fix: add frontend web dir and improve AMD GPU detection

.codewhale/instructions.md

@@ -0,0 +1,21 @@
+# Project Context (Auto-generated)
+
+> This file was automatically generated by CodeWhale.
+> You can edit or delete it at any time.
+
+## Bounded Project Overview
+
+```json
+{
+  "project_name": "xpusys",
+  "directory_structure": [],
+  "readme": null,
+  "config_files": [],
+  "key_source_files": [],
+  "counts": {
+    "config_files": 0,
+    "directory_entries": 0,
+    "key_source_files": 0
+  }
+}
+```

.deepseek/state/subagents.v1.json

@@ -0,0 +1,30 @@
+{
+  "schema_version": 1,
+  "agents": [
+    {
+      "id": "agent_81924c2d",
+      "session_name": "clone-fork",
+      "fork_context": false,
+      "workspace": "/home/pe1085/development/xpusys",
+      "agent_type": "implementer",
+      "prompt": "Clone the repository https://github.com/forkless/ComfyUI-XPUSYS-Monitor-AMD into the current directory /home/pe1085/development/xpusys. Use `git clone https://github.com/forkless/ComfyUI-XPUSYS-Monitor-AMD .` to clone into the existing directory.",
+      "assignment": {
+        "objective": "Clone the repository https://github.com/forkless/ComfyUI-XPUSYS-Monitor-AMD into the current directory /home/pe1085/development/xpusys. Use `git clone https://github.com/forkless/ComfyUI-XPUSYS-Monitor-AMD .` to clone into the existing directory.",
+        "role": "implementer"
+      },
+      "model": "deepseek-v4-flash",
+      "nickname": "Beaked",
+      "status": {
+        "Failed": "Sub-agent requested unavailable tools: exec_shell\n(child model `deepseek-v4-flash` may be unavailable under the current access profile — retry agent_open with a different `model`, or remove `model` to inherit the parent's)"
+      },
+      "result": null,
+      "steps_taken": 0,
+      "duration_ms": 17,
+      "allowed_tools": [
+        "exec_shell"
+      ],
+      "updated_at_ms": 1780951716988,
+      "session_boot_id": "boot_3aa40f1d-b3c"
+    }
+  ]
+}

providers/__init__.py

@@ -137,13 +137,23 @@ def _is_amd_rocme() -> bool:
     """
     Check if torch.cuda is backed by AMD ROCm.
 
-    Returns True if torch.version.roc is not None (AMD ROCm PyTorch).
-    Returns False if torch.version.roc is None (NVIDIA or standard CUDA).
+    Primary signal: torch.version.roc is not None (AMD ROCm PyTorch).
+    Fallback signal: GPU device name contains AMD/Radeon markers
+    (catches Windows ROCm builds that don't set version.roc).
+
+    Returns False for NVIDIA or standard CUDA.
     """
     try:
         import torch
-        # torch.version.roc: True/str = AMD ROCm, None = NVIDIA/other
-        return torch.version.roc is not None
+        # Primary: ROCm version string
+        if torch.version.roc is not None:
+            return True
+        # Fallback: check GPU name for AMD identifiers
+        name = torch.cuda.get_device_name(0).lower()
+        for marker in ["amd", "radeon", "advanced micro devices"]:
+            if marker in name:
+                return True
+        return False
     except Exception:
         return False