A question about determining potential costs on Az...
# support
t
A question about determining potential costs on Azure. Maybe @Max @azul (she/her) (nixtla) @Cristian Challu or @Mariana Menchero can help?
โœ… 1
"Franco, Joao Vitor" <joao.vitor.franco@effem.com>
Hello, I hope you're doing well.
Can you please share a cost calculation example using the TimeGPT model in the Azure environment?
Iโ€™m particularly interested to understand how the cost works when finetuning it.
How `finetune_steps`parameter impact cost? Should I consider all 4 types of costs when performing a finetuning?
Please share a detailed example covering all possible scenarios, with finetuning or not.
m
@Cristian Challu, do you have the updated calculator?
@Cristian Challu, please dont forget about this.
c
We don't have a calculator for Microsoft prices. I can do the calculator once I return to the hotel at night ``````
๐Ÿ‘ 1
t
Putting this one back in the queue, to get Joao at effem.com a response about a price calculation for Azure. @Cristian Challu Thanks!
m
I think @Cristian Challu will be back tomorrow and surely he will come with something up. MSFT is also asking for something similar.
๐Ÿ‘ 1
c
Answer to his question:
Copy code
The finetune steps determine the type of token you will be charged. If you are finetuning the model, then only the finetuned type of tokens apply.
t
ok, thanks, I'll share this with him
In a PS email, he also asked "PS.: Please also explain to me how the
level
parameter impacts the cost when using it. For example, letโ€™s assume a confidence interval of 80%"
c
It increases the number of outputs. Each time you specify a level it returns to additional values per timestamp. The actual formula is
output_tokens=num_series*h*(1+2*num_levels)
๐Ÿ™ 1
t
Replied and put you on cc @Cristian Challu