Blame for app.py

import gradio as gr
import random
import re
import json

# ========== 可选：加载免费开源小模型（CPU本地推理，无需Key） ==========
USE_LLM = True
tokenizer = model = None
MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"
try:
    import torch
    from transformers import AutoTokenizer, AutoModelForCausalLM
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_NAME,
        trust_remote_code=True,
        torch_dtype=torch.float32
    )
    model.eval()
except Exception:
    USE_LLM = False

# ===== 名字合法性：拒绝拼音/奇怪符号，仅允许一个单词 =====
ALLOWLIST = {"Chloe","Charlotte","Xavier","Zach","Shawn","Shane","Zoe","Zoey","Chase","Charles","Xena","Phoenix"}
PINYIN_FRAGMENTS = [
    "zh","ch","sh","xue","xiao","ming","jing","qiang","qian","qin","qiu","xuan","xun",
    "yue","yu","yang","ying","yong","yan","jun","hao","lei","peng","lin","zhao","zhou","zhang",
    "wang","li","liu","chen","gao","guo","wu","he","luo","sun","tang","song"
]
NAME_RE = re.compile(r"^[A-Z][a-z]{2,15}$")

def looks_like_pinyin(name:str)->bool:
    return (name not in ALLOWLIST) and any(f in name.lower() for f in PINYIN_FRAGMENTS)

def name_sanity(name:str)->bool:
    return bool(NAME_RE.match(name or "")) and (not looks_like_pinyin(name))

# ===== 基于中文的“主题/意象”提取 =====
# 主题 -> 候选英文“常规名”和“创意词名”
THEME_POOL = {
    "bright": {
        "male": ["Leo","Lucian","Clarence","Ray"],
        "female": ["Clara","Lucy","Claire","Eleanor"],
        "word": ["Bright","Dawn","Halo","Glow","Solar"]
    },
    "wisdom": {
        "male": ["Solomon","Hugo","Sage"],
        "female": ["Sophia","Sage","Minerva"],
        "word": ["Oracle","Logic","Riddle","Cipher"]
    },
    "brave": {
        "male": ["Andrew","Conrad","Leon","Victor"],
        "female": ["Valerie","Andrea","Bridget"],
        "word": ["Valor","Brave","Vanguard","Grit","Steel"]
    },
    "strong": {
        "male": ["Ethan","Arnold","Magnus"],
        "female": ["Matilda","Valerie"],
        "word": ["Atlas","Titan","Forge","Anchor"]
    },
    "peace": {
        "male": ["Irenaeus","Pax","Felix"],
        "female": ["Irene","Serena","Pax"],
        "word": ["Truce","Calm","Zen","Peace"]
    },
    "snow": {
        "male": ["Nevin","Frost"],
        "female": ["Neve","Elsa"],
        "word": ["Snow","Frost","Blanc","Crystal"]
    },
    "pure": {
        "male": ["Aaron"],
        "female": ["Agnes","Catherine"],
        "word": ["Pure","Prism"]
    },
    "dawn": {
        "male": ["Aiden"],
        "female": ["Aurora","Zora"],
        "word": ["Dawn","Eos","Sunrise","Firstlight"]
    },
    "sunny": {
        "male": ["Sunny","Helios"],
        "female": ["Sunny","Solana"],
        "word": ["Sunny","Sunbeam","Sol"]
    },
    "cosmic": {
        "male": ["Cosmo","Orion"],
        "female": ["Nova","Stella"],
        "word": ["Nova","Stellar","Comet","Nebula","Orbit"]
    },
    "grand": {
        "male": ["August","Grant"],
        "female": ["Gloria","Augusta"],
        "word": ["Grand","Majesty","Prime"]
    },
    "beauty": {
        "male": ["Beau"],
        "female": ["Bella"],
        "word": ["Beauty","Belle"]
    },
    "serene": {
        "male": ["Irenaeus"],
        "female": ["Serena","Irene"],
        "word": ["Serene","Still","Quiet"]
    },
    "talented": {
        "male": ["Aiden","Gavin","Kevin"],
        "female": ["Ava","Mia"],
        "word": ["Talent","Flair","Spark"]
    },
    "positive": {
        "male": ["Felix"],
        "female": ["Joy"],
        "word": ["Joy","Lucky","Fortune"]
    },
    # 通用兜底
    "general": {
        "male": ["Henry","Daniel","Samuel","Arthur","Miles","James","William","Michael"],
        "female": ["Emma","Olivia","Emily","Grace","Lucy","Clara","Jane","Rachel"],
        "word": ["Echo","River","Willow","Raven","Skye","Blaze","Rider","Fox","Bolt"]
    }
}

# 中文字符到主题映射
CHAR2THEMES = {
    "明": ["bright","wisdom","dawn"],
    "文": ["wisdom"],
    "静": ["serene","peace"],
    "勇": ["brave"],
    "伟": ["grand","strong"],
    "强": ["strong"],
    "俊": ["talented"],
    "晨": ["dawn","bright"],
    "雪": ["snow","pure"],
    "宁": ["peace","serene"],
    "美": ["beauty"],
    "慧": ["wisdom"],
    "宇": ["cosmic","grand"],
    "阳": ["sunny","positive","bright"],
}

# 引用库（扩展版，含一些“词名”的流行文化对照）
REFERENCE_DB = {
    "Elizabeth": ["Elizabeth Bennet（《傲慢与偏见》）"],
    "Darcy": ["Mr. Darcy（《傲慢与偏见》）"],
    "Emma": ["Emma Woodhouse（《爱玛》）"],
    "Jane": ["Jane Eyre（《简·爱》）"],
    "Alice": ["Alice（《爱丽丝梦游仙境》）"],
    "Hermione": ["Hermione Granger（《哈利·波特》）"],
    "Harry": ["Harry Potter（《哈利·波特》）"],
    "Ron": ["Ron Weasley（《哈利·波特》）"],
    "Luna": ["Luna Lovegood（《哈利·波特》）"],
    "Clara": ["Clara Oswald（《神秘博士》）"],
    "Lucy": ["Lucy Pevensie（《纳尼亚传奇》）"],
    "Peter": ["Peter Parker（蜘蛛侠）"],
    "Tony": ["Tony Stark（钢铁侠）"],
    "Bruce": ["Bruce Wayne（蝙蝠侠）"],
    "Arthur": ["Arthur（亚瑟王传说）"],
    "Victor": ["Victor Frankenstein（《弗兰肯斯坦》）"],
    "Magnus": ["Magnus Bane（《凡人圣器》）"],
    "Gatsby": ["Jay Gatsby（《了不起的盖茨比》）"],
    "Daisy": ["Daisy Buchanan（《了不起的盖茨比》）"],
    "Holden": ["Holden Caulfield（《麦田里的守望者》）"],
    "Atticus": ["Atticus Finch（《杀死一只知更鸟》）"],
    "Romeo": ["Romeo（《罗密欧与朱丽叶》）"],
    "Juliet": ["Juliet（《罗密欧与朱丽叶》）"],
    "Arya": ["Arya Stark（《权力的游戏》）"],
    "Sansa": ["Sansa Stark（《权力的游戏》）"],
    "Jon": ["Jon Snow（《权力的游戏》）"],
    "Daenerys": ["Daenerys Targaryen（《权力的游戏》）"],
    "Sam": ["Samwise Gamgee（《指环王》）"],
    "Frodo": ["Frodo Baggins（《指环王》）"],
    "Gandalf": ["Gandalf（《指环王》）"],
    "Katniss": ["Katniss Everdeen（《饥饿游戏》）"],
    "Tris": ["Tris Prior（《分歧者》）"],
    "Leo": ["Leo McGarry（《白宫风云》）"],
    "Felix": ["Felix Krull（《骗子费利克斯》）"],
    "Aurora": ["Aurora（《睡美人》）"],
    "Holly": ["Holly Golightly（《蒂凡尼的早餐》）"],
    "Henry": ["Henry Higgins（《窈窕淑女》）"],
    "Emily": ["Emily Dickinson（诗人）"],
    "Olivia": ["Olivia Pope（《丑闻》）"],
    "Rachel": ["Rachel Green（《老友记》）"],
    "Monica": ["Monica Geller（《老友记》）"],
    "Phoebe": ["Phoebe Buffay（《老友记》）"],
    "Chandler": ["Chandler Bing（《老友记》）"],
    "Ross": ["Ross Geller（《老友记》）"],
    "Joey": ["Joey Tribbiani（《老友记》）"],
    "Michael": ["Michael Corleone（《教父》）"],
    "James": ["James Bond（007）"],
    "Ethan": ["Ethan Hunt（《碟中谍》）"],
    "Liam": ["Liam Neeson（演员）"],
    "Ava": ["Ava（《机械姬》）"],
    "Mia": ["Mia Thermopolis（《公主日记》）"],
    "Chloe": ["Chloe Price（《奇异人生》）"],
    "Willow": ["Willow Rosenberg（《吸血鬼猎人巴菲》）"],
    "River": ["River Tam（《萤火虫》）"],
    "Raven": ["Raven（《少年泰坦》）"],
    "Echo": ["Echo / Maya Lopez（漫威）"],
    "Nova": ["Nova（漫威/多部作品）"],
    "Logan": ["Logan / Wolverine（《X战警》）"],
    "Edward": ["Edward Cullen（《暮光之城》）"],
    "Bella": ["Bella Swan（《暮光之城》）"],
    "Skye": ["Skye / Daisy Johnson（《神盾局特工》）"],
    "Atlas": ["Atlas（希腊神话）"],
    "Orion": ["Orion（猎户座/神话）"],
    "Comet": ["Comet（多部作品/角色名常见称呼）"],
    "Blaze": ["Blaze the Cat（《索尼克》）"],
    "Fox": ["Fox Mulder（《X档案》）"],
}

NORMAL_POOL_MALE   = ["Ethan","Liam","Noah","Mason","Leo","Andrew","Victor","Felix","Grant","Conrad","Kevin","Gavin","Miles","Henry","Arthur","Daniel","Michael","James","Samuel","Benjamin","Alexander","Logan","Edward","William","Joseph"]
NORMAL_POOL_FEMALE = ["Emma","Olivia","Emily","Grace","Lucy","Clara","Jane","Rachel","Ava","Mia","Chloe","Stella","Bella","Claire","Phoebe","Serena","Irene","Aurora","Zoe","Sophie","Hannah","Sarah","Natalie","Victoria","Willow"]

MALE_SET   = set(NORMAL_POOL_MALE)
FEMALE_SET = set(NORMAL_POOL_FEMALE)

FEMALE_HINT = set(list("芳娜丽敏静婷雪梅玲颖倩婧薇诗妍悦宁珂伊菲茜可璇欣美慧"))
MALE_HINT   = set(list("强军伟刚涛磊鹏杰阳龙峰超辉宁宇航晨鑫豪剑霖勇健俊"))

def gender_hint_from_cn(name: str) -> str:
    score = 0
    for ch in name:
        if ch in FEMALE_HINT: score -= 1
        if ch in MALE_HINT: score += 1
    return "male" if score > 0 else "female"

# —— 从中文名提取主题列表（按频次和优先级） ——
def themes_from_cn(cn_name: str):
    themes = []
    for ch in cn_name:
        themes += CHAR2THEMES.get(ch, [])
    if not themes:
        themes = ["general"]
    # 去重保序
    seen = set()
    out = []
    for t in themes:
        if t not in seen:
            seen.add(t)
            out.append(t)
    if "general" not in out:
        out.append("general")
    return out

# —— 动态含义模板（避免每次一样） ——
MEANING_TEMPLATES = [
    "取“{chars}”的意象（{themes_cn}），所以选了 {name}。",
    "参考“{chars}”带来的感觉（{themes_cn}），给你 {name}。",
    "从“{chars}”联想到 {themes_cn}，因此是 {name}。",
    "因为“{chars}”→{themes_cn}，所以用了 {name} 这个名字。",
    "由“{chars}”启发（{themes_cn}），名字定为 {name}。"
]
THEME_ZH = {
    "bright":"明亮/光芒","wisdom":"智慧/学识","brave":"勇气/无畏","strong":"强壮/坚韧",
    "peace":"平和/宁静","snow":"雪意/纯净","pure":"纯洁","dawn":"黎明/新生","sunny":"阳光/积极",
    "cosmic":"宇宙/星空","grand":"宏大/气势","beauty":"美感/优雅","serene":"安静/从容",
    "talented":"才华/灵气","positive":"好运/积极","general":"通用/耐看"
}

def render_meaning(cn_name, used_themes, name):
    chars = "".join([ch for ch in cn_name if ch in CHAR2THEMES]) or cn_name
    zh = "、".join([THEME_ZH.get(t,t) for t in used_themes[:2]])  # 取前两个主题做说明
    tpl = random.choice(MEANING_TEMPLATES)
    return tpl.format(chars=chars, themes_cn=zh, name=name)

# —— 主题驱动选名：可能是常规英文名，也可能是“词名” ——
def themed_pick(cn_name: str, gender: str, creative_ratio=0.4, seed=None):
    rnd = random.Random(seed)
    themes = themes_from_cn(cn_name)
    t = rnd.choice(themes[:2] or themes)  # 更偏向前两个主题
    pool = THEME_POOL.get(t, THEME_POOL["general"])

    creative = rnd.random() < creative_ratio
    if creative:
        # 允许词名（不做男女白名单）
        cand = [w for w in pool["word"] if name_sanity(w)]
        if not cand:
            cand = [w for w in THEME_POOL["general"]["word"] if name_sanity(w)]
        pick = rnd.choice(cand)
        kind = "word"
    else:
        base = pool["male"] if gender=="male" else pool["female"]
        # 性别白名单 + 合法性
        cand = [n for n in base if name_sanity(n) and ((n in MALE_SET) if gender=="male" else (n in FEMALE_SET))]
        if not cand:
            base2 = NORMAL_POOL_MALE if gender=="male" else NORMAL_POOL_FEMALE
            cand = [n for n in base2 if name_sanity(n)]
        pick = rnd.choice(cand)
        kind = "given"

    meaning = render_meaning(cn_name, [t], pick)
    # 引用：优先直接命中；如果词名没命中，用若干主题近似兜底
    refs = REFERENCE_DB.get(pick, [])
    if not refs:
        approx = {
            "River":["River Tam（《萤火虫》）"],
            "Willow":["Willow Rosenberg（《吸血鬼猎人巴菲》）"],
            "Raven":["Raven（《少年泰坦》）"],
            "Echo":["Echo / Maya Lopez（漫威）"],
            "Nova":["Nova（漫威）》"],
            "Blaze":["Blaze the Cat（《索尼克》）"],
            "Fox":["Fox Mulder（《X档案》）"],
            "Atlas":["Atlas（希腊神话）"],
            "Orion":["Orion（神话/星座）"],
            "Comet":["Comet（常见角色/代号）"],
        }
        refs = approx.get(pick, [])
    if not refs:
        refs = ["未找到特别典型的文学/影视角色"]

    return {"name": pick, "meaning": meaning, "references": refs, "kind": kind}

# —— LLM生成（仍保留，但确保和主题相关；失败则用 themed_pick） ——
def llm_one(cn_name: str, gender: str, seed: int = None):
    if not USE_LLM or tokenizer is None or model is None:
        return themed_pick(cn_name, gender, seed=seed)

    themes = themes_from_cn(cn_name)[:3]
    sys_prompt = (
        "你是英文起名专家。给用户生成1个名字，它可以是：\n"
        "A) 常规英文名（Given name，单词），需符合指定性别；或\n"
        "B) 一个英文单词式的“词名”（如 Nova, Raven, River, Echo, Blaze 等，单词）。\n"
        "要求：\n"
        "1) 名字必须与中文名的意象/主题“稍有关联”，主题如下：" + ", ".join(themes) + "；\n"
        "2) 名字必须是一个单词；禁止中文拼音/音译（Li, Ming, Xue 等）；\n"
        "3) 若选择常规英文名，需符合性别（male/female）；\n"
        "4) 输出JSON对象：{\"name\":..., \"meaning\":..., \"references\":[...], \"kind\":\"given|word\"}；\n"
        "   meaning 用中文，简述与哪个汉字/主题有关；references 给2-4个文学/影视/名人同名（若无写“未找到特别典型的文学/影视角色”）。"
    )
    user = {"cn_name": cn_name, "gender": gender}
    prompt = f"<|im_start|>system\n{sys_prompt}\n<|im_end|>\n<|im_start|>user\n{json.dumps(user, ensure_ascii=False)}\n<|im_end|>\n<|im_start|>assistant\n"

    if seed and hasattr(torch, "manual_seed"):
        torch.manual_seed(seed)

    def try_once():
        inputs = tokenizer(prompt, return_tensors="pt")
        with torch.no_grad():
            out = model.generate(
                **inputs,
                max_new_tokens=320,
                do_sample=True,
                top_p=0.9,
                temperature=0.9,
                eos_token_id=tokenizer.eos_token_id
            )
        text = tokenizer.decode(out[0], skip_special_tokens=True)
        m = re.findall(r"\{.*\}", text, flags=re.S)
        if not m:
            return None
        try:
            obj = json.loads(m[-1])
            name = (obj.get("name") or "").strip()
            kind = (obj.get("kind") or "").strip().lower()
            meaning = (obj.get("meaning") or "").strip()
            refs = obj.get("references") or []
            if not name_sanity(name):
                return None
            if kind != "word":  # 常规名要性别匹配
                if gender=="male" and name not in MALE_SET: return None
                if gender=="female" and name not in FEMALE_SET: return None
                kind = "given"
            # 引用兜底
            if not isinstance(refs, list) or not refs:
                refs = REFERENCE_DB.get(name, []) or ["未找到特别典型的文学/影视角色"]
            if not meaning:
                meaning = render_meaning(cn_name, themes[:1] or ["general"], name)
            return {"name": name, "meaning": meaning, "references": refs[:4], "kind": kind}
        except Exception:
            return None

    for _ in range(4):
        r = try_once()
        if r:
            return r
        if hasattr(torch, "manual_seed"):
            torch.manual_seed(random.randint(1, 10_000_000))

    return themed_pick(cn_name, gender, seed=seed)

# ======== 页面交互 =========
def do_generate(cn_name, gender, seed):
    cn_name = (cn_name or "").strip()
    if not cn_name:
        return gr.update(visible=False), gr.update(visible=True), "请输入中文名字。", "", "—"
    if any(not ('\u4e00' <= ch <= '\u9fff') for ch in cn_name):
        return gr.update(visible=False), gr.update(visible=True), "请输入**中文**名字（例如：李明 / 王静）。", "", "—"
    if gender not in ("male","female"):
        gender = "male"

    res = llm_one(cn_name, gender, seed=seed)
    name, meaning, refs = res["name"], res["meaning"], res["references"]
    refs_md = "、".join(refs)
    title = f"🎉 你的英文名字：**{name}**"
    return gr.update(visible=False), gr.update(visible=True), title, f"含义/灵感：{meaning}", f"同名角色：{refs_md}"

def go_home():
    return gr.update(visible=True), gr.update(visible=False), "", "", ""

with gr.Blocks(title="给自己起一个英文名字", fill_height=True) as demo:
    seed_state = gr.State(value=random.randint(1, 10_000_000))

    # 首页
    gr.Markdown("## 给自己起一个英文名字")
    with gr.Group(visible=True) as home:
        gr.Markdown("输入中文名与性别，然后点击 **起名** 👇")
        with gr.Row():
            cn_name = gr.Textbox(label="中文名字", placeholder="例如：李明 / 王静 / 陈宇", lines=1)
        with gr.Row():
            gender = gr.Radio(choices=["male","female"], value="male", label="性别")
        btn_go = gr.Button("起名", variant="primary")

    # 结果页
    with gr.Group(visible=False) as result:
        result_title = gr.Markdown("")
        result_meaning = gr.Markdown("")
        result_refs = gr.Markdown("")
        with gr.Row():
            btn_shuffle = gr.Button("换名字")
            btn_back = gr.Button("返回首页", variant="secondary")

    # 绑定事件
    btn_go.click(
        do_generate,
        inputs=[cn_name, gender, seed_state],
        outputs=[home, result, result_title, result_meaning, result_refs]
    )

    def on_shuffle(cn, g, sd):
        new_seed = (sd or 0) + random.randint(7, 9999)
        return (*do_generate(cn, g, new_seed),)

    btn_shuffle.click(
        on_shuffle,
        inputs=[cn_name, gender, seed_state],
        outputs=[home, result, result_title, result_meaning, result_refs]
    )

    btn_back.click(
        go_home,
        inputs=[],
        outputs=[home, result, result_title, result_meaning, result_refs]
    )

if __name__ == "__main__":
    demo.launch()