使用小尺寸大模型和 Dify 清洗数据：Qwen 2.5 7B

docker pull pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime

docker run --gpus=all  -it -v `pwd`/Qwen:/models/Qwen -p 8000:8000 pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime bash

cd /models

python -m pip install --upgrade pip
pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple

pip install vllm

vllm serve Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4

# vllm serve Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4

INFO 09-3006:47:30 api_server.py:526] vLLM API server version 0.6.1.dev238+ge2c6e0a82
INFO 09-3006:47:30 api_server.py:527] args:Namespace(model_tag='Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4', config='', host=None, port=8000, uvicorn_log_level='info', allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_auto_tool_choice=False, tool_call_parser=None, model='Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4', tokenizer=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, download_dir=None, load_format='auto', config_format='auto', dtype='auto', kv_cache_dtype='auto', quantization_param_path=None, max_model_len=None, guided_decoding_backend='outlines', distributed_executor_backend=None, worker_use_ray=False, pipeline_parallel_size=1, tensor_parallel_size=1, max_parallel_loading_workers=None, ray_workers_use_nsight=False, block_size=16, enable_prefix_caching=False, disable_sliding_window=False, use_v2_block_manager=False, num_lookahead_slots=0, seed=0, swap_space=4, cpu_offload_gb=0, gpu_memory_utilization=0.9, num_gpu_blocks_override=None, max_num_batched_tokens=None, max_num_seqs=256, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, enforce_eager=False, max_context_len_to_capture=None, max_seq_len_to_capture=8192, disable_custom_all_reduce=False, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config=None, limit_mm_per_prompt=None, mm_processor_kwargs=None, enable_lora=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, enable_prompt_adapter=False, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', num_scheduler_steps=1, multi_step_stream_outputs=False, scheduler_delay_factor=0.0, enable_chunked_prefill=None, speculative_model=None, speculative_model_quantization=None, num_speculative_tokens=None, speculative_draft_tensor_parallel_size=None, speculative_max_model_len=None, speculative_disable_by_batch_size=None, ngram_prompt_lookup_max=None, ngram_prompt_lookup_min=None, spec_decoding_acceptance_method='rejection_sampler', typical_acceptance_sampler_posterior_threshold=None, typical_acceptance_sampler_posterior_alpha=None, disable_logprobs_during_spec_decoding=None, model_loader_extra_config=None, ignore_patterns=[], preemption_mode=None, served_model_name=None, qlora_adapter_name_or_path=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, override_neuron_config=None, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False, dispatch_function=<function serve at 0x74954f205760>)
INFO 09-3006:47:30 api_server.py:164]Multiprocessing frontend to use ipc:///tmp/7c3171d1-993e-4f8f-a466-95f0877a42a5for IPC Path.
INFO 09-3006:47:30 api_server.py:177]Started engine process with PID 45
INFO 09-3006:47:30 gptq_marlin.py:107]The model is convertible to gptq_marlin during runtime.Using gptq_marlin kernel.
INFO 09-3006:47:32 gptq_marlin.py:107]The model is convertible to gptq_marlin during runtime.Using gptq_marlin kernel.
INFO 09-3006:47:32 llm_engine.py:226]Initializing an LLM engine (v0.6.1.dev238+ge2c6e0a82) with config: model='Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4', speculative_config=None, tokenizer='Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, rope_scaling=None, rope_theta=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.float16, max_seq_len=32768, download_dir=None, load_format=LoadFormat.AUTO, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=gptq_marlin, enforce_eager=False, kv_cache_dtype=auto, quantization_param_path=None, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='outlines'), observability_config=ObservabilityConfig(otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=0, served_model_name=Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4, use_v2_block_manager=False, num_scheduler_steps=1, multi_step_stream_outputs=False, enable_prefix_caching=False, use_async_output_proc=True, use_cached_outputs=True, mm_processor_kwargs=None)
INFO 09-3006:47:33 model_runner.py:1014]Starting to load model Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4...
INFO 09-3006:47:33 gptq_marlin.py:198]UsingMarlinLinearKernelforGPTQMarlinLinearMethod
Loading safetensors checkpoint shards:0%Completed|0/2[00:00<?,?it/s]
Loading safetensors checkpoint shards:50%Completed|1/2[00:00<00:00,6.33it/s]
Loading safetensors checkpoint shards:100%Completed|2/2[00:00<00:00,3.06it/s]
Loading safetensors checkpoint shards:100%Completed|2/2[00:00<00:00,3.32it/s]

INFO 09-3006:47:34 model_runner.py:1025]Loading model weights took 5.1810 GB
INFO 09-3006:47:37 gpu_executor.py:122]# GPU blocks: 11455, # CPU blocks: 4681
INFO 09-3006:47:39 model_runner.py:1329]Capturing the model for CUDA graphs.This may lead to unexpected consequences if the model is not static.To run the model in eager mode,set'enforce_eager=True' or use '--enforce-eager'in the CLI.
INFO 09-3006:47:39 model_runner.py:1333] CUDA graphs can take additional 1~3GiB memory per GPU.If you are running out of memory, consider decreasing `gpu_memory_utilization` or enforcing eager mode.You can also reduce the `max_num_seqs` as needed to decrease memory usage.
INFO 09-3006:47:46 model_runner.py:1456]Graph capturing finished in7 secs.
INFO 09-3006:47:46 api_server.py:230] vLLM to use /tmp/tmpjt18oh55 as PROMETHEUS_MULTIPROC_DIR
WARNING 09-3006:47:46 serving_embedding.py:189] embedding_mode is False.Embedding API will not work.
INFO 09-3006:47:46 launcher.py:19]Available routes are:
INFO 09-3006:47:46 launcher.py:27]Route:/openapi.json,Methods: GET, HEAD
INFO 09-3006:47:46 launcher.py:27]Route:/docs,Methods: GET, HEAD
INFO 09-3006:47:46 launcher.py:27]Route:/docs/oauth2-redirect,Methods: GET, HEAD
INFO 09-3006:47:46 launcher.py:27]Route:/redoc,Methods: GET, HEAD
INFO 09-3006:47:46 launcher.py:27]Route:/health,Methods: GET
INFO 09-3006:47:46 launcher.py:27]Route:/tokenize,Methods: POST
INFO 09-3006:47:46 launcher.py:27]Route:/detokenize,Methods: POST
INFO 09-3006:47:46 launcher.py:27]Route:/v1/models,Methods: GET
INFO 09-3006:47:46 launcher.py:27]Route:/version,Methods: GET
INFO 09-3006:47:46 launcher.py:27]Route:/v1/chat/completions,Methods: POST
INFO 09-3006:47:46 launcher.py:27]Route:/v1/completions,Methods: POST
INFO 09-3006:47:46 launcher.py:27]Route:/v1/embeddings,Methods: POST
INFO:Started server process [37]
INFO:Waitingfor application startup.
INFO:Application startup complete.
INFO:Uvicorn running on http://0.0.0.0:8000(Press CTRL+C to quit)

function main({arg1}){
let input = arg1;
    input = input.split("\n").filter((line)=>{
return!line.includes("热点精选")
}).join("\n");
return{
result: input
}
}

app:
  description:''
icon:🤖
icon_background:'#FFEAD5'
mode:advanced-chat
name:快来帮我洗数据！(copy)
use_icon_as_answer_icon:false
kind:app
version:0.1.2
workflow:
conversation_variables:[]
environment_variables:[]
features:
file_upload:
image:
enabled:false
number_limits:3
transfer_methods:
-remote_url
-local_file
opening_statement:''
retriever_resource:
enabled:true
sensitive_word_avoidance:
configs:[]
enabled:false
type:''
speech_to_text:
enabled:false
suggested_questions:[]
suggested_questions_after_answer:
enabled:false
text_to_speech:
enabled:false
language:''
voice:''
graph:
edges:
-data:
isInIteration:false
sourceType:llm
targetType:llm
id:llm-source-1727679645155-target
source:llm
sourceHandle:source
target:'1727679645155'
targetHandle:target
type:custom
zIndex:0
-data:
isInIteration:false
sourceType:llm
targetType:answer
id:1727679645155-source-answer-target
source:'1727679645155'
sourceHandle:source
target:answer
targetHandle:target
type:custom
zIndex:0
-data:
isInIteration:false
sourceType:start
targetType:code
id:start-source-1727679927101-target
source:start
sourceHandle:source
target:'1727679927101'
targetHandle:target
type:custom
zIndex:0
-data:
isInIteration:false
sourceType:code
targetType:llm
id:1727679927101-source-llm-target
source:'1727679927101'
sourceHandle:source
target:llm
targetHandle:target
type:custom
zIndex:0
nodes:
-data:
selected:false
title:开始处理
type:start
variables:
-default:''
description:null
hint:null
label:content
max_length:null
options:null
required:true
type:paragraph
variable:content
-label:内容类型
max_length:48
options:[]
required:true
type:text-input
variable:type
height:116
id:start
position:
x:30
y:258
positionAbsolute:
x:30
y:258
selected:false
type:custom
width:244
-data:
context:
enabled:false
variable_selector:null
memory:
role_prefix:
assistant:''
user:''
window:
enabled:false
model:
completion_params:
stop:[]
temperature:0.1
mode:chat
name:Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4
provider:openai_api_compatible
prompt_template:
-id:8591e02f-6d72-480d-9b61-b39e3802bf43
role:user
text:'你是资深的记者，擅长严谨客观的表达内容，对下面的内容进行内容标题改写，使用陈述句表达。


            # 示例


            改写前


`text

            半导体需求火爆！韩国芯片库存以2009年以来最快速度减少

            财联社

            1小时前

`


            改写后


`text

            [财联社] 韩国芯片库存以2009年以来最快速度减少，半导体需求火爆

`


            ## 待处理内容


`text

{{#1727679927101.result#}}

`


            ## 处理结果


            '
selected:false
title:内容改写
type:llm
vision:
configs:null
enabled:false
variable_selector:null
height:98
id:llm
position:
x:638
y:258
positionAbsolute:
x:638
y:258
selected:false
type:custom
width:244
-data:
answer:'{{#1727679645155.text#}}'
selected:false
title:输出结果
type:answer
height:107
id:answer
position:
x:1246
y:258
positionAbsolute:
x:1246
y:258
selected:false
type:custom
width:244
-data:
context:
enabled:false
variable_selector:[]
desc:''
model:
completion_params:
temperature:0.1
mode:chat
name:Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4
provider:openai_api_compatible
prompt_template:
-id:3411a24a-f57d-4f26-9078-67a01915df57
role:system
text:'你是资深的记者，擅长严谨客观的表达内容，只输出待处理内容中和 “{{#start.type#}}“ 相关的内容。


            ## 待处理内容


`text

{{#llm.text#}}

`


            ## 处理结果




            '
selected:false
title:筛选内容
type:llm
variables:[]
vision:
enabled:false
height:98
id:'1727679645155'
position:
x:942
y:258
positionAbsolute:
x:942
y:258
selected:false
sourcePosition:right
targetPosition:left
type:custom
width:244
-data:
code:"\nfunction main({arg1}) {\n    let input = arg1;\n    input = input.split(\"\
          \\n\").filter((line)=>{\n        return !line.includes(\"热点精选\")\n    }).join(\"\
          \\n\");\n    return { \n        result: input\n    }\n}\n"
code_language:javascript
desc:''
outputs:
result:
children:null
type:string
selected:false
title:处理输入内容
type:code
variables:
-value_selector:
-start
-content
variable:arg1
height:54
id:'1727679927101'
position:
x:334
y:258
positionAbsolute:
x:334
y:258
selected:true
sourcePosition:right
targetPosition:left
type:custom
width:244
viewport:
x:-158.82992408947268
y:90.16096026215655
zoom: 0.961545363400739