bilisum 1.20.0 → 1.20.1-alpha.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bilisum",
-  "version": "1.20.0",
+  "version": "1.20.1-alpha.2",
   "description": "BiliSum CLI: control the installed BiliSum app, or run standalone with its own runtime.",
   "bin": {
     "bilisum": "bin/bilisum.js"

package/runtime/VERSION

@@ -1 +1 @@
-1.20.0
+1.20.1-alpha.2

package/runtime/apps/service/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "video-sum-service"
-version = "1.20.0"
+version = "1.20.1-alpha.2"
 description = "Local backend service for the video summarizer."
 requires-python = ">=3.12"
 dependencies = [

package/runtime/apps/service/src/video_sum_service/integrations.py

@@ -3,6 +3,7 @@ import binascii
 import importlib.resources
 import io
 import json
+import logging
 import os
 from pathlib import Path
 import struct
@@ -28,6 +29,8 @@ from video_sum_infra.runtime import service_log_path
 from video_sum_service.context import COVER_CACHE_DIR, settings_manager
 from video_sum_service.settings_manager import SettingsUpdatePayload, is_blank_or_masked_secret
 
+logger = logging.getLogger(__name__)
+
 MAX_LOG_CHARS = 20_000
 MAX_LOG_LINE_CHARS = 1_000
 
@@ -134,7 +137,7 @@ def cache_cover_image(source_url: str, canonical_id: str, referer_url: str | Non
 
     # 视频封面 URL 白名单
     ALLOWED_COVER_DOMAINS = [
-        "bilibili.com", "hdslb.com", "bilivideo.com",  # B站
+        "bilibili.com", "hdslb.com", "bilivideo.com", "biliimg.com",  # B站
         "ytimg.com", "ggpht.com", "googleusercontent.com",  # YouTube
         "twimg.com",  # Twitter
     ]
@@ -330,7 +333,6 @@ def probe_llm_connection(payload: SettingsUpdatePayload | None = None) -> dict[s
             ],
             "temperature": 0,
             "max_tokens": 256,
-            "response_format": {"type": "json_object"},
         }
     else:
         request_payload = {

package/runtime/apps/service/src/video_sum_service/knowledge/local_llm.py

@@ -97,6 +97,12 @@ def chat_knowledge_llm(
     require_json: bool = False,
     _allow_empty_retry: bool = True,
 ) -> tuple[str, dict[str, object] | None]:
+    """调用知识库 LLM。
+
+    注意：`require_json` 不再设置 `response_format: json_object` ——
+    许多本地 LLM 后端（Ollama、llama.cpp、vLLM）会因该参数返回 400。
+    输出格式靠 system prompt 指示模型只输出合法 JSON。
+    """
     base_url, model, api_key, provider = ensure_knowledge_llm_settings(settings)
     use_anthropic = is_anthropic_llm(provider, base_url)
     request_model = model if use_anthropic else normalize_openai_compatible_model_name(model)
@@ -117,8 +123,9 @@ def chat_knowledge_llm(
         "temperature": temperature,
         "max_tokens": max_tokens,
     }
-    if require_json:
-        payload["response_format"] = {"type": "json_object"}
+    # 说明：有意不传 response_format（json_object）
+    # 本地 LLM 后端（Ollama/llama.cpp/vLLM 等）不支持该参数，会返回 400
+    # 输出格式靠 system prompt 中"只输出合法 JSON"约束
     if not use_anthropic and _should_disable_deepseek_thinking(base_url, request_model):
         payload["thinking"] = {"type": "disabled"}
     request_url = anthropic_messages_url(base_url) if use_anthropic else openai_chat_completions_url(base_url)