Skip to content

Fix auto device falling back when a provider fails to load #1674

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Suh0161 wants to merge 1 commit into
base: main
Choose a base branch
from
+31 −1
Open

Fix auto device falling back when a provider fails to load #1674

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 31 additions & 1 deletion packages/transformers/src/models/session.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@ import {
} from '../backends/onnx.js';
import { getCacheNames } from '../configs.js';
import { DATA_TYPES, DEFAULT_DTYPE_SUFFIX_MAPPING, isWebGpuFp16Supported, selectDtype } from '../utils/dtypes.js';
import { selectDevice } from '../utils/devices.js';
import { selectDevice, DEVICE_TYPES } from '../utils/devices.js';
import { apis } from '../env.js';
import { getCoreModelFile, getModelDataFiles } from '../utils/model-loader.js';
import { Tensor } from '../utils/tensor.js';
Expand Down Expand Up @@ -154,6 +154,36 @@ export async function constructSessions(pretrained_model_name_or_path, names, op
cache_config,
name,
);

// When device='auto', try each execution provider individually so that a failing
// accelerator (e.g. CUDA not installed) falls back cleanly to the next one.
// For any explicit device the caller already knows what they want — no fallback.
const isAuto =

@nico-martin nico-martin Jun 8, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The isAuto check re-reads the raw option/config value instead of using the resolved selectedDevice already computed in getSession and returned in session_config.device:

Please base this on the resolved device, for example session_config.device === DEVICE_TYPES.auto, or return selectedDevice explicitly from getSession and use that.

(options.device ?? options.config?.['transformers.js_config']?.device) === DEVICE_TYPES.auto;

if (isAuto) {
const candidates = /** @type {import('onnxruntime-common').InferenceSession.ExecutionProviderConfig[]} */ (
session_options.executionProviders
);
let lastError;
for (const provider of candidates) {
try {
const opts = { ...session_options, executionProviders: [provider] };
const session = await createInferenceSession(buffer_or_path, opts, {
...session_config,
device: typeof provider === 'string' ? provider : provider.name,
});
return [name, session];
} catch (err) {
logger.warn(
`Failed to create session with provider "${typeof provider === 'string' ? provider : provider.name}": ${err?.message ?? err}. Trying next provider.`,
);
lastError = err;
}
}
throw lastError ?? new Error('All execution providers failed to initialize.');
}

const session = await createInferenceSession(buffer_or_path, session_options, session_config);
return [name, session];
}),
Expand Down