Skip to content

Commit bc3be47

Browse files
committed
docs(gemini): document embedding-model filtering and validation errors
The Gemini provider page did not mention that embedding-only models (those without generateContent support) are filtered out of the model catalog, and had no troubleshooting section for validation errors. Both additions are grounded in src/lib/inference/provider-models.ts and src/lib/inference/onboard-probes.ts. Fixes #8971 Signed-off-by: Ashish Shrees <47479956+ashrees@users.noreply.github.com>
1 parent deee72e commit bc3be47

1 file changed

Lines changed: 12 additions & 0 deletions

File tree

docs/inference/use-google-gemini.mdx

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -44,6 +44,18 @@ NemoClaw validates the selected provider and model before creating the sandbox.
4444
NemoClaw validates Gemini inference through its OpenAI-compatible Chat Completions path.
4545
When you enter a custom Gemini model ID, NemoClaw checks Google's native model catalog and accepts IDs with or without the `models/` prefix.
4646
It skips the Responses API probe because Gemini does not support `/v1/responses`.
47+
When NemoClaw reads the native Google model catalog, it keeps only models that support `generateContent`.
48+
Embedding-only models are filtered out of the catalog, so they do not appear as onboarding choices.
49+
50+
## Troubleshooting
51+
52+
Model validation can fail with these messages:
53+
54+
- `Unexpected Gemini model catalog response: expected a top-level models array`: The Google model catalog did not return the expected data. Make sure that `GEMINI_API_KEY` is set and that the host can reach `generativelanguage.googleapis.com`.
55+
- `Gemini model catalog pagination repeated page token '<token>'`: The catalog returned the same page twice. This condition is usually transient. Run the onboarding wizard again.
56+
- `Gemini model catalog pagination exceeded the page budget`: The catalog search reached its page limit before it found the model. Run the wizard again, or select one of the curated models.
57+
- `Unexpected model catalog response: expected a top-level data array`: The OpenAI-compatible endpoint did not return the expected data. Check the provider configuration.
58+
- `Onboard inference smoke check failed.`: The post-selection probe failed. The output shows the provider, model, and API base URL. Compare these values with your configuration.
4759

4860
## Related Topics
4961

0 commit comments

Comments
 (0)