From 8e00613e693e7a30bf83fca29de6fabd8425a696 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=80=AA=E7=A8=8B=E4=BC=9F?= <ncw1992120@gmail.com>
Date: Wed, 20 May 2026 09:22:06 +0800
Subject: [PATCH] fix(embedding): use chatModel field for protocol routing
 instead of providerId matching (#170)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Closes #166

EmbeddingModelFactory used EmbeddingProtocol.fromProviderId() to pick the
embedding protocol, which substring-matches 'dashscope' / 'qwen' / 'aliyun'
in the providerId. The dashscope-compat provider carries 'dashscope' in its
id but runs in OpenAI compatible mode (chatModel='OpenAIChatModel',
baseUrl='https://dashscope.aliyuncs.com/compatible-mode/v1'). Routing it to
DASHSCOPE_EMBEDDING made DashScopeApi build its native path against the
compat base, producing 404s on every embedding call.

Switch to the chatModel column instead — the same signal ModelProtocol
.fromChatModel() uses for the chat path. chatModel='DashScopeChatModel'
takes the native protocol; everything else (including dashscope-compat)
takes OpenAI-compatible.

EmbeddingProtocol.fromProviderId() is retained for reference but is no
longer called; future callers should follow the chatModel pattern.
---
 .../mate/llm/embedding/EmbeddingModelFactory.java   | 13 ++++++++++---
 .../java/vip/mate/llm/model/EmbeddingProtocol.java  |  4 +++-
 2 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/mateclaw-server/src/main/java/vip/mate/llm/embedding/EmbeddingModelFactory.java b/mateclaw-server/src/main/java/vip/mate/llm/embedding/EmbeddingModelFactory.java
index 4d13b4ca..d47b44a8 100644
--- a/mateclaw-server/src/main/java/vip/mate/llm/embedding/EmbeddingModelFactory.java
+++ b/mateclaw-server/src/main/java/vip/mate/llm/embedding/EmbeddingModelFactory.java
@@ -111,9 +111,16 @@ public class EmbeddingModelFactory {
                     "Embedding provider '" + modelConfig.getProvider() + "' not found in mate_model_provider");
         }
 
-        EmbeddingProtocol protocol = EmbeddingProtocol.fromProviderId(provider.getProviderId());
-        log.info("[EmbeddingFactory] Building embedding model: provider={}, model={}, protocol={}",
-                provider.getProviderId(), modelConfig.getModelName(), protocol);
+        // Use chatModel column (same signal as ModelProtocol.fromChatModel) rather than
+        // providerId substring matching. dashscope-compat has "dashscope" in its id but
+        // uses OpenAIChatModel + compatible-mode URL — routing it to DASHSCOPE_EMBEDDING
+        // causes DashScopeApi to construct a native path that returns 404 against the
+        // compat base URL.
+        EmbeddingProtocol protocol = "DashScopeChatModel".equals(provider.getChatModel())
+                ? EmbeddingProtocol.DASHSCOPE_EMBEDDING
+                : EmbeddingProtocol.OPENAI_EMBEDDING;
+        log.info("[EmbeddingFactory] Building embedding model: provider={}, chatModel={}, model={}, protocol={}",
+                provider.getProviderId(), provider.getChatModel(), modelConfig.getModelName(), protocol);
 
         return switch (protocol) {
             case DASHSCOPE_EMBEDDING -> buildDashScope(provider, modelConfig);
diff --git a/mateclaw-server/src/main/java/vip/mate/llm/model/EmbeddingProtocol.java b/mateclaw-server/src/main/java/vip/mate/llm/model/EmbeddingProtocol.java
index 2a381d10..a6530f88 100644
--- a/mateclaw-server/src/main/java/vip/mate/llm/model/EmbeddingProtocol.java
+++ b/mateclaw-server/src/main/java/vip/mate/llm/model/EmbeddingProtocol.java
@@ -10,7 +10,9 @@ package vip.mate.llm.model;
  *   <li>OpenAI 兼容协议的 embedding 统一走 /v1/embeddings</li>
  * </ul>
  * <p>
- * 通过 {@link #fromProviderId} 从 providerId 推断协议，新增 provider 时只需扩展这里。
+ * <b>注意</b>：{@link EmbeddingModelFactory} 现在优先通过 {@code chatModel} 列判断协议
+ * （与 {@link ModelProtocol#fromChatModel} 保持一致），{@link #fromProviderId} 已不再使用。
+ * 保留该方法仅作参考；不要在新代码中调用它。
  *
  * @author MateClaw Team
  */