app.py

import streamlit as st
import cshogi
from IPython.display import display
from transformers import T5ForConditionalGeneration, T5Tokenizer
import pandas as pd

#モデルの読み込み
tokenizer = T5Tokenizer.from_pretrained("pizzagatakasugi/shogi_t5", is_fast=True)
model = T5ForConditionalGeneration.from_pretrained("pizzagatakasugi/shogi_t5")
model.eval()

st.title("将棋解説文の自動生成")
df = pd.read_csv("./dataset10.csv")
num = st.text_input("0から9の数字を入力")

KIFU_TO_SQUARE_NAMES = [
                    '１一', '１二', '１三', '１四', '１五', '１六', '１七', '１八', '１九',
                    '２一', '２二', '２三', '２四', '２五', '２六', '２七', '２八', '２九',
                    '３一', '３二', '３三', '３四', '３五', '３六', '３七', '３八', '３九',
                    '４一', '４二', '４三', '４四', '４五', '４六', '４七', '４八', '４九',
                    '５一', '５二', '５三', '５四', '５五', '５六', '５七', '５八', '５九',
                    '６一', '６二', '６三', '６四', '６五', '６六', '６七', '６八', '６九',
                    '７一', '７二', '７三', '７四', '７五', '７六', '７七', '７八', '７九',
                    '８一', '８二', '８三', '８四', '８五', '８六', '８七', '８八', '８九',
                    '９一', '９二', '９三', '９四', '９五', '９六', '９七', '９八', '９九',
                ]
KIFU_FROM_SQUARE_NAMES = [
                    '11', '12', '13', '14', '15', '16', '17', '18', '19',
                    '21', '22', '23', '24', '25', '26', '27', '28', '29',
                    '31', '32', '33', '34', '35', '36', '37', '38', '39',
                    '41', '42', '43', '44', '45', '46', '47', '48', '49',
                    '51', '52', '53', '54', '55', '56', '57', '58', '59',
                    '61', '62', '63', '64', '65', '66', '67', '68', '69',
                    '71', '72', '73', '74', '75', '76', '77', '78', '79',
                    '81', '82', '83', '84', '85', '86', '87', '88', '89',
                    '91', '92', '93', '94', '95', '96', '97', '98', '99',
                ]

if  num in [str(x) for x in list(range(10))]:
    df = df.iloc[int(num)]
    st.write(df["game_type"],df["precedence_name"],df["follower_name"])
    sfen = df["sfen"].split("\n")
    bestlist = eval(df["bestlist"])
    best2list = eval(df["best2list"])
    te = []
    te_sf = []
    movelist = []

    #文字の正規化
    for x in range(len(sfen)):
        if x < 2:
            continue
        if len(sfen[x]) > 30:
            te_sf.append(sfen[x])
        else:
            #te.append(sfen[x])
            temp = sfen[x].split()
            num = temp[1][0] + temp[1][1]
            for y in range(len(KIFU_FROM_SQUARE_NAMES)):
                if num == KIFU_FROM_SQUARE_NAMES[y]:
                    sq = KIFU_TO_SQUARE_NAMES[y]
            word = sq+temp[1][2:]
            word = word.replace("竜","龍").replace("成銀","全").replace("成桂","圭").replace("成香","杏")
            if sfen[x].split()[1] not in ["投了" , "千日手" , "持将棋" , "反則勝ち"]:
                te.append(temp[0]+" "+word)
                movelist.append(word)
            else:
                movelist.append(sfen[x].split()[1])

    #盤面表示
    s = st.selectbox(label="手数を選択",options=te)

    with st.expander("parameter"):
        temp = st.slider("temperature",min_value=0.0,max_value=1.0,step=0.01,value=0.3,key=1)
        beams = st.slider("num_beams",min_value=1,max_value=5,step=1,value=1,key=2)
        tokens = st.slider("min_new_tokens",min_value=0,max_value=50,value=20,key=3)
    
    reload = st.button('盤面生成',key=0)
    if s in te and reload == True:
        reload = False
        idx = te.index(s)
        board = cshogi.Board(sfen=te_sf[idx+1])
        st.markdown(board.to_svg(),unsafe_allow_html=True)

        #入力文作成
        kifs=""
        cnt = 0
        for kif in movelist:
            if cnt > idx:
                break
            kif = kif.split("(")[0]
            kifs += kif
            cnt += 1
            
        best = ""
        for x in bestlist[idx]:
            best += x.split("(")[0]

        best2 = ""
        for y in best2list[idx]:
            best2 += y.split("(")[0]

        #st.write(idx,"入力",input)
        with st.spinner("推論中です..."):
            input = sfen[0]+sfen[1]+kifs+"最善手の予測手順は"+best+"次善手の予測手順は"+best2
            tokenized_inputs = tokenizer.encode(
                                    input, max_length= 512, truncation=True, 
                                    padding="max_length", return_tensors="pt"
                                )
        
            output_ids = model.generate(input_ids=tokenized_inputs,
                            max_length=512,
                            repetition_penalty=10.0, # 同じ文の繰り返しへのペナルティ
                            temperature = temp,
                            num_beams = beams,
                            min_new_tokens = tokens,
                            )
        
            output_text = tokenizer.decode(output_ids[0], skip_special_tokens=True,
                                                clean_up_tokenization_spaces=False)
            st.write(output_text)

    
        # temperature = st.slider("temperature",min_value=0.0,max_value=1.0,step=0.01,value=0.3,key=1)
        # num_beams = st.slider("num_beams",min_value=1,max_value=5,step=1,value=1,key=2)
        # min_new_tokens = st.slider("min_new_tokens",min_value=0,max_value=100,value=30,key=3)