feat(qoder): refresh model catalog, add capability mapping and image pass-through

- Registry/constants: drop qmodel_preview/gm51model, add lite, qmodel_38max (Qwen3.8-Max), qfmodel (Qwen3.8-Flash), gmodel (GLM-5.3), gfmodel (GLM-5.3-Flash) - capabilities: add PROVIDER_CAPABILITIES['qoder'] so opaque internal ids resolve to their real models' context windows and limits - executor: preserve image blocks instead of flattening away, convert Claude-style image blocks, and hash images into chat_record_id - tests: cover image preservation, data-URI and Claude-block conversion - build(docker): use CN mirrors for apk and npm
2026-09-03 23:01:53 +07:00
parent c08efdbe2b
commit 2ab6a4c949
6 changed files with 176 additions and 11 deletions
--- a/5
+++ b/5
@@ -2,14 +2,15 @@
 ARG NODE_IMAGE=node:22-alpine
 FROM ${NODE_IMAGE} AS base
 WORKDIR /app
+# CN mirror for apk (used by builder and runner stages)
+RUN sed -i 's|dl-cdn.alpinelinux.org|mirrors.aliyun.com|g' /etc/apk/repositories

 FROM base AS builder

 RUN apk --no-cache upgrade && apk --no-cache add python3 make g++ linux-headers

 COPY package.json ./
-RUN --mount=type=cache,target=/root/.npm \
-  npm install
+RUN npm install --registry=https://registry.npmmirror.com

 COPY . ./
 ENV NEXT_TELEMETRY_DISABLED=1
--- a/open-sse/executors/qoder.js
+++ b/open-sse/executors/qoder.js
@@ -37,10 +37,13 @@ import {
  QODER_MODEL_MAP,
 } from "../shared/qoder/constants.js";
 import { getQoderModelConfig, resolveQoderModels, isQoderPat, resolveQoderCredentials } from "../services/qoderModels.js";
+import { OPENAI_BLOCK, CLAUDE_BLOCK } from "../translator/schema/blocks.js";
+import { encodeDataUri } from "../translator/concerns/image.js";

 /**
 * Hoist role:"system" messages out of the messages array (Qoder rejects
- * system in messages) and flatten any multipart content arrays.
+ * system in messages) and flatten multipart content arrays — EXCEPT image
+ * blocks, which are preserved (see normalizeContent).
 */
 function normalizeMessages(messages) {
  if (!Array.isArray(messages) || messages.length === 0) {
@@ -50,18 +53,72 @@ function normalizeMessages(messages) {
  const out = [];
  for (const msg of messages) {
    if (!msg || typeof msg !== "object") continue;
-    const text = extractText(msg.content);
    if (msg.role === "system") {
+      const text = extractText(msg.content);
      if (text) systemParts.push(text);
      continue;
    }
    const cloned = { ...msg };
-    cloned.content = text;
+    cloned.content = normalizeContent(msg.content);
    out.push(cloned);
  }
  return { messages: out, systemText: systemParts.join("\n\n") };
 }

+/**
+ * Normalize one message's content for Qoder.
+ *
+ * Text-only content is flattened to a plain string (Qoder's historical
+ * shape). When images are present the content stays an array and image
+ * blocks are kept as OpenAI-style `image_url` parts — verified against the
+ * upstream: it accepts both http(s) URLs and inline base64 data: URIs
+ * directly, no pre-upload to the /image/upload OSS flow required (that is
+ * a qodercli client-side choice, not a protocol requirement). The legacy
+ * top-level `image_urls` / `chat_context.imageUrls` slots stay null —
+ * qodercli leaves them null too.
+ *
+ * Claude-style `{type:"image", source:{...}}` blocks are converted to
+ * `image_url` so claude-format clients also round-trip.
+ */
+function normalizeContent(content) {
+  if (typeof content === "string") return content;
+  if (content == null) return "";
+  if (!Array.isArray(content)) return String(content);
+
+  const blocks = [];
+  const textParts = [];
+  let hasImage = false;
+  for (const item of content) {
+    if (!item || typeof item !== "object") continue;
+    if (item.type === OPENAI_BLOCK.IMAGE_URL && typeof item.image_url?.url === "string" && item.image_url.url) {
+      blocks.push({ type: OPENAI_BLOCK.IMAGE_URL, image_url: { url: item.image_url.url } });
+      hasImage = true;
+    } else if (item.type === CLAUDE_BLOCK.IMAGE && item.source) {
+      // Claude base64/url image → OpenAI image_url equivalent.
+      const src = item.source;
+      const url = src.type === "base64" && src.data
+        ? encodeDataUri(src.media_type || "image/png", src.data)
+        : typeof src.url === "string" && src.url ? src.url : null;
+      if (url) {
+        blocks.push({ type: OPENAI_BLOCK.IMAGE_URL, image_url: { url } });
+        hasImage = true;
+      }
+    } else if (typeof item.text === "string" && item.text) {
+      if (hasImage || blocks.length) {
+        // Keep ordering faithful once images are in play.
+        blocks.push({ type: OPENAI_BLOCK.TEXT, text: item.text });
+      } else {
+        textParts.push(item.text);
+      }
+    }
+  }
+
+  if (!hasImage) return textParts.join("\n");
+  // Prepend any text collected before the first image block.
+  if (textParts.length) blocks.unshift({ type: OPENAI_BLOCK.TEXT, text: textParts.join("\n") });
+  return blocks;
+}
+
 function extractText(content) {
  if (typeof content === "string") return content;
  if (content == null) return "";
@@ -84,9 +141,9 @@ function extractText(content) {
 function lastUserText(messages) {
  for (let i = messages.length - 1; i >= 0; i--) {
    const m = messages[i];
-    if (m?.role === "user" && typeof m.content === "string") {
-      return m.content;
-    }
+    if (m?.role !== "user") continue;
+    if (typeof m.content === "string") return m.content;
+    if (Array.isArray(m.content)) return extractText(m.content);
  }
  return "";
 }
@@ -110,6 +167,11 @@ function stableChatRecordId(model, messages, tools, maxTokens) {
    if (m.role) { h.update("\0"); h.update(m.role); }
    if (typeof m.content === "string" && m.content) {
      h.update("\0"); h.update(m.content);
+    } else if (Array.isArray(m.content)) {
+      // Include image refs so the same prompt with a different image gets
+      // a distinct chat_record_id.
+      h.update("\0");
+      try { h.update(JSON.stringify(m.content)); } catch {}
    }
  }
  if (tools) {
--- a/open-sse/providers/capabilities.js
+++ b/open-sse/providers/capabilities.js
@@ -206,6 +206,38 @@ export const PROVIDER_CAPABILITIES = {
    "deepseek-v4-flash":  { vision: true, reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 50000 },
    "deepseek-v3-2-volc": { reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 96000, maxOutput: 32000 },
  },
+  // Qoder — upstream exposes opaque internal ids (dfmodel, kmodel, …); the
+  // registry `name` is display-only and capability lookup matches on the raw
+  // id, so every qoder model would fall through to DEFAULT_CAPABILITIES
+  // (200K) without this map. contextWindow follows the real model family's
+  // spec: the /algo/api/v2/model/list max_input_tokens under-reports some
+  // windows (GLM-5.3 / Kimi-K3 / Qwen3.8-Max claim 180K but accept more).
+  // max_output_tokens arrives as 0 for every model, so outputs are
+  // best-guess from the real model family. Vision tags below follow the
+  // upstream is_vl flag per explicit request, even though the executor
+  // currently sends image_urls:null (image pass-through over the agent_chat
+  // SSE protocol is unverified). reasoning:true on all of them — every model can
+  // reason; the upstream is_reasoning flag only drives model_config selection.
+  // thinkingFormat keeps the true-model family for documentation/UI, but
+  // thinkingCanDisable:false everywhere: the executor only forwards
+  // messages/tools/max_tokens, and thinking is fixed upstream via
+  // modelConfig.is_reasoning — client thinking intent is dropped, so "none"
+  // must never be offered as an option.
+  "qoder": {
+    "ultimate":       { vision: true, reasoning: true, thinkingFormat: "claude-adaptive", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 128000 }, // Claude Opus 5
+    "performance":    { vision: true, reasoning: true, thinkingFormat: "claude-adaptive", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 128000 }, // Claude Sonnet 5
+    "dmodel":         { reasoning: true, thinkingFormat: "deepseek", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },  // DeepSeek-V4-Pro
+    "dfmodel":        { reasoning: true, thinkingFormat: "deepseek", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },  // DeepSeek-V4-Flash
+    "gmodel":         { reasoning: true, thinkingFormat: "zai", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 128000 },      // GLM-5.3
+    "gfmodel":        { vision: true, reasoning: true, thinkingFormat: "zai", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 128000 }, // GLM-5.3-Flash
+    "kmodel_latest":  { vision: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },      // Kimi-K3
+    "kmodel":         { vision: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 256000, maxOutput: 65536 },  // Kimi-K2.7-Code
+    "mmodel":         { reasoning: true, thinkingFormat: "minimax", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 512000 }, // MiniMax-M3
+    "qmodel_latest":  { vision: true, reasoning: true, thinkingFormat: "qwen", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },  // Qwen3.7-Max
+    "qmodel":         { vision: true, reasoning: true, thinkingFormat: "qwen", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },  // Qwen3.7-Plus
+    "qfmodel":        { vision: true, reasoning: true, thinkingFormat: "qwen", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },  // Qwen3.8-Flash
+    "qmodel_38max":   { vision: true, reasoning: true, thinkingFormat: "qwen", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 65536 },      // Qwen3.8-Max
+  },
  // Poolside Laguna — OpenAI-compatible, all reasoning-capable (32K max output).
  "poolside": {
    "laguna-s-2.1":  { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 32000 },
--- a/open-sse/providers/registry/qoder.js
+++ b/open-sse/providers/registry/qoder.js
@@ -30,12 +30,15 @@ export default {
    { id: "auto", name: "Auto" },
    { id: "performance", name: "Performance" },
    { id: "efficient", name: "Efficient" },
-    { id: "qmodel_preview", name: "Qwen3.8-Max-Preview" },
+    { id: "lite", name: "Lite" },
+    { id: "qmodel_38max", name: "Qwen3.8-Max" },
    { id: "qmodel_latest", name: "Qwen3.7-Max" },
    { id: "qmodel", name: "Qwen3.7-Plus" },
+    { id: "qfmodel", name: "Qwen3.8-Flash" },
    { id: "kmodel_latest", name: "Kimi-K3" },
    { id: "kmodel", name: "Kimi-K2.7-Code" },
-    { id: "gm51model", name: "GLM-5.2" },
+    { id: "gmodel", name: "GLM-5.3" },
+    { id: "gfmodel", name: "GLM-5.3-Flash" },
    { id: "dmodel", name: "DeepSeek-V4-Pro" },
    { id: "dfmodel", name: "DeepSeek-V4-Flash" },
    { id: "mmodel", name: "MiniMax-M3" },
--- a/open-sse/shared/qoder/constants.js
+++ b/open-sse/shared/qoder/constants.js
@@ -54,10 +54,13 @@ export const QODER_MODEL_MAP = {
  lite: "lite",
  // Frontier models
  qmodel: "qmodel",
+  qfmodel: "qfmodel",
  qmodel_latest: "qmodel_latest",
+  qmodel_38max: "qmodel_38max",
  dmodel: "dmodel",
  dfmodel: "dfmodel",
-  gm51model: "gm51model",
+  gmodel: "gmodel",
+  gfmodel: "gfmodel",
  kmodel: "kmodel",
  mmodel: "mmodel",
 };
--- a/tests/unit/qoder.test.js
+++ b/tests/unit/qoder.test.js
@@ -367,6 +367,70 @@ describe("normalizeMessages", () => {
    expect(result.messages).toEqual([]);
    expect(result.systemText).toBe("");
  });
+
+  it("preserves image_url blocks (http URL) instead of dropping them", () => {
+    const result = normalizeMessages([
+      {
+        role: "user",
+        content: [
+          { type: "text", text: "describe" },
+          { type: "image_url", image_url: { url: "https://example.com/a.png" } },
+        ],
+      },
+    ]);
+    const content = result.messages[0].content;
+    expect(Array.isArray(content)).toBe(true);
+    expect(content).toContainEqual({ type: "text", text: "describe" });
+    expect(content).toContainEqual({ type: "image_url", image_url: { url: "https://example.com/a.png" } });
+  });
+
+  it("preserves base64 data: URI images (no OSS upload needed)", () => {
+    const dataUri = "data:image/png;base64,iVBORw0KGgo=";
+    const result = normalizeMessages([
+      {
+        role: "user",
+        content: [
+          { type: "image_url", image_url: { url: dataUri } },
+          { type: "text", text: "what color?" },
+        ],
+      },
+    ]);
+    const content = result.messages[0].content;
+    expect(Array.isArray(content)).toBe(true);
+    expect(content[0]).toEqual({ type: "image_url", image_url: { url: dataUri } });
+    expect(content.some((b) => b.type === "text" && b.text === "what color?")).toBe(true);
+  });
+
+  it("converts claude-style base64 image blocks to image_url data URIs", () => {
+    const result = normalizeMessages([
+      {
+        role: "user",
+        content: [
+          { type: "text", text: "see this" },
+          { type: "image", source: { type: "base64", media_type: "image/jpeg", data: "AAAA" } },
+        ],
+      },
+    ]);
+    const content = result.messages[0].content;
+    expect(content).toContainEqual({
+      type: "image_url",
+      image_url: { url: "data:image/jpeg;base64,AAAA" },
+    });
+  });
+
+  it("drops image blocks with no usable url but keeps the text", () => {
+    const result = normalizeMessages([
+      {
+        role: "user",
+        content: [
+          { type: "text", text: "hi" },
+          { type: "image_url", image_url: {} },
+          { type: "image", source: { type: "base64" } },
+        ],
+      },
+    ]);
+    expect(result.messages[0].content).toBe("hi");
+  });
 });

 describe("wrapQoderSSE", () => {