The vscode commands in the README.md for Fine-tuning is outdated.
- It reads
AI Toolkit instead of Foundry Toolkit
- It reads
HF_TOKEN instead of hf-token
- Step 1 refers to make sure your subscription has enough GPU capacity, but this is unclear, it leads to confusion. Ex., I requested for ACA service, SKU SubscriptionNCA100Gpus and West US 3 region, quota increase from 0 to 1, but it wasn't until I ran command
Foundry Toolkit: Provision Azure Container Apps job for fine-tuning that I found this error -below- and checking with the Microsoft Staff over Discord that we found the minimum requested capacity is 2. A better experience would be to list in a table the models supported what minimum GPU capacity is required. In my case Meta-Llama-3-8B requires 2.
{
"code": "MaxRegionalQuotaExceeded",
"message": "Regional max quota limit threshold hit. The subscription dd36d329-c225-4803-8c5e-d5deca941777 cannot have more than 1 SubscriptionNCA100Gpus in region West US 3."
}
The vscode commands in the README.md for Fine-tuning is outdated.
AI Toolkitinstead ofFoundry ToolkitHF_TOKENinstead ofhf-tokenFoundry Toolkit: Provision Azure Container Apps job for fine-tuningthat I found this error -below- and checking with the Microsoft Staff over Discord that we found the minimum requested capacity is 2. A better experience would be to list in a table the models supported what minimum GPU capacity is required. In my case Meta-Llama-3-8B requires 2.