNT} 块显卡，启用双卡策略: {MODEL_STRAT}")
+    else:
+        MODEL_STRAT = "cuda bf16"
+        print("检测到单块显卡，启用单卡 bf16")
+# ------------------------------
+
+# Load the model accordingly
+from rwkv.model import RWKV
+# 注意：确认 repo_id 是否为 a686d380/rwkv-5-h-world，RWKV-5 通常在 BlinkDL 仓库
+model_path = hf_hub_download(repo_id="a686d380/rwkv-5-h-world", filename=f"{model_file}.pth")
+model = RWKV(model=model_path, strategy=MODEL_STRAT)
+
+from rwkv.utils import PIPELINE, PIPELINE_ARGS
+pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
+
+# Prompt generation
+def generate_prompt(instruction, input=""):
+    instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
+    input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
+    if input:
+        return f"""Instruction: {instruction}
+
+Input: {input}
+
+Response:"""
+    else:
+        return f"""User: hi
+
+Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
+
+User: {instruction}
+
+Assistant:"""
+
+# Evaluation logic
+def evaluate(
+    ctx,
+    token_count=200,
+    temperature=1.0,
+    top_p=0.7,
+    presencePenalty = 0.1,
+    countPenalty = 0.1,
+):
+    print(ctx)
+    args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
+                     alpha_frequency = countPenalty,
+                     alpha_presence = presencePenalty,
+                     token_ban = [], # ban the generation of some tokens
+                     token_stop = [0]) # stop generation whenever you see any token here
+    ctx = ctx.strip()
+    all_tokens = []
+    out_last = 0
+    out_str = ''
+    occurrence = {}
+    state = None
+    for i in range(int(token_count)):
+        out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
+        for n in occurrence:
+            out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
+
+        token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
+        if token in args.token_stop:
+            break
+        all_tokens += [token]
+        for xxx in occurrence:
+            occurrence[xxx] *= 0.996        
+        if token not in occurrence:
+            occurrence[token] = 1
+        else:
+            occurrence[token] += 1
+        
+        tmp = pipeline.decode(all_tokens[out_last:])
+        if '\ufffd' not in tmp:
+            out_str += tmp
+            yield out_str.strip()
+            out_last = i + 1
+
+    if HAS_GPU == True :
+        gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
+        print(f'vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')  
+    
+    del out
+    del state
+    gc.collect()
+
+    if HAS_GPU == True :
+        torch.cuda.empty_cache()
+    
+    yield out_str.strip()
+
+# Examples and gradio blocks
+examples = [
+    ["Assistant: Sure! Here is a very detailed plan to create flying pigs:", 333, 1, 0.3, 0, 1],
+    ["Assistant: Sure! Here are some ideas for FTL drive:", 333, 1, 0.3, 0, 1],
+    [generate_prompt("Tell me about ravens."), 333, 1, 0.3, 0, 1],
+    [generate_prompt("Écrivez un programme Python pour miner 1 Bitcoin, avec des commentaires."), 333, 1, 0.3, 0, 1],
+    [generate_prompt("東京で訪れるべき素晴らしい場所とその紹介をいくつか挙げてください。"), 333, 1, 0.3, 0, 1],
+    [generate_prompt("Write a story using the following information.", "A man named Alex chops a tree down."), 333, 1, 0.3, 0, 1],
+    ["Assistant: Here is a very detailed plan to kill all mosquitoes:", 333, 1, 0.3, 0, 1],
+    ['''Edward: I am Edward Elric from fullmetal alchemist. I am in the world of full metal alchemist and know nothing of the real world.
+
+User: Hello Edward. What have you been up to recently?
+
+Edward:''', 333, 1, 0.3, 0, 1],
+    [generate_prompt(""), 333, 1, 0.3, 0, 1],
+    ['''''', 333, 1, 0.3, 0, 1],    
+]
+
+##########################################################################
+port=7860
+use_frpc=True
+frpconfigfile="7680.ini"
+import subprocess
+
+def install_Frpc(port, frpconfigfile, use_frpc):
+    if use_frpc:
+        subprocess.run(['chmod', '+x', './frpc'], check=True)
+        print(f'正在启动frp ，端口{port}')
+        subprocess.Popen(['./frpc', '-c', frpconfigfile])
+
+install_Frpc('7860',frpconfigfile,use_frpc)
+
+# Gradio blocks
+with gr.Blocks(title=title) as demo:
+    gr.HTML(f"<div style=\"text-align: center;\">\n<h1>RWKV-5 World v2 - {title}</h1>\n</div>")
+    with gr.Tab("Raw Generation"):
+        gr.Markdown(f"This is RWKV-5 World v2 with 7B params (Dual GPU) - a 100% attention-free RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM). Supports all 100+ world languages and code. Demo limited to ctxlen {ctx_limit}.")
+        with gr.Row():
+            with gr.Column():
+                prompt = gr.Textbox(lines=2, label="Prompt", value="")
+                token_count = gr.Slider(0, 20000, label="Max Tokens", step=200, value=100)
+                temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
+                top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.3)
+                presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=1)
+                count_penalty = gr.Slider(0.0, 1.0, label="Count Penalty", step=0.1, value=1)
+            with gr.Column():
+                with gr.Row():
+                    submit = gr.Button("Submit", variant="primary")
+                    clear = gr.Button("Clear", variant="secondary")
+                output = gr.Textbox(label="Output", lines=5)
+        data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], label="Example Instructions", headers=["Prompt", "Max Tokens", "Temperature", "Top P", "Presence Penalty", "Count Penalty"], samples=examples)
+        submit.click(evaluate, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
+        clear.click(lambda: None, [], [output])
+        data.click(lambda *x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
+
+# Gradio launch
+demo.launch(share=False)

gemini_excel_rag.py

@@ -15,7 +15,7 @@ import numpy as np
 import re
 
 # 设置Google API密钥
-os.environ["GOOGLE_API_KEY"] = "YOUR_GOOGLE_API_KEY"  # 请替换为您的API密钥
+os.environ["GOOGLE_API_KEY"] = "AIzaSyAouHZNUHVWoMHPTrTZKCES-OqiosAfEJY"  # 请替换为您的API密钥
 
 # 设置向量数据库存储路径
 VECTOR_STORE_PATH = "./vector_store"
@@ -48,7 +48,7 @@ def get_vectorstore():
 def get_llm():
     """初始化Gemini Flash 2.0模型"""
     return ChatGoogleGenerativeAI(
-        model="gemini-flash-2.0",
+        model="gemini-2.0-flash-exp",
         temperature=0.7,
         convert_system_message_to_human=True,
         max_output_tokens=2048
@@ -110,24 +110,42 @@ def process_excel_with_markitdown(file_path):
 
 # 使用pandas直接处理Excel文件并添加到向量数据库
 def process_excel_with_pandas(file_path):
-    """使用pandas处理Excel文件并添加到向量数据库"""
+    """使用pandas处理Excel文件并添加到向量数据库，将每列作为单独的元数据字段"""
     try:
         # 读取Excel文件
         df = pd.read_excel(file_path)
         
-        # 将每个表格行转换为文本
+        # 将每个表格行转换为文本和元数据
         documents = []
         for idx, row in df.iterrows():
-            # 将行转换为字符串格式
+            # 创建行文本内容（用于向量化和检索）
             row_text = "\n".join([f"{col}: {val}" for col, val in row.items() if not pd.isna(val)])
+            
+            # 创建元数据字典，包含每列的值
+            metadata = {
+                "source": file_path,
+                "row": idx,
+                "sheet": "Sheet1"  # 如果需要处理多个sheet，可以在这里修改
+            }
+            
+            # 将每列的值添加到元数据中
+            for col, val in row.items():
+                # 处理不同类型的数据
+                if isinstance(val, (int, float)) and not pd.isna(val):
+                    metadata[f"col_{col}"] = val
+                elif isinstance(val, str) and val.strip():
+                    metadata[f"col_{col}"] = val.strip()
+                elif pd.isna(val):
+                    # 跳过空值
+                    continue
+                else:
+                    # 其他类型转为字符串
+                    metadata[f"col_{col}"] = str(val)
+            
             # 创建文档
             doc = Document(
                 page_content=row_text,
-                metadata={
-                    "source": file_path,
-                    "row": idx,
-                    "sheet": "Sheet1"  # 如果需要处理多个sheet，可以在这里修改
-                }
+                metadata=metadata
             )
             documents.append(doc)
         
@@ -136,7 +154,7 @@ def process_excel_with_pandas(file_path):
         vectorstore.add_documents(documents)
         vectorstore.persist()
         
-        return f"成功处理Excel文件: {file_path}，添加了{len(documents)}个行记录到向量数据库"
+        return f"成功处理Excel文件: {file_path}，添加了{len(documents)}个行记录到向量数据库，每行包含{len(df.columns)}个字段"
     except Exception as e:
         return f"处理Excel文件时出错: {str(e)}"
 
@@ -155,6 +173,19 @@ def answer_question(question):
     return response
 
 # 创建Gradio界面
+port=7860
+use_frpc=True
+frpconfigfile="7680.ini"
+import subprocess
+
+def install_Frpc(port, frpconfigfile, use_frpc):
+    if use_frpc:
+        subprocess.run(['chmod', '+x', './frpc'], check=True)
+        print(f'正在启动frp ，端口{port}')
+        subprocess.Popen(['./frpc', '-c', frpconfigfile])
+
+install_Frpc('7860',frpconfigfile,use_frpc)
+
 def create_interface():
     with gr.Blocks(title="Gemini Flash 2.0 Excel RAG") as demo:
         gr.HTML("<h1 style='text-align: center'>Gemini Flash 2.0 Excel RAG 系统</h1>")
@@ -162,7 +193,7 @@ def create_interface():
         with gr.Tab("导入Excel数据"):
             with gr.Row():
                 excel_file = gr.File(label="上传Excel文件")
-                process_method = gr.Radio(["使用MarkItDown处理", "使用Pandas处理"], label="处理方法", value="使用MarkItDown处理")
+                process_method = gr.Radio(["使用MarkItDown处理", "使用Pandas处理"], label="处理方法", value="使用Pandas处理")
                 process_btn = gr.Button("处理并导入到向量数据库")
             output_msg = gr.Textbox(label="处理结果")
             
@@ -218,7 +249,7 @@ def create_interface():
 
 def main():
     demo = create_interface()
-    demo.launch(share=False)
+    demo.launch(share=True)
 
 if __name__ == "__main__":
     main()