{"repo": "chuifengsky-lang/pdf-translate", "files": {"llm.py": {"b8c5b10d937c28fe73fb7787365d7c2c69807509": {"t": 1785746635.4015348, "n": 5031, "h": "e6f1d95f9b777afc", "c": "\"\"\"LLM client wrapper. DeepSeek is the initial provider.\n\nDeepSeek exposes an OpenAI-compatible API, so we use the `openai` SDK and point\nbase_url at DeepSeek. Both translate and define_word are STREAMING generators so\nthe UI can render output token-by-token.\n\"\"\"\n\nfrom openai import OpenAI\n\nPROVIDERS = {\n    \"deepseek\": {\n        \"base_url\": \"https://api.deepseek.com\",          # OpenAI-compatible\n        \"models\": [\"deepseek-v4-flash\", \"deepseek-v4-pro\"],\n        \"default_model\": \"deepseek-v4-flash\",\n    },\n    \"openai\": {\n        \"base_url\": \"https://api.openai.com/v1\",\n        \"models\": [\"gpt-4o-mini\", \"gpt-4o\"],\n        \"default_model\": \"gpt-4o-mini\",\n    },\n}\n\n\ndef models_for(provider):\n    return PROVIDERS.get(provider, {}).get(\"models\", [])\n\n\nclass LLMClient:\n    def __init__(self, provider=\"deepseek\", api_key=\"\", model=None):\n        if provider not in PROVIDERS:\n            raise ValueError(\"Unknown provider: %s\" % provider)\n        cfg = PROVIDERS[provider]\n        self.provider = provider\n        self.model = model or cfg[\"default_model\"]\n        self.client = OpenAI(api_key=api_key, base_url=cfg[\"base_url\"])\n\n    def translate_stream(self, text):\n        \"\"\"Yield Chinese translation token-by-token for one English paragraph.\"\"\"\n        prompt = (\n            \"Translate the following academic English text into Simplified Chinese.\\n\"\n            \"Rules:\\n\"\n            \"- Keep technical terms accurate.\\n\"\n            \"- Preserve all special symbols, mathematical notation, formulas, \"\n            \"inline equations, citation markers (e.g. [12], (Smith et al., 2020)), \"\n            \"URLs, and numbers exactly as they appear — do NOT translate or alter them.\\n\"\n            \"- Preserve the original spacing and line structure where meaningful.\\n\"\n            \"- Return ONLY the translation, with no notes or explanations.\\n\\n\"\n            + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.2,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def study_summary_stream(self, papers):\n        \"\"\"papers = [(name, text), ...]. Yield a Chinese study summary + advice:\n        per-paper summary, links between them, and a step-by-step learning plan.\"\"\"\n        parts = []\n        for i, (name, text) in enumerate(papers, 1):\n            parts.append(\"【论文%d：%s】\\n%s\" % (i, name, text))\n        body = \"\\n\\n\".join(parts)\n        prompt = (\n            \"你是一位学术导师。下面是我挑选要学习的论文内容。请用简体中文：\\n\"\n            \"1) 对每篇论文给出简明摘要（研究问题、方法、主要结论）；\\n\"\n            \"2) 指出它们的共同主题与相互联系；\\n\"\n            \"3) 给出循序渐进的学习建议：建议的阅读顺序、需要补充的背景知识、\"\n            \"每篇的重点章节、以及可延伸的研究方向。\\n\"\n            \"条理清晰，使用小标题（用 Markdown）。\\n\\n\" + body\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.4,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def explain_stream(self, text):\n        \"\"\"Yield a Chinese explanation of a selected passage: what it means plus\n        relevant concepts/background. This powers the selection popup (it is a\n        look-up / explanation, NOT a translation).\"\"\"\n        prompt = (\n            \"用简体中文解释下面这段选中的学术内容：\\n\"\n            \"1) 先简要说明它在说什么（含义/要点）；\\n\"\n            \"2) 再补充相关的概念、背景或知识点，帮助理解。\\n\"\n            \"只输出解释，不要逐字翻译原文。\\n\\n选中内容：\\n\" + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def define_word_stream(self, word, context):\n        \"\"\"Yield a Chinese definition + concept note token-by-token.\"\"\"\n        prompt = (\n            'Explain the term \"%s\" as used in this sentence:\\n\"%s\"\\n\\n'\n            \"Reply in Simplified Chinese with two short parts:\\n\"\n            \"1) 释义: a concise definition.\\n\"\n            \"2) 相关概念: 2-3 sentences of relevant background/conceptual knowledge.\"\n            % (word, context)\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n"}, "fcd7e545776bfe0a195719cb4b5d42b5c486c64d": {"t": 1785766434.3356311, "n": 5031, "h": "e6f1d95f9b777afc", "c": "\"\"\"LLM client wrapper. DeepSeek is the initial provider.\n\nDeepSeek exposes an OpenAI-compatible API, so we use the `openai` SDK and point\nbase_url at DeepSeek. Both translate and define_word are STREAMING generators so\nthe UI can render output token-by-token.\n\"\"\"\n\nfrom openai import OpenAI\n\nPROVIDERS = {\n    \"deepseek\": {\n        \"base_url\": \"https://api.deepseek.com\",          # OpenAI-compatible\n        \"models\": [\"deepseek-v4-flash\", \"deepseek-v4-pro\"],\n        \"default_model\": \"deepseek-v4-flash\",\n    },\n    \"openai\": {\n        \"base_url\": \"https://api.openai.com/v1\",\n        \"models\": [\"gpt-4o-mini\", \"gpt-4o\"],\n        \"default_model\": \"gpt-4o-mini\",\n    },\n}\n\n\ndef models_for(provider):\n    return PROVIDERS.get(provider, {}).get(\"models\", [])\n\n\nclass LLMClient:\n    def __init__(self, provider=\"deepseek\", api_key=\"\", model=None):\n        if provider not in PROVIDERS:\n            raise ValueError(\"Unknown provider: %s\" % provider)\n        cfg = PROVIDERS[provider]\n        self.provider = provider\n        self.model = model or cfg[\"default_model\"]\n        self.client = OpenAI(api_key=api_key, base_url=cfg[\"base_url\"])\n\n    def translate_stream(self, text):\n        \"\"\"Yield Chinese translation token-by-token for one English paragraph.\"\"\"\n        prompt = (\n            \"Translate the following academic English text into Simplified Chinese.\\n\"\n            \"Rules:\\n\"\n            \"- Keep technical terms accurate.\\n\"\n            \"- Preserve all special symbols, mathematical notation, formulas, \"\n            \"inline equations, citation markers (e.g. [12], (Smith et al., 2020)), \"\n            \"URLs, and numbers exactly as they appear — do NOT translate or alter them.\\n\"\n            \"- Preserve the original spacing and line structure where meaningful.\\n\"\n            \"- Return ONLY the translation, with no notes or explanations.\\n\\n\"\n            + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.2,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def study_summary_stream(self, papers):\n        \"\"\"papers = [(name, text), ...]. Yield a Chinese study summary + advice:\n        per-paper summary, links between them, and a step-by-step learning plan.\"\"\"\n        parts = []\n        for i, (name, text) in enumerate(papers, 1):\n            parts.append(\"【论文%d：%s】\\n%s\" % (i, name, text))\n        body = \"\\n\\n\".join(parts)\n        prompt = (\n            \"你是一位学术导师。下面是我挑选要学习的论文内容。请用简体中文：\\n\"\n            \"1) 对每篇论文给出简明摘要（研究问题、方法、主要结论）；\\n\"\n            \"2) 指出它们的共同主题与相互联系；\\n\"\n            \"3) 给出循序渐进的学习建议：建议的阅读顺序、需要补充的背景知识、\"\n            \"每篇的重点章节、以及可延伸的研究方向。\\n\"\n            \"条理清晰，使用小标题（用 Markdown）。\\n\\n\" + body\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.4,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def explain_stream(self, text):\n        \"\"\"Yield a Chinese explanation of a selected passage: what it means plus\n        relevant concepts/background. This powers the selection popup (it is a\n        look-up / explanation, NOT a translation).\"\"\"\n        prompt = (\n            \"用简体中文解释下面这段选中的学术内容：\\n\"\n            \"1) 先简要说明它在说什么（含义/要点）；\\n\"\n            \"2) 再补充相关的概念、背景或知识点，帮助理解。\\n\"\n            \"只输出解释，不要逐字翻译原文。\\n\\n选中内容：\\n\" + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def define_word_stream(self, word, context):\n        \"\"\"Yield a Chinese definition + concept note token-by-token.\"\"\"\n        prompt = (\n            'Explain the term \"%s\" as used in this sentence:\\n\"%s\"\\n\\n'\n            \"Reply in Simplified Chinese with two short parts:\\n\"\n            \"1) 释义: a concise definition.\\n\"\n            \"2) 相关概念: 2-3 sentences of relevant background/conceptual knowledge.\"\n            % (word, context)\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n"}, "e8201b62bbdf75331df05cc59f2aa465bd6f6e5d": {"t": 1785766435.7315009, "n": 3967, "h": "6002b6346abd1dac", "c": "\"\"\"LLM client wrapper. DeepSeek is the initial provider.\n\nDeepSeek exposes an OpenAI-compatible API, so we use the `openai` SDK and point\nbase_url at DeepSeek. Both translate and define_word are STREAMING generators so\nthe UI can render output token-by-token.\n\"\"\"\n\nfrom openai import OpenAI\n\nPROVIDERS = {\n    \"deepseek\": {\n        \"base_url\": \"https://api.deepseek.com\",          # OpenAI-compatible\n        \"models\": [\"deepseek-v4-flash\", \"deepseek-v4-pro\"],\n        \"default_model\": \"deepseek-v4-flash\",\n    },\n    \"openai\": {\n        \"base_url\": \"https://api.openai.com/v1\",\n        \"models\": [\"gpt-4o-mini\", \"gpt-4o\"],\n        \"default_model\": \"gpt-4o-mini\",\n    },\n}\n\n\ndef models_for(provider):\n    return PROVIDERS.get(provider, {}).get(\"models\", [])\n\n\nclass LLMClient:\n    def __init__(self, provider=\"deepseek\", api_key=\"\", model=None):\n        if provider not in PROVIDERS:\n            raise ValueError(\"Unknown provider: %s\" % provider)\n        cfg = PROVIDERS[provider]\n        self.provider = provider\n        self.model = model or cfg[\"default_model\"]\n        self.client = OpenAI(api_key=api_key, base_url=cfg[\"base_url\"])\n\n    def translate_stream(self, text):\n        \"\"\"Yield Chinese translation token-by-token for one English paragraph.\"\"\"\n        prompt = (\n            \"Translate the following academic English text into Simplified Chinese.\\n\"\n            \"Rules:\\n\"\n            \"- Keep technical terms accurate.\\n\"\n            \"- Preserve all special symbols, mathematical notation, formulas, \"\n            \"inline equations, citation markers (e.g. [12], (Smith et al., 2020)), \"\n            \"URLs, and numbers exactly as they appear — do NOT translate or alter them.\\n\"\n            \"- Preserve the original spacing and line structure where meaningful.\\n\"\n            \"- Return ONLY the translation, with no notes or explanations.\\n\\n\"\n            + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.2,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def explain_stream(self, text):\n        \"\"\"Yield a Chinese explanation of a selected passage: what it means plus\n        relevant concepts/background. This powers the selection popup (it is a\n        look-up / explanation, NOT a translation).\"\"\"\n        prompt = (\n            \"用简体中文解释下面这段选中的学术内容：\\n\"\n            \"1) 先简要说明它在说什么（含义/要点）；\\n\"\n            \"2) 再补充相关的概念、背景或知识点，帮助理解。\\n\"\n            \"只输出解释，不要逐字翻译原文。\\n\\n选中内容：\\n\" + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def define_word_stream(self, word, context):\n        \"\"\"Yield a Chinese definition + concept note token-by-token.\"\"\"\n        prompt = (\n            'Explain the term \"%s\" as used in this sentence:\\n\"%s\"\\n\\n'\n            \"Reply in Simplified Chinese with two short parts:\\n\"\n            \"1) 释义: a concise definition.\\n\"\n            \"2) 相关概念: 2-3 sentences of relevant background/conceptual knowledge.\"\n            % (word, context)\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n"}, "4d9d8d08f7a620eed6f00677fcab8e17c2a89641": {"t": 1785766438.1235468, "n": 3127, "h": "602d369938964401", "c": "\"\"\"LLM client wrapper. DeepSeek is the initial provider.\n\nDeepSeek exposes an OpenAI-compatible API, so we use the `openai` SDK and point\nbase_url at DeepSeek. Both translate and define_word are STREAMING generators so\nthe UI can render output token-by-token.\n\"\"\"\n\nfrom openai import OpenAI\n\nPROVIDERS = {\n    \"deepseek\": {\n        \"base_url\": \"https://api.deepseek.com\",          # OpenAI-compatible\n        \"models\": [\"deepseek-v4-flash\", \"deepseek-v4-pro\"],\n        \"default_model\": \"deepseek-v4-flash\",\n    },\n    \"openai\": {\n        \"base_url\": \"https://api.openai.com/v1\",\n        \"models\": [\"gpt-4o-mini\", \"gpt-4o\"],\n        \"default_model\": \"gpt-4o-mini\",\n    },\n}\n\n\ndef models_for(provider):\n    return PROVIDERS.get(provider, {}).get(\"models\", [])\n\n\nclass LLMClient:\n    def __init__(self, provider=\"deepseek\", api_key=\"\", model=None):\n        if provider not in PROVIDERS:\n            raise ValueError(\"Unknown provider: %s\" % provider)\n        cfg = PROVIDERS[provider]\n        self.provider = provider\n        self.model = model or cfg[\"default_model\"]\n        self.client = OpenAI(api_key=api_key, base_url=cfg[\"base_url\"])\n\n    def translate_stream(self, text):\n        \"\"\"Yield Chinese translation token-by-token for one English paragraph.\"\"\"\n        prompt = (\n            \"Translate the following academic English text into Simplified Chinese.\\n\"\n            \"Rules:\\n\"\n            \"- Keep technical terms accurate.\\n\"\n            \"- Preserve all special symbols, mathematical notation, formulas, \"\n            \"inline equations, citation markers (e.g. [12], (Smith et al., 2020)), \"\n            \"URLs, and numbers exactly as they appear — do NOT translate or alter them.\\n\"\n            \"- Preserve the original spacing and line structure where meaningful.\\n\"\n            \"- Return ONLY the translation, with no notes or explanations.\\n\\n\"\n            + text\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.2,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n\n    def define_word_stream(self, word, context):\n        \"\"\"Yield a Chinese definition + concept note token-by-token.\"\"\"\n        prompt = (\n            'Explain the term \"%s\" as used in this sentence:\\n\"%s\"\\n\\n'\n            \"Reply in Simplified Chinese with two short parts:\\n\"\n            \"1) 释义: a concise definition.\\n\"\n            \"2) 相关概念: 2-3 sentences of relevant background/conceptual knowledge.\"\n            % (word, context)\n        )\n        stream = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n            stream=True,\n        )\n        for chunk in stream:\n            if not chunk.choices:\n                continue\n            delta = chunk.choices[0].delta.content\n            if delta:\n                yield delta\n"}, "5b072f36f666f9b35e0c2a33dcb4be825c43f720": {"t": 1785766439.826134, "n": 2124, "h": "7aee9f549842c4c8", "c": "\"\"\"LLM client wrapper. DeepSeek is the initial provider.\n\nDeepSeek exposes an OpenAI-compatible API, so we use the `openai` SDK and just\npoint base_url at DeepSeek. Adding another provider later = one more dict entry.\n\"\"\"\n\nfrom openai import OpenAI\n\nPROVIDERS = {\n    \"deepseek\": {\n        \"base_url\": \"https://api.deepseek.com/v1\",\n        \"default_model\": \"deepseek-chat\",\n    },\n    \"openai\": {\n        \"base_url\": \"https://api.openai.com/v1\",\n        \"default_model\": \"gpt-4o-mini\",\n    },\n}\n\n\nclass LLMClient:\n    def __init__(self, provider=\"deepseek\", api_key=\"\", model=None):\n        if provider not in PROVIDERS:\n            raise ValueError(\"Unknown provider: %s\" % provider)\n        cfg = PROVIDERS[provider]\n        self.provider = provider\n        self.model = model or cfg[\"default_model\"]\n        self.client = OpenAI(api_key=api_key, base_url=cfg[\"base_url\"])\n\n    def translate(self, text):\n        \"\"\"Translate an English paragraph into Chinese, layout-neutral.\"\"\"\n        prompt = (\n            \"Translate the following academic English text into Simplified Chinese. \"\n            \"Keep technical terms accurate. Return ONLY the translation, no notes.\\n\\n\"\n            + text\n        )\n        resp = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.2,\n        )\n        return resp.choices[0].message.content.strip()\n\n    def define_word(self, word, context):\n        \"\"\"Return a Chinese definition + concept note for a word in context.\"\"\"\n        prompt = (\n            'Explain the term \"%s\" as used in this sentence:\\n\"%s\"\\n\\n'\n            \"Reply in Simplified Chinese with two short parts:\\n\"\n            \"1) 释义: a concise definition.\\n\"\n            \"2) 相关概念: 2-3 sentences of relevant background/conceptual knowledge.\"\n            % (word, context)\n        )\n        resp = self.client.chat.completions.create(\n            model=self.model,\n            messages=[{\"role\": \"user\", \"content\": prompt}],\n            temperature=0.3,\n        )\n        return resp.choices[0].message.content.strip()\n"}}}}