feat(思维导图助手)：优化打包
parent
59cbda30e2
commit
4b50b6bf4e
|
|
@ -38,10 +38,11 @@ RUN python3 -m pip install -U pip -i https://mirrors.aliyun.com/pypi/simple && \
|
|||
python3 -m pip install \
|
||||
"numpy==1.26.4" \
|
||||
"opencv-python==4.11.0.86" \
|
||||
"pydantic>=2.12,<3" \
|
||||
"modelscope" \
|
||||
"magic-pdf" \
|
||||
-i https://mirrors.aliyun.com/pypi/simple && \
|
||||
python3 -m pip install --force-reinstall --no-cache-dir "pydantic>=2.12,<3" -i https://mirrors.aliyun.com/pypi/simple && \
|
||||
python3 -c "import pydantic; print('pydantic', pydantic.__version__); from pydantic import TypeAdapter; print('pydantic TypeAdapter ok')" && \
|
||||
python3 -m pip cache purge
|
||||
|
||||
WORKDIR /app
|
||||
|
|
|
|||
|
|
@ -18,10 +18,12 @@ RUN python3 -m pip install -U pip -i https://mirrors.aliyun.com/pypi/simple && \
|
|||
numpy==1.26.4 \
|
||||
opencv-python==4.11.0.86 \
|
||||
-i https://mirrors.aliyun.com/pypi/simple && \
|
||||
python3 -m pip install --force-reinstall --no-cache-dir "pydantic>=2.12,<3" -i https://mirrors.aliyun.com/pypi/simple && \
|
||||
python3 -c "import pydantic; print('pydantic', pydantic.__version__); from pydantic import TypeAdapter; print('pydantic TypeAdapter ok')" && \
|
||||
python3 -m pip cache purge
|
||||
|
||||
# Download models and update the configuration file
|
||||
RUN /bin/bash -c "mineru-models-download -s modelscope -m all"
|
||||
|
||||
# Set the entry point to activate the virtual environment and run the command line tool
|
||||
ENTRYPOINT ["/bin/bash", "-c", "export MINERU_MODEL_SOURCE=local && exec \"$@\"", "--"]
|
||||
ENTRYPOINT ["/bin/bash", "-c", "export MINERU_MODEL_SOURCE=local && exec \"$@\"", "--"]
|
||||
|
|
|
|||
|
|
@ -361,10 +361,22 @@ def get_infer_result(file_suffix_identifier: str, pdf_name: str, parse_dir: str)
|
|||
api_router = APIRouter(prefix="/api")
|
||||
|
||||
|
||||
class MindmapLlmConfig(BaseModel):
|
||||
base_url: Optional[str] = None
|
||||
api_key: Optional[str] = None
|
||||
model: Optional[str] = None
|
||||
timeout: Optional[int] = None
|
||||
temperature: Optional[float] = None
|
||||
max_context_tokens: Optional[int] = None
|
||||
max_output_tokens: Optional[int] = None
|
||||
safety_tokens: Optional[int] = None
|
||||
|
||||
|
||||
class MindmapOrganizeRequest(BaseModel):
|
||||
markdown: str
|
||||
mode: str = "smart"
|
||||
prompt: Optional[str] = None
|
||||
llm_config: Optional[MindmapLlmConfig] = None
|
||||
|
||||
|
||||
def _extract_json_object(text: Optional[str]) -> str:
|
||||
|
|
@ -436,16 +448,34 @@ def _estimate_tokens(text: str) -> int:
|
|||
return max(1, int(ascii_chars / 4) + int(non_ascii_chars * 1.5))
|
||||
|
||||
|
||||
def _get_mindmap_max_output_tokens() -> int:
|
||||
return int(os.getenv("MINDMAP_LLM_MAX_OUTPUT_TOKENS", "4096"))
|
||||
def _optional_str(value: Optional[str], env_name: str, default: str = "") -> str:
|
||||
configured = (value or "").strip()
|
||||
return configured or os.getenv(env_name, default)
|
||||
|
||||
|
||||
def _get_mindmap_context_budget(prompt: str, reserve_output_tokens: Optional[int] = None) -> tuple[int, int]:
|
||||
def _optional_int(value: Optional[int], env_name: str, default: int) -> int:
|
||||
return int(value if value is not None else os.getenv(env_name, str(default)))
|
||||
|
||||
|
||||
def _optional_float(value: Optional[float], env_name: str, default: float) -> float:
|
||||
return float(value if value is not None else os.getenv(env_name, str(default)))
|
||||
|
||||
|
||||
def _get_mindmap_max_output_tokens(llm_config: Optional[MindmapLlmConfig] = None) -> int:
|
||||
value = llm_config.max_output_tokens if llm_config else None
|
||||
return _optional_int(value, "MINDMAP_LLM_MAX_OUTPUT_TOKENS", 4096)
|
||||
|
||||
|
||||
def _get_mindmap_context_budget(
|
||||
prompt: str,
|
||||
reserve_output_tokens: Optional[int] = None,
|
||||
llm_config: Optional[MindmapLlmConfig] = None
|
||||
) -> tuple[int, int]:
|
||||
if reserve_output_tokens is None:
|
||||
reserve_output_tokens = _get_mindmap_max_output_tokens()
|
||||
max_context_tokens = int(os.getenv("MINDMAP_LLM_MAX_CONTEXT_TOKENS", "32768"))
|
||||
reserve_output_tokens = _get_mindmap_max_output_tokens(llm_config)
|
||||
max_context_tokens = _optional_int(llm_config.max_context_tokens if llm_config else None, "MINDMAP_LLM_MAX_CONTEXT_TOKENS", 32768)
|
||||
prompt_tokens = _estimate_tokens(prompt)
|
||||
safety_tokens = int(os.getenv("MINDMAP_LLM_SAFETY_TOKENS", "1024"))
|
||||
safety_tokens = _optional_int(llm_config.safety_tokens if llm_config else None, "MINDMAP_LLM_SAFETY_TOKENS", 1024)
|
||||
input_budget_tokens = max(2048, max_context_tokens - prompt_tokens - reserve_output_tokens - safety_tokens)
|
||||
logger.info(
|
||||
"Mindmap context budget max_context_tokens={} prompt_tokens={} reserve_output_tokens={} safety_tokens={} input_budget_tokens={}",
|
||||
|
|
@ -527,12 +557,20 @@ def _chunk_markdown_by_headings(markdown: str, max_tokens: int) -> list[str]:
|
|||
return chunks or [markdown]
|
||||
|
||||
|
||||
def _call_mindmap_llm(markdown: str, mode: str = "smart", custom_prompt: Optional[str] = None, task_id: Optional[str] = None, request_role: str = "organize") -> str:
|
||||
base_url = os.getenv("MINDMAP_LLM_BASE_URL", "").rstrip("/")
|
||||
model = os.getenv("MINDMAP_LLM_MODEL", "gemma-4-26B")
|
||||
api_key = os.getenv("MINDMAP_LLM_API_KEY", "")
|
||||
timeout = int(os.getenv("MINDMAP_LLM_TIMEOUT", "180"))
|
||||
max_output_tokens = _get_mindmap_max_output_tokens()
|
||||
def _call_mindmap_llm(
|
||||
markdown: str,
|
||||
mode: str = "smart",
|
||||
custom_prompt: Optional[str] = None,
|
||||
task_id: Optional[str] = None,
|
||||
request_role: str = "organize",
|
||||
llm_config: Optional[MindmapLlmConfig] = None
|
||||
) -> str:
|
||||
base_url = _optional_str(llm_config.base_url if llm_config else None, "MINDMAP_LLM_BASE_URL").rstrip("/")
|
||||
model = _optional_str(llm_config.model if llm_config else None, "MINDMAP_LLM_MODEL", "gemma-4-26B")
|
||||
api_key = _optional_str(llm_config.api_key if llm_config else None, "MINDMAP_LLM_API_KEY")
|
||||
timeout = _optional_int(llm_config.timeout if llm_config else None, "MINDMAP_LLM_TIMEOUT", 180)
|
||||
max_output_tokens = _get_mindmap_max_output_tokens(llm_config)
|
||||
temperature = _optional_float(llm_config.temperature if llm_config else None, "MINDMAP_LLM_TEMPERATURE", 0.2)
|
||||
|
||||
if not base_url:
|
||||
raise RuntimeError("未配置智能整理模型服务，请设置 MINDMAP_LLM_BASE_URL")
|
||||
|
|
@ -545,8 +583,9 @@ def _call_mindmap_llm(markdown: str, mode: str = "smart", custom_prompt: Optiona
|
|||
{compact_markdown}
|
||||
"""
|
||||
logger.info(
|
||||
"Mindmap LLM request start task_id={} role={} model={} base_url={} mode={} input_chars={} input_tokens_est={} prompt_chars={} max_tokens={}",
|
||||
task_id or "-", request_role, model, base_url, mode, len(compact_markdown), _estimate_tokens(compact_markdown), len(prompt_template), max_output_tokens
|
||||
"Mindmap LLM request start task_id={} role={} model={} base_url={} mode={} input_chars={} input_tokens_est={} prompt_chars={} max_tokens={} timeout={} temperature={} api_key_configured={}",
|
||||
task_id or "-", request_role, model, base_url, mode, len(compact_markdown), _estimate_tokens(compact_markdown),
|
||||
len(prompt_template), max_output_tokens, timeout, temperature, bool(api_key)
|
||||
)
|
||||
|
||||
payload = {
|
||||
|
|
@ -555,7 +594,7 @@ def _call_mindmap_llm(markdown: str, mode: str = "smart", custom_prompt: Optiona
|
|||
{"role": "system", "content": "你擅长把长文档整理成结构清晰、层次合理的思维导图 Markdown，并严格保留原文语言。"},
|
||||
{"role": "user", "content": prompt},
|
||||
],
|
||||
"temperature": float(os.getenv("MINDMAP_LLM_TEMPERATURE", "0.2")),
|
||||
"temperature": temperature,
|
||||
"max_tokens": max_output_tokens,
|
||||
}
|
||||
data = json.dumps(payload, ensure_ascii=False).encode("utf-8")
|
||||
|
|
@ -600,9 +639,15 @@ def _call_mindmap_llm(markdown: str, mode: str = "smart", custom_prompt: Optiona
|
|||
return organized
|
||||
|
||||
|
||||
def _organize_mindmap_markdown(markdown: str, mode: str, custom_prompt: Optional[str], task_id: str) -> str:
|
||||
def _organize_mindmap_markdown(
|
||||
markdown: str,
|
||||
mode: str,
|
||||
custom_prompt: Optional[str],
|
||||
task_id: str,
|
||||
llm_config: Optional[MindmapLlmConfig]
|
||||
) -> str:
|
||||
prompt_template = (custom_prompt or "").strip() or DEFAULT_MINDMAP_ORGANIZE_PROMPT
|
||||
_, input_budget_tokens = _get_mindmap_context_budget(prompt_template)
|
||||
_, input_budget_tokens = _get_mindmap_context_budget(prompt_template, llm_config=llm_config)
|
||||
source_tokens = _estimate_tokens(markdown)
|
||||
logger.info(
|
||||
"Mindmap organize strategy task_id={} source_chars={} source_tokens_est={} input_budget_tokens={}",
|
||||
|
|
@ -611,7 +656,7 @@ def _organize_mindmap_markdown(markdown: str, mode: str, custom_prompt: Optional
|
|||
|
||||
if source_tokens <= input_budget_tokens:
|
||||
_update_task_progress(task_id, 35, "调用智能整理模型")
|
||||
return _call_mindmap_llm(markdown, mode, prompt_template, task_id, "single")
|
||||
return _call_mindmap_llm(markdown, mode, prompt_template, task_id, "single", llm_config)
|
||||
|
||||
chunks = _chunk_markdown_by_headings(markdown, input_budget_tokens)
|
||||
logger.info("Mindmap large input split task_id={} chunks={}", task_id, len(chunks))
|
||||
|
|
@ -623,14 +668,14 @@ def _organize_mindmap_markdown(markdown: str, mode: str, custom_prompt: Optional
|
|||
"Mindmap chunk organize task_id={} chunk={}/{} chars={} tokens_est={}",
|
||||
task_id, index, len(chunks), len(chunk), _estimate_tokens(chunk)
|
||||
)
|
||||
partial = _call_mindmap_llm(chunk, mode, prompt_template, task_id, f"chunk-{index}")
|
||||
partial = _call_mindmap_llm(chunk, mode, prompt_template, task_id, f"chunk-{index}", llm_config)
|
||||
partial_results.append(partial)
|
||||
|
||||
merged_input = "\n\n".join(
|
||||
f"<!-- chunk {index} -->\n{partial}"
|
||||
for index, partial in enumerate(partial_results, start=1)
|
||||
)
|
||||
_, merge_budget_tokens = _get_mindmap_context_budget(MINDMAP_MERGE_PROMPT)
|
||||
_, merge_budget_tokens = _get_mindmap_context_budget(MINDMAP_MERGE_PROMPT, llm_config=llm_config)
|
||||
merge_tokens = _estimate_tokens(merged_input)
|
||||
if merge_tokens > merge_budget_tokens:
|
||||
logger.warning(
|
||||
|
|
@ -641,14 +686,20 @@ def _organize_mindmap_markdown(markdown: str, mode: str, custom_prompt: Optional
|
|||
merged_round: list[str] = []
|
||||
for index, chunk in enumerate(merge_chunks, start=1):
|
||||
_update_task_progress(task_id, 78 + int(index / max(len(merge_chunks), 1) * 10), f"合并局部大纲 {index}/{len(merge_chunks)}")
|
||||
merged_round.append(_call_mindmap_llm(chunk, mode, MINDMAP_MERGE_PROMPT, task_id, f"merge-round-{index}"))
|
||||
merged_round.append(_call_mindmap_llm(chunk, mode, MINDMAP_MERGE_PROMPT, task_id, f"merge-round-{index}", llm_config))
|
||||
merged_input = "\n\n".join(merged_round)
|
||||
|
||||
_update_task_progress(task_id, 90, "全局整理标题结构")
|
||||
return _call_mindmap_llm(merged_input, mode, MINDMAP_MERGE_PROMPT, task_id, "merge")
|
||||
return _call_mindmap_llm(merged_input, mode, MINDMAP_MERGE_PROMPT, task_id, "merge", llm_config)
|
||||
|
||||
|
||||
async def _run_mindmap_organize_task(task_id: str, markdown: str, mode: str, prompt: Optional[str]):
|
||||
async def _run_mindmap_organize_task(
|
||||
task_id: str,
|
||||
markdown: str,
|
||||
mode: str,
|
||||
prompt: Optional[str],
|
||||
llm_config: Optional[MindmapLlmConfig]
|
||||
):
|
||||
try:
|
||||
_store_task_progress(task_id, {
|
||||
"progress": 10,
|
||||
|
|
@ -658,7 +709,7 @@ async def _run_mindmap_organize_task(task_id: str, markdown: str, mode: str, pro
|
|||
"file_names": "",
|
||||
"result_md": None,
|
||||
})
|
||||
organized = await asyncio.to_thread(_organize_mindmap_markdown, markdown, mode, prompt, task_id)
|
||||
organized = await asyncio.to_thread(_organize_mindmap_markdown, markdown, mode, prompt, task_id, llm_config)
|
||||
state = _get_task_progress(task_id) or {}
|
||||
state.update({
|
||||
"progress": 100,
|
||||
|
|
@ -722,10 +773,10 @@ async def create_mindmap_task(task_id: str, request: MindmapOrganizeRequest):
|
|||
"result_md": None,
|
||||
}
|
||||
_store_task_progress(task_id, state)
|
||||
asyncio.create_task(_run_mindmap_organize_task(task_id, markdown, request.mode, request.prompt))
|
||||
asyncio.create_task(_run_mindmap_organize_task(task_id, markdown, request.mode, request.prompt, request.llm_config))
|
||||
logger.info(
|
||||
"Registered mindmap organize task pid={} task_id={} mode={} input_chars={} custom_prompt={}",
|
||||
os.getpid(), task_id, request.mode, len(markdown), bool((request.prompt or "").strip())
|
||||
"Registered mindmap organize task pid={} task_id={} mode={} input_chars={} custom_prompt={} request_llm_config={}",
|
||||
os.getpid(), task_id, request.mode, len(markdown), bool((request.prompt or "").strip()), bool(request.llm_config)
|
||||
)
|
||||
return state
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,17 @@ export interface MindmapOrganizeProgress extends ParseProgress {
|
|||
result_md?: string | null
|
||||
}
|
||||
|
||||
export interface MindmapLlmConfig {
|
||||
base_url?: string
|
||||
api_key?: string
|
||||
model?: string
|
||||
timeout?: number
|
||||
temperature?: number
|
||||
max_context_tokens?: number
|
||||
max_output_tokens?: number
|
||||
safety_tokens?: number
|
||||
}
|
||||
|
||||
export const documentApi = {
|
||||
/**
|
||||
* 解析文档
|
||||
|
|
@ -102,11 +113,18 @@ export const documentApi = {
|
|||
})
|
||||
},
|
||||
|
||||
createMindmapTask(taskId: string, markdown: string, mode = 'smart', prompt?: string): Promise<MindmapOrganizeProgress> {
|
||||
createMindmapTask(
|
||||
taskId: string,
|
||||
markdown: string,
|
||||
mode = 'smart',
|
||||
prompt?: string,
|
||||
llmConfig?: MindmapLlmConfig
|
||||
): Promise<MindmapOrganizeProgress> {
|
||||
return request.post(`/api/mindmap_tasks/${encodeURIComponent(taskId)}`, {
|
||||
markdown,
|
||||
mode,
|
||||
prompt
|
||||
prompt,
|
||||
llm_config: llmConfig
|
||||
}).then(result => {
|
||||
return result as unknown as MindmapOrganizeProgress
|
||||
})
|
||||
|
|
|
|||
|
|
@ -130,6 +130,99 @@
|
|||
</div>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="大模型服务地址" class="form-item">
|
||||
<el-input
|
||||
v-model="draftConfig.mindmapLlmBaseUrl"
|
||||
placeholder="例如：http://10.100.53.199:9527/v1"
|
||||
class="input"
|
||||
clearable
|
||||
/>
|
||||
<div class="form-item-description">
|
||||
OpenAI 兼容接口地址，后端会请求该地址下的 /chat/completions；留空时使用服务端 MINDMAP_LLM_BASE_URL。
|
||||
</div>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="大模型 API Key" class="form-item">
|
||||
<el-input
|
||||
v-model="draftConfig.mindmapLlmApiKey"
|
||||
type="password"
|
||||
placeholder="留空则使用服务端 MINDMAP_LLM_API_KEY"
|
||||
class="input"
|
||||
show-password
|
||||
clearable
|
||||
/>
|
||||
<div class="form-item-description">
|
||||
仅随智能整理请求发送给后端，后端日志不会输出明文 Key。
|
||||
</div>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="大模型名称" class="form-item">
|
||||
<el-input
|
||||
v-model="draftConfig.mindmapLlmModel"
|
||||
placeholder="gemma-4-26B"
|
||||
class="input"
|
||||
clearable
|
||||
/>
|
||||
</el-form-item>
|
||||
|
||||
<div class="llm-grid">
|
||||
<el-form-item label="请求超时（秒）" class="form-item">
|
||||
<el-input-number
|
||||
v-model="draftConfig.mindmapLlmTimeout"
|
||||
:min="30"
|
||||
:max="3600"
|
||||
:step="30"
|
||||
controls-position="right"
|
||||
class="number-input"
|
||||
/>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="温度" class="form-item">
|
||||
<el-input-number
|
||||
v-model="draftConfig.mindmapLlmTemperature"
|
||||
:min="0"
|
||||
:max="2"
|
||||
:step="0.1"
|
||||
:precision="1"
|
||||
controls-position="right"
|
||||
class="number-input"
|
||||
/>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="最大上下文 Token" class="form-item">
|
||||
<el-input-number
|
||||
v-model="draftConfig.mindmapLlmMaxContextTokens"
|
||||
:min="4096"
|
||||
:max="262144"
|
||||
:step="1024"
|
||||
controls-position="right"
|
||||
class="number-input"
|
||||
/>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="最大输出 Token" class="form-item">
|
||||
<el-input-number
|
||||
v-model="draftConfig.mindmapLlmMaxOutputTokens"
|
||||
:min="512"
|
||||
:max="32768"
|
||||
:step="512"
|
||||
controls-position="right"
|
||||
class="number-input"
|
||||
/>
|
||||
</el-form-item>
|
||||
|
||||
<el-form-item label="安全预留 Token" class="form-item">
|
||||
<el-input-number
|
||||
v-model="draftConfig.mindmapLlmSafetyTokens"
|
||||
:min="0"
|
||||
:max="8192"
|
||||
:step="256"
|
||||
controls-position="right"
|
||||
class="number-input"
|
||||
/>
|
||||
</el-form-item>
|
||||
</div>
|
||||
|
||||
<el-form-item label="智能整理提示词" class="form-item">
|
||||
<el-input
|
||||
v-model="draftConfig.mindmapPrompt"
|
||||
|
|
@ -323,6 +416,16 @@ const getBackendDescription = (backend: string) => {
|
|||
line-height: 1.4;
|
||||
}
|
||||
|
||||
.llm-grid {
|
||||
display: grid;
|
||||
grid-template-columns: repeat(2, minmax(0, 1fr));
|
||||
gap: 0 12px;
|
||||
}
|
||||
|
||||
.number-input {
|
||||
width: 100%;
|
||||
}
|
||||
|
||||
.config-actions {
|
||||
display: flex;
|
||||
justify-content: flex-end;
|
||||
|
|
|
|||
|
|
@ -15,6 +15,14 @@ export interface DocumentConfig {
|
|||
serverUrl: string
|
||||
exportResolution: number
|
||||
mindmapPrompt: string
|
||||
mindmapLlmBaseUrl: string
|
||||
mindmapLlmApiKey: string
|
||||
mindmapLlmModel: string
|
||||
mindmapLlmTimeout: number
|
||||
mindmapLlmTemperature: number
|
||||
mindmapLlmMaxContextTokens: number
|
||||
mindmapLlmMaxOutputTokens: number
|
||||
mindmapLlmSafetyTokens: number
|
||||
tableEnable: boolean
|
||||
formulaEnable: boolean
|
||||
language: string
|
||||
|
|
@ -49,6 +57,14 @@ export const DEFAULT_DOCUMENT_CONFIG: DocumentConfig = {
|
|||
serverUrl: 'http://localhost:30000',
|
||||
exportResolution: 7680,
|
||||
mindmapPrompt: DEFAULT_MINDMAP_PROMPT,
|
||||
mindmapLlmBaseUrl: '',
|
||||
mindmapLlmApiKey: '',
|
||||
mindmapLlmModel: 'gemma-4-26B',
|
||||
mindmapLlmTimeout: 180,
|
||||
mindmapLlmTemperature: 0.2,
|
||||
mindmapLlmMaxContextTokens: 32768,
|
||||
mindmapLlmMaxOutputTokens: 4096,
|
||||
mindmapLlmSafetyTokens: 1024,
|
||||
tableEnable: true,
|
||||
formulaEnable: true,
|
||||
language: 'ch',
|
||||
|
|
|
|||
|
|
@ -456,7 +456,16 @@ const startSmartOrganize = async () => {
|
|||
smartOrganizeError.value = ''
|
||||
|
||||
try {
|
||||
await documentApi.createMindmapTask(taskId, markdown, 'hybrid', config.mindmapPrompt)
|
||||
await documentApi.createMindmapTask(taskId, markdown, 'hybrid', config.mindmapPrompt, {
|
||||
base_url: config.mindmapLlmBaseUrl || undefined,
|
||||
api_key: config.mindmapLlmApiKey || undefined,
|
||||
model: config.mindmapLlmModel || undefined,
|
||||
timeout: config.mindmapLlmTimeout,
|
||||
temperature: config.mindmapLlmTemperature,
|
||||
max_context_tokens: config.mindmapLlmMaxContextTokens,
|
||||
max_output_tokens: config.mindmapLlmMaxOutputTokens,
|
||||
safety_tokens: config.mindmapLlmSafetyTokens
|
||||
})
|
||||
startSmartOrganizePolling(taskId)
|
||||
} catch (err) {
|
||||
isSmartOrganizing.value = false
|
||||
|
|
|
|||
Loading…
Reference in New Issue