fix(models): platform-aware activation + download cancel

dream-server/extensions/services/dashboard-api/routers/models.py

@@ -214,6 +214,13 @@ def download_model(model_id: str, api_key: str = Depends(verify_api_key)):
     return result
 
 
+@router.post("/api/models/download/cancel")
+def cancel_download(api_key: str = Depends(verify_api_key)):
+    """Cancel an in-progress model download."""
+    result = _call_agent_model("/v1/model/download/cancel", {})
+    return result
+
+
 @router.post("/api/models/{model_id}/load")
 def load_model(model_id: str, api_key: str = Depends(verify_api_key)):
     """Activate a model — update config and restart llama-server."""

dream-server/extensions/services/dashboard-api/tests/test_host_agent.py

@@ -449,3 +449,44 @@ def test_500_missing_schema(self, env_update_env):
 
         assert handler.response_code == 500
         assert ".env.schema.json not found" in handler.parse_response()["error"]
+
+
+class TestHandleModelDownloadCancel:
+
+    def test_returns_no_download_when_idle(self, monkeypatch):
+        handler = _FakeHandler(b"")
+        monkeypatch.setattr(_mod, "AGENT_API_KEY", "test-key")
+        monkeypatch.setattr(_mod, "_model_download_thread", None)
+        _mod._model_download_cancel.clear()
+
+        _mod.AgentHandler._handle_model_download_cancel(handler)
+
+        assert handler.response_code == 200
+        assert handler.parse_response()["status"] == "no_download"
+        assert _mod._model_download_cancel.is_set() is False
+
+    def test_sets_cancel_flag_and_kills_active_proc(self, monkeypatch):
+        class _AliveThread:
+            def is_alive(self):
+                return True
+
+        class _FakeProc:
+            def __init__(self):
+                self.killed = False
+
+            def kill(self):
+                self.killed = True
+
+        handler = _FakeHandler(b"")
+        proc = _FakeProc()
+        monkeypatch.setattr(_mod, "AGENT_API_KEY", "test-key")
+        monkeypatch.setattr(_mod, "_model_download_thread", _AliveThread())
+        monkeypatch.setattr(_mod, "_model_download_proc", proc)
+        _mod._model_download_cancel.clear()
+
+        _mod.AgentHandler._handle_model_download_cancel(handler)
+
+        assert handler.response_code == 200
+        assert handler.parse_response()["status"] == "cancelling"
+        assert _mod._model_download_cancel.is_set() is True
+        assert proc.killed is True

dream-server/extensions/services/dashboard-api/tests/test_model_activate.py

@@ -16,6 +16,8 @@
 _check_lemonade_health = _mod._check_lemonade_health
 _send_lemonade_warmup = _mod._send_lemonade_warmup
 _write_lemonade_config = _mod._write_lemonade_config
+_compose_restart_llama_server = _mod._compose_restart_llama_server
+_launch_native_llama_server = _mod._launch_native_llama_server
 
 
 # --- _check_lemonade_health ---
@@ -136,6 +138,78 @@ def test_file_path(self, tmp_path):
         assert (litellm_dir / "lemonade.yaml").exists()
 
 
+class TestComposeRestartLlamaServer:
+
+    def test_amd_uses_stop_then_up(self, monkeypatch, tmp_path):
+        calls = []
+
+        def fake_run(cmd, **kwargs):
+            calls.append(cmd)
+            return subprocess.CompletedProcess(cmd, 0, stdout="", stderr="")
+
+        monkeypatch.setattr(_mod, "INSTALL_DIR", tmp_path)
+        monkeypatch.setattr(
+            _mod,
+            "resolve_compose_flags",
+            lambda: ["--env-file", ".env", "-f", "docker-compose.base.yml"],
+        )
+        monkeypatch.setattr(subprocess, "run", fake_run)
+
+        _compose_restart_llama_server({"GPU_BACKEND": "amd"})
+
+        assert calls == [
+            [
+                "docker", "compose", "--env-file", ".env", "-f",
+                "docker-compose.base.yml", "stop", "llama-server",
+            ],
+            [
+                "docker", "compose", "--env-file", ".env", "-f",
+                "docker-compose.base.yml", "up", "-d", "llama-server",
+            ],
+        ]
+
+
+class TestLaunchNativeLlamaServer:
+
+    def test_reads_env_and_writes_pid(self, monkeypatch, tmp_path):
+        env_path = tmp_path / ".env"
+        env_path.write_text(
+            "GGUF_FILE=test-model.gguf\nCTX_SIZE=8192\nLLAMA_REASONING=on\n",
+            encoding="utf-8",
+        )
+        (tmp_path / "data" / "models").mkdir(parents=True)
+        (tmp_path / "data").mkdir(exist_ok=True)
+        llama_bin = tmp_path / "bin" / "llama-server"
+        llama_bin.parent.mkdir(parents=True)
+        llama_bin.write_text("", encoding="utf-8")
+        llama_log = tmp_path / "data" / "llama-server.log"
+        pid_file = tmp_path / "data" / ".llama-server.pid"
+
+        calls = []
+
+        class _FakeProc:
+            pid = 4321
+
+        def fake_popen(cmd, **kwargs):
+            calls.append((cmd, kwargs))
+            return _FakeProc()
+
+        monkeypatch.setattr(_mod, "INSTALL_DIR", tmp_path)
+        monkeypatch.setattr(subprocess, "Popen", fake_popen)
+
+        _launch_native_llama_server(env_path, llama_bin, llama_log, pid_file)
+
+        assert pid_file.read_text(encoding="utf-8") == "4321"
+        cmd, _kwargs = calls[0]
+        assert cmd[0] == str(llama_bin)
+        assert "--model" in cmd
+        assert str(tmp_path / "data" / "models" / "test-model.gguf") in cmd
+        assert "--ctx-size" in cmd
+        assert "8192" in cmd
+        assert "--reasoning-format" in cmd
+        assert "deepseek" in cmd
+
+
 # --- Rollback integration ---
 
 

dream-server/extensions/services/dashboard/src/hooks/useDownloadProgress.js

@@ -34,14 +34,14 @@
       } else if (data.status === 'complete' || data.status === 'idle') {
         setIsDownloading(false)
         setProgress(null)
-      } else if (data.status === 'failed' || data.status === 'error') {
+      } else if (data.status === 'failed' || data.status === 'error' || data.status === 'cancelled') {
         setIsDownloading(false)
         setProgress({
-          error: data.error || data.message || 'Download failed',
+          error: data.error || data.message || (data.status === 'cancelled' ? 'Download cancelled' : 'Download failed'),
           model: data.model
         })
       }
     } catch (err) {
       // Silently fail - API might not be available
     }
   }, [])
@@ -76,11 +76,21 @@
     return eta
   }
 
+  const cancelDownload = useCallback(async () => {
+    try {
+      await fetch('/api/models/download/cancel', { method: 'POST' })
+      fetchProgress()
+    } catch (err) {
+      console.error('Failed to cancel download:', err)
+    }
+  }, [fetchProgress])
+
   return {
     isDownloading,
     progress,
     formatBytes,
     formatEta,
-    refresh: fetchProgress
+    refresh: fetchProgress,
+    cancelDownload
   }
 }

dream-server/extensions/services/dashboard/src/pages/Models.jsx

@@ -319,9 +319,19 @@ function DownloadProgressBar({ progress, helpers }) {
             </p>
           </div>
         </div>
-        <span className="text-lg font-bold text-theme-accent">
-          {progress.percent?.toFixed(0) || 0}%
-        </span>
+        <div className="flex items-center gap-3">
+          <span className="text-lg font-bold text-theme-accent">
+            {progress.percent?.toFixed(0) || 0}%
+          </span>
+          {helpers.cancelDownload && (
+            <button
+              onClick={helpers.cancelDownload}
+              className="px-3 py-1 text-xs font-medium text-red-400 border border-red-400/30 rounded-lg hover:bg-red-400/10 transition-colors"
+            >
+              Cancel
+            </button>
+          )}
+        </div>
       </div>
 
       <div className="h-3 bg-theme-border rounded-full overflow-hidden">