Initial commit

merge_model.py

+import torch
+from peft import PeftModel
+from transformers import AutoTokenizer, AutoModel
+
+base_model_path = '../Finllm/Finllm_computing'
+lora_model_path = '../saves/Finllm_computing_train/lora/sft/checkpoint-4000'
+new_model_path = '../Finllm/Finllm_computing-merged'
+
+
+# Loading the base model and tokenzier
+model = AutoModel.from_pretrained(
+       base_model_path, torch_dtype=torch.float16, trust_remote_code=True
+    )
+
+tokenzier = AutoTokenizer.from_pretrained(
+       base_model_path, trust_remote_code=True
+    )
+
+# Loading the LoRA adapter
+lora_model = PeftModel.from_pretrained(model, lora_model_path, torch_dtype=torch.float16,)
+
+# Applying the LoRA
+new_model = lora_model.merge_and_unload()
+
+# Saving the target model to new model paths
+new_model.save_pretrained(new_model_path)
+tokenzier.save_pretrained(new_model_path)
\ No newline at end of file

model.properties

+# 模型唯一标识
+modelCode = 631
+# 模型名称
+modelName=DISC-FinLLM_pytorch           
+# 模型描述
+modelDescription=开源免费可商用的预训练多任务金融大模型可用于情绪分析, 关键词提取, 标题生成以及金融词提取,问答等。
+# 应用场景
+appScenario=推理,训练,金融,教育,政府,科研
+# 框架类型
+frameType=pytorch

multi_dcu_train.sh

+#!/bin/bash
+
+deepspeed --master_port $(shuf -n 1 -i 10000-65535) --include="localhost:3,4" /DISC-FinLLM_ModelZoo/LLaMA-Factory/src/train_bash.py \
+    --deepspeed /DISC-FinLLM_ModelZoo/LLaMA-Factory/deepspeed.json \
+    --stage sft \
+    --do_train \
+    --model_name_or_path /DISC-FinLLM_ModelZoo/FinLLM \
+    --dataset retrieval_part \
+    --dataset_dir /DISC-FinLLM_ModelZoo/LLaMA-Factory/data \
+    --template baichuan2 \
+    --finetuning_type lora \
+    --lora_target all \
+    --output_dir /DISC-FinLLM_ModelZoo/saves/lora_multi.py \
+    --overwrite_output_dir \
+    --cutoff_len 1024 \
+    --preprocessing_num_workers 4 \
+    --per_device_train_batch_size 16 \
+    --per_device_eval_batch_size 4 \
+    --gradient_accumulation_steps 1 \
+    --lr_scheduler_type cosine \
+    --logging_steps 10 \
+    --warmup_steps 20 \
+    --save_steps 100 \
+    --eval_steps 100 \
+    --evaluation_strategy steps \
+    --load_best_model_at_end \
+    --learning_rate 5e-5 \
+    --num_train_epochs 10.0 \
+    --max_samples 3000 \
+    --val_size 0.1 \
+    --ddp_timeout 180000000 \
+    --plot_loss \
+    --fp16

requirements.txt

+torch
+transformers
+accelerate
+colorama
+cpm_kernels
+sentencepiece
+streamlit
+transformers_stream_generator
+peft
+trl>=0.7.6
+gradio>=3.38.0,<4.0.0
+scipy
+einops
+sentencepiece
+protobuf
+jieba
+rouge-chinese
+nltk
+uvicorn
+pydantic
+fastapi
+sse-starlette
\ No newline at end of file

sft_work_dtk.sh

+HIP_VISIBLE_DEVICES=6  python /DISC-FinLLM_ModelZoo/LLaMA-Factory/src/train_bash.py \
+    --stage sft \
+    --do_train \
+    --model_name_or_path /DISC-FinLLM_ModelZoo/FinLLM \
+    --dataset retrieval_part \
+    --dataset_dir /DISC-FinLLM_ModelZoo/LLaMA-Factory/data \
+    --template  baichuan2 \
+    --finetuning_type lora \
+    --lora_target all \
+    --output_dir /DISC-FinLLM_ModelZoo/saves/lora \
+    --overwrite_cache \
+    --overwrite_output_dir \
+    --cutoff_len 1024 \
+    --preprocessing_num_workers 1 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 1 \
+    --lr_scheduler_type cosine \
+    --logging_steps 10 \
+    --warmup_steps 20 \
+    --save_steps 100 \
+    --eval_steps 100 \
+    --evaluation_strategy steps \
+    --learning_rate 5e-5 \
+    --num_train_epochs 3.0 \
+    --max_samples 3000 \
+    --val_size 0.1 \
+    --ddp_timeout 180000000 \
+    --plot_loss True \
+    --fp16
+

sha_calculator.py

+import hashlib
+
+def calculate_sha1(file_path):
+    sha1_hash = hashlib.sha1()
+    with open(file_path,"rb") as f:
+        # Read and update hash string value in blocks of 4K
+        for byte_block in iter(lambda: f.read(4096),b""):
+            sha1_hash.update(byte_block)
+    return sha1_hash.hexdigest()
+
+print(calculate_sha1("/home/wanglch/projects/LLaMA-Factory/data/retrieval_part.json")) # c25232270e00337ca3889a04b5ae023129c7c154
\ No newline at end of file

web_demo.py

+import json
+import torch
+import streamlit as st
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers.generation.utils import GenerationConfig
+
+
+st.set_page_config(page_title="FudanDISC-FinLLM")
+st.title("FudanDISC-FinLLM🤖️")
+
+
+@st.cache_resource
+def init_model():
+    model_path = "/DISC-FinLLM/FinLLM"
+    model = AutoModelForCausalLM.from_pretrained(
+        model_path, torch_dtype=torch.float16, device_map="auto", trust_remote_code=True
+    )
+    model.generation_config = GenerationConfig.from_pretrained(model_path)
+    tokenizer = AutoTokenizer.from_pretrained(
+        model_path, use_fast=False, trust_remote_code=True
+    )
+    return model, tokenizer
+
+
+def clear_chat_history():
+    del st.session_state.messages
+
+
+def init_chat_history():
+    with st.chat_message("assistant", avatar="🤖"):
+        st.markdown("您好，我是复旦 DISC-FinLLM，很高兴为您服务💖")
+
+    if "messages" in st.session_state:
+        for message in st.session_state.messages:
+            avatar = "🙋‍♂️" if message["role"] == "user" else "🤖"
+            with st.chat_message(message["role"], avatar=avatar):
+                st.markdown(message["content"])
+    else:
+        st.session_state.messages = []
+
+    return st.session_state.messages
+
+
+def main():
+    model, tokenizer = init_model()
+    messages = init_chat_history()
+    if prompt := st.chat_input("Shift + Enter 换行，Enter 发送"):
+        with st.chat_message("user", avatar="🙋‍♂️"):
+            st.markdown(prompt)
+        messages.append({"role": "user", "content": prompt})
+        print(f"[user] {prompt}", flush=True)
+        with st.chat_message("assistant", avatar="🤖"):
+            placeholder = st.empty()
+            for response in model.chat(tokenizer, messages, stream=True):
+                placeholder.markdown(response)
+                if torch.backends.mps.is_available():
+                    torch.mps.empty_cache()
+        messages.append({"role": "assistant", "content": response})
+        print(json.dumps(messages, ensure_ascii=False), flush=True)
+
+        st.button("清空对话", on_click=clear_chat_history)
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file