Merge pull request #1167 from transformerlab/fix/run-button-compatible-message

Suggest a compatible loader plugin for when one is not installed and a model is selected

Author: deep1401

api/transformerlab/routers/plugins.py

@@ -80,6 +80,84 @@ async def plugin_gallery():
     return gallery
 
 
+@router.get("/suggest_loader", summary="Suggest a compatible loader plugin for a model architecture.")
+async def suggest_loader_plugin(model_architecture: str):
+    """
+    Suggest a compatible loader plugin based on model architecture and platform.
+    Returns the best matching loader plugin that:
+    1. Supports the model architecture
+    2. Is compatible with the current platform/hardware
+    3. Is not already installed
+    """
+    # Import here to avoid circular dependency
+    import transformerlab.routers.serverinfo as serverinfo_module
+
+    device_type = serverinfo_module.system_info.get("device_type", "cpu")
+
+    # Map device_type to supported_hardware_architectures
+    # device_type: nvidia -> cuda, apple_silicon -> mlx, amd -> amd, cpu -> cpu
+    hardware_arch_map = {
+        "nvidia": "cuda",
+        "apple_silicon": "mlx",
+        "amd": "amd",
+        "cpu": "cpu",
+    }
+    required_hardware = hardware_arch_map.get(device_type, "cpu")
+
+    # Get all plugins from gallery
+    gallery = await plugin_gallery()
+
+    # Filter for loader plugins that:
+    # 1. Are of type "loader"
+    # 2. Are not installed
+    # 3. Support the model architecture
+    # 4. Support the current hardware architecture
+    compatible_plugins = []
+
+    for plugin in gallery:
+        # Must be a loader plugin
+        if plugin.get("type") != "loader":
+            continue
+
+        # Must not be installed
+        if plugin.get("installed", False):
+            continue
+
+        # Must support the model architecture
+        model_architectures = plugin.get("model_architectures", [])
+        if not isinstance(model_architectures, list):
+            continue
+
+        architecture_match = False
+        for arch in model_architectures:
+            if arch and arch.lower() == model_architecture.lower():
+                architecture_match = True
+                break
+
+        if not architecture_match:
+            continue
+
+        # Must support the current hardware architecture
+        supported_hardware = plugin.get("supported_hardware_architectures", [])
+        if not isinstance(supported_hardware, list):
+            continue
+
+        hardware_match = required_hardware in supported_hardware
+
+        if hardware_match:
+            compatible_plugins.append(plugin)
+
+    # If no compatible plugins found, return None
+    if not compatible_plugins:
+        return None
+
+    # Sort alphabetically by name and return the first one
+    compatible_plugins.sort(key=lambda p: p.get("name", ""))
+
+    # Return the first match
+    return compatible_plugins[0]
+
+
 async def copy_plugin_files_to_workspace(plugin_id: str):
     plugin_id = secure_filename(plugin_id)
 

src/renderer/components/Experiment/Foundation/RunModelButton.tsx

@@ -26,7 +26,8 @@ import {
 import InferenceEngineModal from './InferenceEngineModal';
 import * as chatAPI from 'renderer/lib/transformerlab-api-sdk';
 import OneTimePopup from 'renderer/components/Shared/OneTimePopup';
-import { useAPI } from 'renderer/lib/transformerlab-api-sdk';
+import { useAPI, fetcher } from 'renderer/lib/transformerlab-api-sdk';
+import { useSWRWithAuth as useSWR } from 'renderer/lib/authContext';
 import React from 'react';
 
 import { Link } from 'react-router-dom';
@@ -83,6 +84,13 @@ export default function RunModelButton({
 
   const archTag = experimentInfo?.config?.foundation_model_architecture ?? '';
 
+  // Fetch suggested compatible loader plugin from API (platform-aware)
+  const { data: suggestedLoaderPlugin, isLoading: suggestedPluginLoading } =
+    useSWR(
+      archTag ? chatAPI.Endpoints.Plugins.SuggestLoader(archTag) : null,
+      fetcher,
+    );
+
   const supportedEngines = React.useMemo(() => {
     if (!data || pipelineTagLoading) return [];
 
@@ -472,12 +480,30 @@ export default function RunModelButton({
           <Alert startDecorator={<TriangleAlertIcon />} color="warning">
             <Typography level="body-sm">
               None of the installed Engines currently support this model
-              architecture. You can try a different engine in{' '}
-              <Link to="/plugins">
-                <Plug2Icon size="15px" />
-                Plugins
-              </Link>{' '}
-              , or you can try running it with an unsupported Engine by clicking{' '}
+              architecture.
+              {suggestedLoaderPlugin ? (
+                <>
+                  {' '}
+                  <b>{suggestedLoaderPlugin.name}</b> is compatible with this
+                  model architecture. Install it in{' '}
+                  <Link to="/plugins">
+                    <Plug2Icon size="15px" />
+                    Plugins
+                  </Link>
+                  .
+                </>
+              ) : (
+                <>
+                  {' '}
+                  You can try a different engine in{' '}
+                  <Link to="/plugins">
+                    <Plug2Icon size="15px" />
+                    Plugins
+                  </Link>
+                  .
+                </>
+              )}{' '}
+              Or you can try running it with an unsupported Engine by clicking{' '}
               <b>using Engine</b> below and check{' '}
               <b>Show unsupported engines</b>.
             </Typography>
@@ -495,9 +521,21 @@ export default function RunModelButton({
               <Plug2Icon size="15px" />
               Plugins
             </Link>{' '}
-            and install an Inference Engine. <b>FastChat Server</b> is a good
-            default for systems with a GPU. <b>Apple MLX Server</b> is the best
-            default for MacOS with Apple Silicon.
+            and install an Inference Engine.
+            {suggestedLoaderPlugin ? (
+              <>
+                {' '}
+                <b>{suggestedLoaderPlugin.name}</b> is compatible with this
+                model architecture.
+              </>
+            ) : (
+              <>
+                {' '}
+                <b>FastChat Server</b> is a good default for systems with a GPU.{' '}
+                <b>Apple MLX Server</b> is the best default for MacOS with Apple
+                Silicon.
+              </>
+            )}
           </Typography>
         </Alert>
       )}

src/renderer/lib/api-client/endpoints.ts

@@ -275,6 +275,8 @@ Endpoints.Plugins = {
   List: () => `${API_URL()}plugins/list`,
   RunPluginInstallScript: (pluginId: string) =>
     `${API_URL()}plugins/${pluginId}/run_installer_script`,
+  SuggestLoader: (modelArchitecture: string) =>
+    `${API_URL()}plugins/suggest_loader?model_architecture=${encodeURIComponent(modelArchitecture)}`,
 };
 
 // Following is no longer needed as it is replaced with useAPI