a-julia-bert-client

1. download model, then run bert-as-service

    pip install bert-serving-server
   
    bert-serving-start -model_dir ./model_dir -num_worker=4

2. in julia

   julia> include("BertClient.jl")
   julia> BertClient.bert_encode("我能吞下玻璃而不伤身体")
               768-element Array{Float32,1}:
                 0.63951504 
                 0.11443944 
                -0.09068978 
                 0.011962767
                 0.19503269 
                -0.5136112  
                -0.28722548 
                -0.24000126 
                -0.09366071 
                 0.9944896  
                 0.11499034 
               ⋮ 

   bert_encode( [str1,str2,...] ) is also okay

   bert_init() is optional,

   default: tcp_server_address="127.0.0.1", port_client_input=5555, port_server_output=5556

---

remember to install tensorflow-gpu for (server side) better performance

notice

bert-as-service may not be stable when not optimized