dflash-console 0.3.241__tar.gz → 0.3.249__tar.gz

{dflash_console-0.3.241 → dflash_console-0.3.249}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: dflash-console
-Version: 0.3.241
+Version: 0.3.249
 Summary: Local control panel and terminal CLI for DFlash stacks and local model runtimes
 Project-URL: Homepage, https://github.com/ilan4ever/Dflash-Console
 Project-URL: Repository, https://github.com/ilan4ever/Dflash-Console
@@ -674,7 +674,7 @@ from one UI, then talk to them through a single OpenAI-compatible port.
 > supported platform**; Linux and macOS are not supported or tested (see
 > [Platform support](#platform-support) below).
 
-**Developer:** ILAN AVIV · **UI:** [http://127.0.0.1:8900/](http://127.0.0.1:8900/) · **Version:** v0.3.241
+**Developer:** ILAN AVIV · **UI:** [http://127.0.0.1:8900/](http://127.0.0.1:8900/) · **Version:** v0.3.249
 
 ## Download (Windows)
 
@@ -753,8 +753,12 @@ Typical first session:
 3. For a DFlash GGUF, right-click and **Find and attach draft** if you want speculative decoding.
 4. Chat in the **Playground**, or point any OpenAI client at `http://127.0.0.1:8001/v1` with optional `X-DFlash-Client: YourApp` so **Engines** shows who is using each model.
 
-### Recent improvements (v0.3.241)
+### Recent improvements (v0.3.249)
 
+- **Model library** — loading clears when the model is on the GPU; Engine and GPU column headers stay visible
+- **Model catalog** — a pasted repo name with a trailing period still opens the real model, and shared shard names are no longer marked installed
+- **Image downloads** — a pipeline folder no longer nests a second copy of the whole model inside the text encoder
+- **Desktop app** — a second launch opens the window that is already running
 - **Engine standby** — Running toggle gates load/chat until you arm the pipeline
 - **External GPU cards** — OneVoice, LM Studio, and other apps on the GPU; compact mobile layout; loading state expires when models are ready
 - **Other GPU processes** — per-process VRAM chips for non-model apps

{dflash_console-0.3.241 → dflash_console-0.3.249}/README.md

@@ -8,7 +8,7 @@ from one UI, then talk to them through a single OpenAI-compatible port.
 > supported platform**; Linux and macOS are not supported or tested (see
 > [Platform support](#platform-support) below).
 
-**Developer:** ILAN AVIV · **UI:** [http://127.0.0.1:8900/](http://127.0.0.1:8900/) · **Version:** v0.3.241
+**Developer:** ILAN AVIV · **UI:** [http://127.0.0.1:8900/](http://127.0.0.1:8900/) · **Version:** v0.3.249
 
 ## Download (Windows)
 
@@ -87,8 +87,12 @@ Typical first session:
 3. For a DFlash GGUF, right-click and **Find and attach draft** if you want speculative decoding.
 4. Chat in the **Playground**, or point any OpenAI client at `http://127.0.0.1:8001/v1` with optional `X-DFlash-Client: YourApp` so **Engines** shows who is using each model.
 
-### Recent improvements (v0.3.241)
+### Recent improvements (v0.3.249)
 
+- **Model library** — loading clears when the model is on the GPU; Engine and GPU column headers stay visible
+- **Model catalog** — a pasted repo name with a trailing period still opens the real model, and shared shard names are no longer marked installed
+- **Image downloads** — a pipeline folder no longer nests a second copy of the whole model inside the text encoder
+- **Desktop app** — a second launch opens the window that is already running
 - **Engine standby** — Running toggle gates load/chat until you arm the pipeline
 - **External GPU cards** — OneVoice, LM Studio, and other apps on the GPU; compact mobile layout; loading state expires when models are ready
 - **Other GPU processes** — per-process VRAM chips for non-model apps

{dflash_console-0.3.241 → dflash_console-0.3.249}/api/app.py

@@ -23,7 +23,7 @@ from pydantic import BaseModel, Field
 from core.config import PACKAGE_ROOT, ROOT, get_dflash_root, get_server, is_embedding_server, list_runtimes, list_servers, load_config, normalize_download_settings, normalize_hardware_settings, normalize_inference_settings, normalize_load_settings, normalize_model_libraries, normalize_remote_nodes, normalize_runtime, normalize_server, normalize_ui_layout, save_config, suggest_server_port, update_server_runtime
 from core.version import APP_VERSION
 from core.model_paths import allowed_model_roots, disk_scan_roots, get_models_root, validate_model_path
-from core.gpu_devices import get_gpu_devices_payload
+from core.gpu_devices import get_gpu_devices_payload, validate_gpu_device
 from core.local_models import (
     friendly_model_dir_label,
     invalidate_model_catalog_cache,
@@ -302,6 +302,17 @@ def _request_client_label(request: Request | None) -> str:
     return resolve_client_label(request)
 
 
+def _request_gpu_device(raw: str | None) -> str | None:
+    """Validate an optional per-load GPU override."""
+    raw = getattr(raw, 'default', raw)
+    if raw is None:
+        return None
+    try:
+        return validate_gpu_device(raw)
+    except ValueError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
+
+
 def _raise_if_pipeline_standby(cfg: dict[str, Any]) -> None:
     from core.engine_state import console_pipeline_active, engine_standby_http_error
 
@@ -729,6 +740,7 @@ class ModelLoadRequest(BaseModel):
     model_id: str | None = None
     runtime_id: str | None = None
     context_size: int | None = Field(default=None, ge=2048, le=1048576)
+    gpu_device: str | None = None
     load_settings: dict[str, Any] | None = None
     inference_settings: dict[str, Any] | None = None
 
@@ -754,6 +766,7 @@ class ServerLoadRequest(BaseModel):
     inference_settings: dict[str, Any] | None = None
     model_path: str | None = None
     model_id: str | None = None
+    gpu_device: str | None = None
 
 
 class GpuProcessUnload(BaseModel):
@@ -805,6 +818,7 @@ class RuntimeLoadRequest(BaseModel):
     voice: str = Field(default='', max_length=120)
     path: str = Field(default='', max_length=1024)
     preset: str = Field(default='', max_length=40)
+    gpu_device: str | None = None
     load_settings: dict[str, Any] | None = None
 
 
@@ -1976,17 +1990,20 @@ def model_load(body: ModelLoadRequest, request: Request) -> dict[str, Any]:
 
     cfg = load_config()
     _ensure_api_pipeline_armed(cfg, server_id=body.server_id)
+    gpu_device = _request_gpu_device(body.gpu_device)
     result = execute_catalog_load(
         path=body.path,
         model_id=body.model_id,
         server_id=body.server_id,
         context_size=body.context_size,
+        gpu_device=gpu_device,
         load_settings=body.load_settings,
         inference_settings=body.inference_settings,
         requested_runtime_id=body.runtime_id,
         loaded_by=_request_client_label(request),
         cfg=cfg,
     )
+    result.setdefault('gpu_device', gpu_device or 'auto')
     _invalidate_status_cache()
     return result
 
@@ -2435,6 +2452,9 @@ def runtime_load(runtime_id: str, body: RuntimeLoadRequest) -> dict[str, Any]:
     if not callable(load_fn):
         raise HTTPException(status_code=400, detail='adapter does not support load')
     payload: dict[str, Any] = {'id': body.voice, 'path': body.path}
+    gpu_device = _request_gpu_device(body.gpu_device)
+    if gpu_device is not None:
+        payload['gpu_device'] = gpu_device
     if body.preset:
         payload['preset'] = body.preset
     if body.load_settings:
@@ -2442,6 +2462,7 @@ def runtime_load(runtime_id: str, body: RuntimeLoadRequest) -> dict[str, Any]:
     result = load_fn(payload)
     if not result.get('success'):
         raise HTTPException(status_code=400, detail=result.get('error') or 'load failed')
+    result.setdefault('gpu_device', gpu_device or 'auto')
     return {'success': True, 'runtime_id': runtime_id, **result}
 
 
@@ -2601,7 +2622,10 @@ def patch_server(server_id: str, body: ServerPatch) -> dict[str, Any]:
     for idx, entry in enumerate(servers):
         if not isinstance(entry, dict) or str(entry.get('id') or '') != server_id:
             continue
-        merged = normalize_server({**entry, **body.model_dump(exclude_none=True), 'id': server_id})
+        patch = body.model_dump(exclude_none=True)
+        if 'gpu_device' in patch:
+            patch['gpu_device'] = _request_gpu_device(patch['gpu_device'])
+        merged = normalize_server({**entry, **patch, 'id': server_id})
         servers[idx] = merged
         found = True
         break
@@ -3088,6 +3112,37 @@ def _ensure_server_ready_for_chat(
             'ready_for_chat': bool(loaded),
         }
 
+    # Already-loaded engines answer /v1/models in a fraction of a second.
+    # The full status build checks GPU, logs, and the Windows process list,
+    # which was adding many seconds in front of every chat.
+    host = str(server.get('host') or '127.0.0.1').strip() or '127.0.0.1'
+    port = int(server.get('port') or 0)
+    api_url = str(server.get('api_url') or '').strip()
+    if port > 0 and tcp_port_open(host, port):
+        if not api_url:
+            api_url = f'http://{host}:{port}/v1'
+        from core.runtime import probe_runtime_state
+
+        loaded_now, _loading_now, _router_now, _progress_now = probe_runtime_state(api_url)
+        if loaded_now:
+            needs_grow = False
+            if required_context and cfg.get('context_auto_grow') is not False:
+                loaded_ctx = _loaded_per_slot_context(server) or _configured_per_slot(server)
+                if loaded_ctx and int(required_context) > int(loaded_ctx):
+                    needs_grow = True
+            if not needs_grow:
+                note_engine_active_client(server_id, client_label=client_label)
+                return {
+                    **server,
+                    'running': True,
+                    'status': 'loaded',
+                    'booting': False,
+                    'load_progress': None,
+                    'loaded_models': loaded_now,
+                    'active_model_id': loaded_now[0],
+                    'ready_for_chat': True,
+                }
+
     live = build_server_status(server, cfg=cfg)
 
     # Another chat is already mid-flight on this engine: never JIT-load (that
@@ -3250,6 +3305,7 @@ def server_load_plan(
     server_id: str,
     model_path: str = Query(default=''),
     model_id: str = Query(default=''),
+    gpu_device: str | None = Query(default=None),
 ) -> dict[str, Any]:
     from core.memory_guardrails import assess_load
     from core.server_boot import (
@@ -3262,6 +3318,9 @@ def server_load_plan(
     cfg = load_config()
     server = _require_server(cfg, server_id)
     candidate = dict(server)
+    requested_gpu = _request_gpu_device(gpu_device)
+    if requested_gpu is not None:
+        candidate['gpu_device'] = requested_gpu
     if model_path.strip():
         candidate['adhoc_model_path'] = model_path.strip()
     if model_id.strip():
@@ -3338,6 +3397,8 @@ def server_load(server_id: str, request: Request, body: ServerLoadRequest | None
         patch = body.model_dump(exclude_none=True)
         model_path = patch.pop('model_path', None)
         model_id = patch.pop('model_id', None)
+        if 'gpu_device' in patch:
+            patch['gpu_device'] = _request_gpu_device(patch['gpu_device'])
         if patch:
             server = _persist_server_merge(cfg, server_id, patch)
     if model_path:
@@ -3354,31 +3415,27 @@ def server_load(server_id: str, request: Request, body: ServerLoadRequest | None
         note_engine_loaded(server_id, loaded_by=_request_client_label(request))
         _invalidate_status_cache()
         return already
-    elsewhere = find_target_loaded_elsewhere(
-        server,
-        cfg=cfg,
-        model_path=model_path,
-        exclude_server_id=server_id,
-    )
-    if elsewhere:
-        raise HTTPException(status_code=409, detail=duplicate_load_detail(elsewhere))
-    check = assess_load(server, cfg=cfg)
-    if check.get('level') == 'block':
-        raise HTTPException(status_code=400, detail=_vram_load_block_detail(check))
     if _auto_stop_other_servers(cfg, server_id):
         _invalidate_status_cache()
         cfg = load_config()
         server = _require_server(cfg, server_id)
-        check = assess_load(server, cfg=cfg)
-        if check.get('level') == 'block':
-            raise HTTPException(status_code=400, detail=_vram_load_block_detail(check))
+        if model_path:
+            server = {**server, 'adhoc_model_path': model_path}
+    # Single VRAM assess here; elsewhere/duplicate checks run inside
+    # load_server_checkpoint under the load lock (avoid a second full scan).
+    check = assess_load(server, cfg=cfg)
+    if check.get('level') == 'block':
+        raise HTTPException(status_code=400, detail=_vram_load_block_detail(check))
     result = load_server_checkpoint(
         server,
         cfg=cfg,
         model_path=model_path,
         model_id=model_id,
+        skip_memory_check=True,
     )
     if not result.get('success'):
+        if result.get('already_loaded_elsewhere'):
+            raise HTTPException(status_code=409, detail=duplicate_load_detail(result))
         raise HTTPException(
             status_code=409 if result.get('repair') else 400,
             detail=result,
@@ -3387,6 +3444,7 @@ def server_load(server_id: str, request: Request, body: ServerLoadRequest | None
     _invalidate_status_cache()
     if check.get('level') == 'warn' and check.get('message'):
         result['memory_warning'] = check['message']
+    result.setdefault('gpu_device', str(server.get('gpu_device') or 'auto'))
     return result