Closed chinyeefan closed 1 month ago
We are sorry to hear about your experience. Please note that while the graphrag library can run on a laptop, only small amounts of data can be indexed. When indexing a large volume of data, GraphRAG currently has a high memory requirement. To ensure a reasonable and stable experience with the solution accelerator for the most general use case, the E series VMs were chosen.
https://github.com/Azure-Samples/graphrag-accelerator/pull/140 https://github.com/Azure-Samples/graphrag-accelerator/pull/141
We have modified the AKS deployment to use a more efficient VM series while only using the E series VMs for indexing, as well as provided warnings in the documentation. If you are testing out graphrag with very small amounts of data, your options to reduce costs are to use the graphrag library directly or to spin down the graphrag accelerator Azure resources when not in use.
Hey @chinyeefan, could you open an Azure Support ticket within the portal about the billing from this sample experience? Hopefully they can help understand the situation and they may be able to waive the costs if this is the first time it's been an issue for your account.
Appreciate you taking the time to open the issue, I think we can work with teams to emphasize the warnings even more for similar projects that could have a large compute cost if run for too long.
Thanks, Jeff Microsoft open source office
I followed the recommended Azure configuration as in here, and forget about it until one month later and I see my bill is !!!!!!!!!! OVER $1300!!!!!!!!!!!!
If the whole accelerator can be done in a laptop, I don't see any reason why it needs a
E16as v5
instance in Azure. This is absolutely an overqualified expensive option for running this project. Here's what aE16as v5
looks likeI am deeply concerned it's a sneaky tactic to sell Azure services. This kind of behaviour is not only unethical but also severely damages the trust users place in open source projects.
I expect immediate action to address this issue, including inform people of the costs and adjust the service to a reasonable tier