Is llama 3.0 not supported anymore or is it just a glitch
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-08-2024 10:14 AM
Hello all,
For some reason, I can't find llama-3.0 models on either the playground or serving endpoints even though we've been using it for months.
I can see both llama-3.1 models on the serving endpoints page and it also says they have been last modified 9 months ago. This, obviously, can't be correct so I suspect it's just a glitch and would expect DB to acknowledge.
Thank you
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-20-2024 10:48 AM
Hi @tefrati, The model is now replaced with v3.1.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-20-2024 11:15 AM
Thank you for the reply Kaniz.
Can you please elaborate regarding the following:
- When did DB replaced 3.0 with 3.1?
- How was it communicated to paying clients? I don't recall any email alerting us about the coming change.
- Is there any rate difference between 3.0 and 3.1?
- We're still using 3.0 endpoint URL. I assume there's an internal re-direct to route requests to 3.1 models. When would we need to update the URL or will DB continue to re-direct our calls?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-27-2024 08:19 AM
@tefrati Thank you for your patience on this one! The pricing remains the same. Our team found version 3.1 to be better than 3 in many aspects, with the added benefit of not requiring any prompt changes. As a result, we have transitioned to 3.1. Our model update policy for pay-per-tokens is listed here. Hope this helps!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-27-2024 08:58 AM
Thank you for your response Sujitha. I understand the rational. However, you can’t make such as decision for your clients without letting them know and give them plenty of time to test and prepare.
I’m sure you thoroughly tested different use cases but, you can’t cover all.
Databricks should have notified its clients at least 60 days in advance and let them make a choice.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
08-28-2024 08:23 AM