Higher LoRA rank limit for fine-tuning
Update4 days agoUnverifiedAdded Oct 2, 2026
You can now train LoRA adapters with a rank of up to 128 for the majority of models, up from 64. The default rank for these models stays at 64, so set lora_r to use a higher one. The model limits response has a new lora_training.default_rank field next to lora_training.max_rank . Run tg fine-tuning model-limits to see both values for a model.
Topics: Developer tools
More Together AI platform releases
Every Together AI platform release| Date | Release | Type |
|---|---|---|
| Sep 294 days ago | Update | Update |
| Sep 294 days ago | Batch API files are now retained for 7 days unverified Update | Update |
| Oct 12 days ago | Together Link beta unverified Beta | Beta |
| Oct 12 days ago | What's included: unverified Update | Update |
| Oct 12 days ago | Code sandbox SDK and CLI unverified Update | Update |
| Sep 249 days ago | TogetherLink beta unverified Beta | Beta |
| Sep 2310 days ago | New serverless models unverified Update | Update |
| Sep 2211 days ago | Pricing update unverified Pricing | Pricing |
Also shipped on Sep 29, 2026
Together AI in September 2026| Date | Company | Release | Product | Type |
|---|---|---|---|---|
| Sep 294 days ago | Databricks Apps | GA | ||
| Sep 294 days ago | New Relic | GA | ||
| Sep 294 days ago | CodeRabbit | Beta | ||
| Sep 294 days ago | Cloud Run | GA | ||
| Sep 294 days ago | Rich link previews for Agent Runners sites unverified | Netlify | Preview | |
| Sep 294 days ago | DigitalOcean | GA |