Add DuckLM service scripts and utility model benchmark

2026-05-22 07:27:59 +08:00
parent e6b82f0376
commit ff98224eb6
24 changed files with 2492 additions and 74 deletions
@@ -3,7 +3,7 @@ import json
 import logging
 import time
 from pathlib import Path
-from typing import Any
+from typing import Any, Literal

 import uvicorn
 from fastapi import FastAPI, HTTPException, Request
@@ -60,6 +60,7 @@ class ChatRequest(BaseModel):
    conversation_id: str | None = None
    workspace: str | None = None
    debug: bool = False
+    reasoning: Literal["auto", "on", "off"] = "auto"


 class ConversationRequest(BaseModel):
@@ -208,6 +209,7 @@ def create_app() -> FastAPI:
            history_messages=history,
            memory_records=memory_records,
            skill_summary=await selected_skill_summary(body.message),
+            reasoning=body.reasoning,
        )
        await conversations.add_message(
            conversation.conversation_id,
@@ -411,7 +413,12 @@ def create_app() -> FastAPI:
                    "Формирую ответ...",
                )
                await event_store.append(task.task_id, "model_call_started", {"role": "thinker"})
-                async for chunk in model_client.stream_chat("thinker", messages):
+                stream = (
+                    model_client.stream_chat("thinker", messages, reasoning=body.reasoning)
+                    if body.reasoning in {"on", "off"}
+                    else model_client.stream_chat("thinker", messages)
+                )
+                async for chunk in stream:
                    delta = str(chunk.get("delta") or "")
                    if chunk.get("type") == "reasoning_delta":
                        generation_stats.record(delta)
@@ -3,13 +3,15 @@ import logging
 import time
 from dataclasses import dataclass
 from pathlib import Path
-from typing import Any
+from typing import Any, Literal

 import httpx
 import yaml

 logger = logging.getLogger(__name__)

+ReasoningMode = Literal["auto", "on", "off"]
+

@dataclass(frozen=True)
 class RoleConfig:
@@ -92,6 +94,20 @@ class ModelClient:
            }
        return {"type": "json_object"}

+    def _reasoning_options(self, reasoning: ReasoningMode | None) -> dict[str, Any]:
+        if reasoning == "on":
+            return {
+                "reasoning_format": "deepseek",
+                "chat_template_kwargs": {"enable_thinking": True},
+            }
+        if reasoning == "off":
+            return {
+                "reasoning_format": "deepseek",
+                "chat_template_kwargs": {"enable_thinking": False},
+                "thinking_budget_tokens": 0,
+            }
+        return {}
+
    async def chat(
        self,
        role: str,
@@ -99,6 +115,7 @@ class ModelClient:
        temperature: float | None = None,
        max_output_tokens: int | None = None,
        response_format: dict[str, Any] | None = None,
+        reasoning: ReasoningMode | None = None,
    ) -> ModelResponse:
        cfg = self.get_role_config(role)
        outbound = list(messages)
@@ -115,6 +132,7 @@ class ModelClient:
        fmt = self._response_format(cfg, response_format)
        if fmt is not None:
            payload["response_format"] = fmt
+        payload.update(self._reasoning_options(reasoning))

        start = time.perf_counter()
        try:
@@ -150,6 +168,7 @@ class ModelClient:
        temperature: float | None = None,
        max_output_tokens: int | None = None,
        response_format: dict[str, Any] | None = None,
+        reasoning: ReasoningMode | None = None,
    ):
        cfg = self.get_role_config(role)
        outbound = list(messages)
@@ -167,6 +186,7 @@ class ModelClient:
        fmt = self._response_format(cfg, response_format)
        if fmt is not None:
            payload["response_format"] = fmt
+        payload.update(self._reasoning_options(reasoning))

        try:
            async with httpx.AsyncClient(timeout=self.timeout, trust_env=False) as client:
@@ -10,7 +10,7 @@ from duck_core.experience.recorder import ExperienceRecorder
 from duck_core.memory.policy import MemoryPolicy
 from duck_core.memory.store import MemoryStore
 from duck_core.memory.vector_memory import VectorMemory
-from duck_core.model_client import ModelClient
+from duck_core.model_client import ModelClient, ReasoningMode
 from duck_core.reflection import Reflection
 from duck_core.tasks.store import TaskStore
 from duck_core.tools.base import ToolResult
@@ -63,6 +63,7 @@ class RuntimeLoop:
        memory_records: list[dict[str, str]] | None = None,
        skill_summary: str | None = None,
        reflect: bool = True,
+        reasoning: ReasoningMode | None = None,
    ) -> ChatResult:
        task = await self.task_store.create_task(message, workspace, debug)
        await self.event_store.append(
@@ -100,7 +101,10 @@ class RuntimeLoop:
            await self.event_store.append(
                task.task_id, "model_call_started", {"role": "thinker"}
            )
-            response = await self.model_client.chat("thinker", messages)
+            if reasoning in {"on", "off"}:
+                response = await self.model_client.chat("thinker", messages, reasoning=reasoning)
+            else:
+                response = await self.model_client.chat("thinker", messages)
            await self.event_store.append(
                task.task_id,
                "cognition_response",
@@ -16,6 +16,100 @@ function escapeText(value) {
  return String(value ?? "");
 }

+function escapeHtml(value) {
+  return String(value ?? "")
+    .replaceAll("&", "&amp;")
+    .replaceAll("<", "&lt;")
+    .replaceAll(">", "&gt;")
+    .replaceAll('"', "&quot;")
+    .replaceAll("'", "&#39;");
+}
+
+function renderInlineMarkdown(text) {
+  let html = escapeHtml(text);
+  html = html.replace(/`([^`]+)`/g, "<code>$1</code>");
+  html = html.replace(/\*\*([^*]+)\*\*/g, "<strong>$1</strong>");
+  html = html.replace(/\*([^*]+)\*/g, "<em>$1</em>");
+  html = html.replace(
+    /(https?:\/\/[^\s<]+)/g,
+    '<a href="$1" target="_blank" rel="noreferrer">$1</a>',
+  );
+  return html;
+}
+
+function renderMarkdown(markdown) {
+  const blocks = [];
+  const lines = String(markdown ?? "").replace(/\r\n/g, "\n").split("\n");
+  let paragraph = [];
+  let listItems = [];
+  let inFence = false;
+  let fenceLines = [];
+
+  const flushParagraph = () => {
+    if (!paragraph.length) return;
+    blocks.push(`<p>${renderInlineMarkdown(paragraph.join(" "))}</p>`);
+    paragraph = [];
+  };
+  const flushList = () => {
+    if (!listItems.length) return;
+    blocks.push(`<ul>${listItems.map((item) => `<li>${renderInlineMarkdown(item)}</li>`).join("")}</ul>`);
+    listItems = [];
+  };
+
+  for (const line of lines) {
+    if (line.startsWith("```")) {
+      if (inFence) {
+        blocks.push(`<pre><code>${escapeHtml(fenceLines.join("\n"))}</code></pre>`);
+        fenceLines = [];
+        inFence = false;
+      } else {
+        flushParagraph();
+        flushList();
+        inFence = true;
+      }
+      continue;
+    }
+    if (inFence) {
+      fenceLines.push(line);
+      continue;
+    }
+
+    const heading = /^(#{1,4})\s+(.+)$/.exec(line);
+    if (heading) {
+      flushParagraph();
+      flushList();
+      const level = heading[1].length + 2;
+      blocks.push(`<h${level}>${renderInlineMarkdown(heading[2])}</h${level}>`);
+      continue;
+    }
+
+    const listItem = /^\s*[-*]\s+(.+)$/.exec(line);
+    if (listItem) {
+      flushParagraph();
+      listItems.push(listItem[1]);
+      continue;
+    }
+
+    if (!line.trim()) {
+      flushParagraph();
+      flushList();
+      continue;
+    }
+    flushList();
+    paragraph.push(line.trim());
+  }
+  if (inFence) blocks.push(`<pre><code>${escapeHtml(fenceLines.join("\n"))}</code></pre>`);
+  flushParagraph();
+  flushList();
+  return blocks.join("");
+}
+
+function setMarkdownContent(node, content) {
+  if (!node) return;
+  node.dataset.markdown = String(content ?? "");
+  node.innerHTML = renderMarkdown(node.dataset.markdown);
+}
+
 function setStatus(id, text, tone = "neutral") {
  const node = document.querySelector(id);
  if (!node) return;
@@ -67,7 +161,12 @@ function addMessage(role, content, meta = "", options = {}) {
  messageMeta.innerHTML = `<strong>${role === "user" ? "You" : "DuckLM"}</strong><span>${escapeText(meta)}</span>`;

  const text = document.createElement("p");
-  text.textContent = content;
+  text.className = role === "assistant" ? "message-body markdown-body" : "message-body";
+  if (role === "assistant") {
+    setMarkdownContent(text, content);
+  } else {
+    text.textContent = content;
+  }

  bubble.append(messageMeta);
  if (role === "assistant" && options.reasoning) {
@@ -204,11 +303,10 @@ function updateToolTerminal(article, eventPayload) {
  terminal.classList.remove("is-waiting");
  status.textContent = result.ok ? "ok" : "error";

-  const title = terminal.querySelector(".tool-terminal-title")?.textContent || body.textContent.trim();
-  const parts = [title];
-  if (result.output) parts.push("\nstdout\n" + result.output.trimEnd());
-  if (result.error) parts.push("\nstderr\n" + result.error.trimEnd());
-  body.textContent = parts.join("\n");
+  const parts = [];
+  if (result.output) parts.push("stdout\n" + result.output.trimEnd());
+  if (result.error) parts.push("stderr\n" + result.error.trimEnd());
+  body.textContent = parts.join("\n\n") || "completed with no output";
  document.querySelector("#messages").scrollTop = document.querySelector("#messages").scrollHeight;
 }

@@ -359,13 +457,22 @@ function humanApprovalDecision(action) {

 function setMessagePending(article, text) {
  const paragraph = article?.querySelector("p");
-  if (paragraph) paragraph.textContent = text;
+  if (!paragraph) return;
+  if (paragraph.classList.contains("markdown-body")) {
+    setMarkdownContent(paragraph, text);
+  } else {
+    paragraph.textContent = text;
+  }
 }

 function appendMessageText(article, delta) {
  const paragraph = article?.querySelector("p");
  if (!paragraph) return;
-  paragraph.textContent += delta;
+  if (paragraph.classList.contains("markdown-body")) {
+    setMarkdownContent(paragraph, `${paragraph.dataset.markdown || ""}${delta}`);
+  } else {
+    paragraph.textContent += delta;
+  }
  document.querySelector("#messages").scrollTop = document.querySelector("#messages").scrollHeight;
 }

@@ -758,7 +865,8 @@ async function sendMessage() {
  setStatus("#task-status", "running", "warn");
  addMessage("user", message, "submitted");
  input.value = "";
-  const pending = addMessage("assistant", "", "thinking", {reasoning: true});
+  const reasoningEnabled = document.querySelector("#reasoning")?.checked || false;
+  const pending = addMessage("assistant", "", "thinking", {reasoning: reasoningEnabled});
  const context = {taskId: "", contentStarted: false};

  try {
@@ -767,6 +875,7 @@ async function sendMessage() {
      conversation_id: state.currentConversationId || null,
      workspace: document.querySelector("#workspace").value,
      debug: document.querySelector("#debug").checked,
+      reasoning: reasoningEnabled ? "on" : "off",
    }, async ({name, data}) => {
      await handleAssistantStreamEvent(pending, name, data, context);
    });
@@ -500,6 +500,60 @@ dd {
  line-height: 1.5;
 }

+.bubble .markdown-body {
+  white-space: normal;
+}
+
+.markdown-body h3,
+.markdown-body h4,
+.markdown-body h5,
+.markdown-body p,
+.markdown-body ul,
+.markdown-body pre {
+  margin: 8px 0 0;
+}
+
+.markdown-body h3 {
+  font-size: 17px;
+}
+
+.markdown-body h4,
+.markdown-body h5 {
+  font-size: 15px;
+}
+
+.markdown-body ul {
+  padding-left: 22px;
+}
+
+.markdown-body code {
+  padding: 1px 4px;
+  border-radius: 5px;
+  background: #e2e8f0;
+  font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
+  font-size: 0.92em;
+}
+
+.markdown-body pre {
+  max-width: 100%;
+  overflow: auto;
+  padding: 10px 12px;
+  border-radius: 8px;
+  background: #0f172a;
+  color: #d1fae5;
+  white-space: pre;
+}
+
+.markdown-body pre code {
+  padding: 0;
+  background: transparent;
+  color: inherit;
+}
+
+.markdown-body a {
+  color: var(--accent);
+}
+
 .message-reasoning {
  display: grid;
  gap: 8px;
@@ -46,6 +46,10 @@
            <input id="debug" type="checkbox" checked>
            <span>Debug mode</span>
          </label>
+          <label class="toggle-row">
+            <input id="reasoning" type="checkbox">
+            <span>Reasoning</span>
+          </label>
        </section>

        <section class="status-panel" aria-labelledby="status-title">