From af509077063db59af60b0523db433bfb38376e27 Mon Sep 17 00:00:00 2001 From: wakqasahmed Date: Fri, 14 Aug 2026 17:12:22 +0200 Subject: [PATCH 1/3] docs: document Google Gemini embedding-model filtering and troubleshooting (#8971) The Google Gemini provider page's Validation section did not mention that embedding-only models (no generateContent support) are filtered out of the model catalog. No troubleshooting entry documented the validation error messages a user hits when a Gemini model fails validation. Document the embedding-only filter in the Validation section of use-google-gemini.mdx, add a new Troubleshooting entry covering both Gemini model-validation error strings, and link the two pages together. Signed-off-by: wakqasahmed --- docs/inference/use-google-gemini.mdx | 4 ++++ docs/reference/troubleshooting.mdx | 23 +++++++++++++++++++++++ 2 files changed, 27 insertions(+) diff --git a/docs/inference/use-google-gemini.mdx b/docs/inference/use-google-gemini.mdx index 3f11f44c3d0..d65d49bf63a 100644 --- a/docs/inference/use-google-gemini.mdx +++ b/docs/inference/use-google-gemini.mdx @@ -45,7 +45,11 @@ NemoClaw validates Gemini inference through its OpenAI-compatible Chat Completio When you enter a custom Gemini model ID, NemoClaw checks Google's native model catalog and accepts IDs with or without the `models/` prefix. It skips the Responses API probe because Gemini does not support `/v1/responses`. +Google's native model catalog also lists embedding-only models, which do not support chat completions. +NemoClaw filters these out of the catalog before matching your selection, so embedding-only models are never offered during onboarding and always fail validation if entered manually. + ## Related Topics - [Choose a Model](../learn-and-choose/choose-model) compares the curated Gemini models by task fit. - [Understand Provider Validation](../validate-inference/understand-provider-validation) describes provider validation behavior. +- [Troubleshooting](../reference/troubleshooting#google-gemini-rejects-a-curated-or-custom-model) lists the error messages produced when a Gemini model fails validation. diff --git a/docs/reference/troubleshooting.mdx b/docs/reference/troubleshooting.mdx index 0c53b560624..3efec4b9e15 100644 --- a/docs/reference/troubleshooting.mdx +++ b/docs/reference/troubleshooting.mdx @@ -1254,6 +1254,29 @@ bash uninstall.sh --yes curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash ``` +### Google Gemini rejects a curated or custom model + +NemoClaw validates a selected Google Gemini model against Google's native model catalog (`https://generativelanguage.googleapis.com/v1beta/models`) before it creates the sandbox. +That catalog includes embedding-only models, which do not support chat completions. +NemoClaw filters out any model whose `supportedGenerationMethods` does not include `generateContent`, so embedding-only models never appear as selectable and always fail validation if entered manually. + +If the catalog request itself fails (invalid API key, network error, rate limit, or an outage), onboarding reports: + +```text +Could not validate model against https://generativelanguage.googleapis.com/v1beta/models: {reason} +``` + +Check that `GEMINI_API_KEY` is set and valid, and that the host can reach `generativelanguage.googleapis.com`, then retry onboarding. + +If the catalog request succeeds but the selected model is not present in the filtered results, onboarding reports: + +```text +Model '{model}' is not available from Google Gemini. Checked https://generativelanguage.googleapis.com/v1beta/models. +``` + +This means the model ID is either misspelled, not available to your API key or account tier, or is an embedding-only model (for example, an `embedding-*` or `text-embedding-*` model) that Google's catalog does not list under `generateContent`. +NemoClaw accepts the model ID with or without the `models/` prefix, so re-check the ID against the [curated Gemini models](../inference/use-google-gemini#model-choices) or Google's published model list rather than the prefix form. + ## Runtime From a947d72db3e6e7e5c48be801cc300e2e2177196a Mon Sep 17 00:00:00 2001 From: wakqasahmed Date: Fri, 14 Aug 2026 20:18:08 +0200 Subject: [PATCH 2/3] docs(gemini): clarify optional models/ prefix wording --- docs/reference/troubleshooting.mdx | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/reference/troubleshooting.mdx b/docs/reference/troubleshooting.mdx index 3efec4b9e15..9fe68a3dfb8 100644 --- a/docs/reference/troubleshooting.mdx +++ b/docs/reference/troubleshooting.mdx @@ -1275,7 +1275,7 @@ Model '{model}' is not available from Google Gemini. Checked https://generativel ``` This means the model ID is either misspelled, not available to your API key or account tier, or is an embedding-only model (for example, an `embedding-*` or `text-embedding-*` model) that Google's catalog does not list under `generateContent`. -NemoClaw accepts the model ID with or without the `models/` prefix, so re-check the ID against the [curated Gemini models](../inference/use-google-gemini#model-choices) or Google's published model list rather than the prefix form. +NemoClaw accepts the model ID with or without the `models/` prefix. Verify the model name against the [curated Gemini models](../inference/use-google-gemini#model-choices) or Google's published model list. Do not treat the optional prefix as a different model. ## Runtime From 50d97e1b427fd250db2fabb37b9b785d5866c169 Mon Sep 17 00:00:00 2001 From: Prekshi Vyas Date: Fri, 14 Aug 2026 13:43:08 -0700 Subject: [PATCH 3/3] docs: fix Gemini troubleshooting route --- docs/inference/use-google-gemini.mdx | 2 +- docs/reference/troubleshooting.mdx | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/inference/use-google-gemini.mdx b/docs/inference/use-google-gemini.mdx index d65d49bf63a..85171a77c9b 100644 --- a/docs/inference/use-google-gemini.mdx +++ b/docs/inference/use-google-gemini.mdx @@ -52,4 +52,4 @@ NemoClaw filters these out of the catalog before matching your selection, so emb - [Choose a Model](../learn-and-choose/choose-model) compares the curated Gemini models by task fit. - [Understand Provider Validation](../validate-inference/understand-provider-validation) describes provider validation behavior. -- [Troubleshooting](../reference/troubleshooting#google-gemini-rejects-a-curated-or-custom-model) lists the error messages produced when a Gemini model fails validation. +- [Troubleshooting](../../reference/troubleshooting#google-gemini-rejects-a-curated-or-custom-model) lists the error messages produced when a Gemini model fails validation. diff --git a/docs/reference/troubleshooting.mdx b/docs/reference/troubleshooting.mdx index 9fe68a3dfb8..2750935b89e 100644 --- a/docs/reference/troubleshooting.mdx +++ b/docs/reference/troubleshooting.mdx @@ -1275,7 +1275,7 @@ Model '{model}' is not available from Google Gemini. Checked https://generativel ``` This means the model ID is either misspelled, not available to your API key or account tier, or is an embedding-only model (for example, an `embedding-*` or `text-embedding-*` model) that Google's catalog does not list under `generateContent`. -NemoClaw accepts the model ID with or without the `models/` prefix. Verify the model name against the [curated Gemini models](../inference/use-google-gemini#model-choices) or Google's published model list. Do not treat the optional prefix as a different model. +NemoClaw accepts the model ID with or without the `models/` prefix. Verify the model name against the [curated Gemini models](../inference/hosted-inference/use-google-gemini#model-choices) or Google's published model list. Do not treat the optional prefix as a different model. ## Runtime