Skip to content

fix(google_integration): request a model that is still available - #49

Open
AkhileshSamba wants to merge 1 commit into
sambanova:mainfrom
AkhileshSamba:fix/google-integration-live-model
Open

fix(google_integration): request a model that is still available#49
AkhileshSamba wants to merge 1 commit into
sambanova:mainfrom
AkhileshSamba:fix/google-integration-live-model

Conversation

@AkhileshSamba

Copy link
Copy Markdown

Problem

All three Apps Script files in google_integration/ hardcode a model that no longer exists on SambaCloud:

"model": "Meta-Llama-3.1-405B-Instruct",
  • google_docs.js:77
  • google_sheets.js:19
  • walkthrough.js:20

Meta-Llama-3.1-405B-Instruct was removed on 6/25/2025 and moved to SambaStack only, so it is not reachable with a SambaCloud API key. Every request these scripts make fails.

Verification

The model endpoint is publicly listable and needs no API key:

$ curl https://api.sambanova.ai/v1/models

Returns 6 live models. Meta-Llama-3.1-405B-Instruct is not among them, and the only live Meta-Llama id is Meta-Llama-3.3-70B-Instruct.

The SambaNova docs record the removal at en/models/deprecations.mdx:126, in the table headed "Models that are being deprecated and moved to SambaStack only", and name Meta-Llama-3.3-70B-Instruct as the recommended replacement.

Why this matters more than a normal stale pin

The Google Workspace integration page has one deliverable: copy one of these scripts into Apps Script. That makes these files the code sample. A reader completes the documented setup correctly and every SambaAI action fails.

The docs page was separately corrected (DOC-2651, DOC-2652) to name Meta-Llama-3.3-70B-Instruct, and it now does so twice, including in a Benefits bullet reading "Leverage state-of-the-art models like Meta-Llama-3.3-70B-Instruct." So the page currently advertises 3.3-70B while shipping a request for a model that is gone. The page reads as correct and fails on first use, which is harder to diagnose than a page that reads as rough.

Change

Swapped to Meta-Llama-3.3-70B-Instruct in all three scripts, and updated the three prose references that also named 405B:

File Line Was
google_docs.js 77 "model": "Meta-Llama-3.1-405B-Instruct"
google_docs.js 18 Llama3 405B generated text with proper grammar
google_docs.js 47 appends Llama3 405B generated text
google_sheets.js 19 "model": "Meta-Llama-3.1-405B-Instruct"
walkthrough.js 20 "model": "Meta-Llama-3.1-405B-Instruct"
README.md 15 will give you access to the Llama 3 405B model

"stop": ["<|eot_id|>"] is unchanged and remains correct: 3.3-70B is the same Llama 3 family and uses the same end-of-turn token.

Six one-line changes, no behavioural change beyond the model.

Note for whoever lands this

The docs page carries a temporary workaround telling readers to edit the "model" line themselves, plus a troubleshooting entry for the failure. Both can be removed once this merges. Happy to send that follow-up.

Separately, walkthrough.js is not referenced anywhere on the docs page, so readers who find it through the repo rather than the docs were hitting the same wall with no guidance at all.

🤖 Generated with Claude Code

All three Apps Script files hardcode "model": "Meta-Llama-3.1-405B-Instruct",
which was removed from SambaCloud on 6/25/2025 and moved to SambaStack only.
Every request from these scripts fails.

Verified against the public model endpoint, which needs no API key:

  curl https://api.sambanova.ai/v1/models

It returns 6 live models. Meta-Llama-3.1-405B-Instruct is not among them,
and the only live Meta-Llama is Meta-Llama-3.3-70B-Instruct. The SambaNova
docs record the removal at en/models/deprecations.mdx:126 and name
Meta-Llama-3.3-70B-Instruct as the recommended replacement.

Switched to Meta-Llama-3.3-70B-Instruct in google_docs.js, google_sheets.js,
and walkthrough.js, and updated the two doc comments plus the README line
that promised 405B on the free tier.

The "stop": ["<|eot_id|>"] value is unchanged and remains correct, since
3.3-70B is the same Llama 3 family and uses the same end-of-turn token.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant