Merge branch 'main' into fix/serialize_local_jobs

Author: dadmobile

.github/dependabot.yml

@@ -106,39 +106,6 @@ updates:
       - dependency-name: '@teamsupercell/typings-for-css-modules-loader'
         update-types: ['version-update:semver-major']
 
-  # Docs site (Docusaurus, managed with yarn classic — dependabot's npm ecosystem auto-detects yarn.lock)
-  # Build runs on Node 22 via transformerlab-docs/netlify.toml. Major bumps are ignored across
-  # the board (manual coordination needed); minor/patch updates flow through for security.
-  - package-ecosystem: 'npm'
-    directory: '/transformerlab-docs'
-    schedule:
-      interval: 'weekly'
-      day: 'monday'
-    open-pull-requests-limit: 5
-    commit-message:
-      prefix: 'docs-deps'
-      include: 'scope'
-    labels:
-      - 'dependencies'
-      - 'docs'
-    groups:
-      docusaurus:
-        patterns:
-          - '@docusaurus/*'
-          - '@mdx-js/*'
-      react:
-        patterns:
-          - 'react'
-          - 'react-dom'
-      minor-and-patch:
-        update-types:
-          - 'minor'
-          - 'patch'
-    ignore:
-      # Major bumps on docs deps need manual coordination; minor/patch is enough for security.
-      - dependency-name: '*'
-        update-types: ['version-update:semver-major']
-
   # API (FastAPI, managed with uv)
   - package-ecosystem: 'uv'
     directory: '/api'

.pre-commit-config.yaml

@@ -18,7 +18,7 @@ repos:
         entry: npm run format
         language: system
         pass_filenames: false
-        files: ^(src|transformerlab-docs)/.*\.(js|jsx|ts|tsx|json|css|scss|md|mdx|ya?ml)$
+        files: ^src/.*\.(js|jsx|ts|tsx|json|css|scss|md|mdx|ya?ml)$
   - repo: local
     hooks:
       - id: react-doctor

api/test/compute_providers/test_runpod_image.py

@@ -56,3 +56,25 @@ def test_default_image_when_no_override():
     config = ClusterConfig(accelerators="A100:1")
     provider.launch_cluster("clust-1", config)
     assert captured["pod_data"]["imageName"] == "runpod/pytorch:1.0.3-cu1281-torch290-ubuntu2204"
+
+
+def test_amd_gpu_defaults_to_rocm_image():
+    provider = _provider()
+    captured = _capture_pod_data(provider)
+    # AMD/ROCm card resolves to a ROCm image, not the default CUDA image.
+    provider._map_gpu_type_to_runpod = lambda _accel: "AMD Instinct MI300X OAM"
+    config = ClusterConfig(accelerators="MI300X:1")
+    provider.launch_cluster("clust-1", config)
+    assert captured["pod_data"]["imageName"] == "rocm/pytorch:latest"
+
+
+def test_amd_gpu_image_name_override_still_honored():
+    provider = _provider()
+    captured = _capture_pod_data(provider)
+    provider._map_gpu_type_to_runpod = lambda _accel: "AMD Instinct MI300X OAM"
+    config = ClusterConfig(
+        accelerators="MI300X:1",
+        provider_config={"image_name": "my-org/rocm-custom:latest"},
+    )
+    provider.launch_cluster("clust-1", config)
+    assert captured["pod_data"]["imageName"] == "my-org/rocm-custom:latest"

api/transformerlab/compute_providers/runpod.py

@@ -68,8 +68,33 @@
     "RTX5000": "NVIDIA RTX 5000 Ada Generation",
     "RTX4000": "NVIDIA RTX 4000 Ada Generation",
     "RTX2000": "NVIDIA RTX 2000 Ada Generation",
+    # AMD Instinct (ROCm). Launching these requires a ROCm container image, not
+    # the default CUDA image — see _is_amd_gpu_id and the image selection in
+    # launch_cluster().
+    "MI300X": "AMD Instinct MI300X OAM",
 }
 
+# Runpod GPU type IDs for AMD/ROCm cards. Used to pick a ROCm container image at
+# launch time instead of the default CUDA image.
+_RUNPOD_AMD_GPU_IDS: frozenset = frozenset(
+    {
+        "AMD Instinct MI300X OAM",
+    }
+)
+
+# Default ROCm container image for AMD pods. Overridable per-launch via the
+# provider config `image_name` (or legacy `template_id`).
+_RUNPOD_DEFAULT_ROCM_IMAGE = "rocm/pytorch:latest"
+# Default CUDA container image for NVIDIA pods.
+_RUNPOD_DEFAULT_CUDA_IMAGE = "runpod/pytorch:1.0.3-cu1281-torch290-ubuntu2204"
+
+
+def _is_amd_gpu_id(gpu_type_id: Optional[str]) -> bool:
+    """Return True if the resolved Runpod GPU type ID is an AMD/ROCm card."""
+    if not gpu_type_id:
+        return False
+    return gpu_type_id in _RUNPOD_AMD_GPU_IDS or "AMD" in gpu_type_id
+
 
 async def fetch_runpod_provider_logs(
     provider_instance: ComputeProvider,
@@ -360,8 +385,12 @@ def launch_cluster(self, cluster_name: str, config: ClusterConfig) -> Dict[str,
                 except ValueError:
                     gpu_count = 1
 
-            # Use GPU-enabled image
-            default_image = "runpod/pytorch:1.0.3-cu1281-torch290-ubuntu2204"
+            # Use a GPU-enabled image. AMD/ROCm cards need a ROCm image; the
+            # default CUDA image will not run on them.
+            if _is_amd_gpu_id(gpu_type_id):
+                default_image = _RUNPOD_DEFAULT_ROCM_IMAGE
+            else:
+                default_image = _RUNPOD_DEFAULT_CUDA_IMAGE
         else:
             # CPU pod
             compute_type = "CPU"

package.json

@@ -32,8 +32,8 @@
     "check-ports": "! lsof -i :1212 -i :8338",
     "start": "dotenv -e .env -- cross-env NODE_ENV=development tsx node_modules/webpack-dev-server/bin/webpack-dev-server.js --config ./.erb/configs/webpack.config.cloud.dev.ts",
     "dev": "npm run check-ports && concurrently -k -n FRONTEND,API -c cyan,magenta \"npm run start\" \"npm run api:start\"",
-    "format:check": "prettier --check \"src/**/*.{js,jsx,ts,tsx,json,css,scss,md}\" \"transformerlab-docs/**/*.{js,jsx,ts,tsx,json,css,scss,md,mdx,yml,yaml}\"",
-    "format": "prettier --write \"src/**/*.{js,jsx,ts,tsx,json,css,scss,md}\" \"transformerlab-docs/**/*.{js,jsx,ts,tsx,json,css,scss,md,mdx,yml,yaml}\"",
+    "format:check": "prettier --check \"src/**/*.{js,jsx,ts,tsx,json,css,scss,md}\"",
+    "format": "prettier --write \"src/**/*.{js,jsx,ts,tsx,json,css,scss,md}\"",
     "format:api": "prettier --write \"api/transformerlab/plugins/**/*.json\"",
     "api:install": "cd api && ./install.sh",
     "api:start": "cd api && ./run.sh",