From 47c301c35f28a26b3c1c26c8f4623c462e4820e7 Mon Sep 17 00:00:00 2001
From: wuyang <5700876+banisherwy@user.noreply.gitee.com>
Date: Wed, 8 Jul 2026 16:50:08 +0800
Subject: [PATCH] Stop overriding Ollama context size

---
 web/README.md |  1 +
 web/app.py    | 11 +++++------
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/web/README.md b/web/README.md
index d89b9a4..7801ac8 100644
--- a/web/README.md
+++ b/web/README.md
@@ -38,6 +38,7 @@ https://lab.k1412.top/
 - Search Agent 查询理解：`ChatGPT-5.6:light`
 - Search Agent 结果简报、搜索追问、论文对话：`ChatGPT-5.6:fast`
 - Ollama 请求单并发执行，并缓存结果到 `web/cache/ai/`
+- 应用侧不下发 `num_ctx`，上下文长度由 Ollama 服务端 runner 统一控制
 - arXiv 摘要缓存到 `web/cache/arxiv/`
 
 ## Knowledge Design
diff --git a/web/app.py b/web/app.py
index bf8c98e..bb00663 100644
--- a/web/app.py
+++ b/web/app.py
@@ -1219,19 +1219,18 @@ def clean_model_response(value: str) -> str:
 def call_ollama(model: str, prompt: str, mode: str) -> dict[str, Any]:
     options = {
         "temperature": 0.2,
-        "num_ctx": 4096,
         "num_predict": 520,
     }
     if mode == "translate":
-        options.update({"num_ctx": 3072, "num_predict": 700, "temperature": 0.1})
+        options.update({"num_predict": 700, "temperature": 0.1})
     elif mode == "deep":
-        options.update({"num_ctx": 8192, "num_predict": 900, "temperature": 0.25})
+        options.update({"num_predict": 900, "temperature": 0.25})
     elif mode in {"atlas", "topic", "path", "compare"}:
-        options.update({"num_ctx": 8192, "num_predict": 950, "temperature": 0.22})
+        options.update({"num_predict": 950, "temperature": 0.22})
     elif mode == "search_plan":
-        options.update({"num_ctx": 4096, "num_predict": 420, "temperature": 0.1})
+        options.update({"num_predict": 420, "temperature": 0.1})
     elif mode in {"search_summary", "search_followup", "paper_chat"}:
-        options.update({"num_ctx": 8192, "num_predict": 780, "temperature": 0.18})
+        options.update({"num_predict": 780, "temperature": 0.18})
 
     payload = {
         "model": model,