You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@chainzero has experimented with the provider per GPU model bid/pricing in a revised scripting logic I created and seems to work nicely.
He modified the default bid script with per GPU model pricing here:
It would be extremely beneficial for the bid script to have the capability to identify the specific GPU that will be allocated when a deployment opts (in SDL) for the "DEFAULT" (*) card model setting.
In the absence of this functionality, I suggest we configure the TARGET_GPU_DEFAULT to the highest-value GPU available from providers. Although this method is not ideal, it will serve as an adequate interim solution.
I'll explore the option of configuring pricing variables for CPU, memory, storage, various storage types (pers.storage aka ceph), leased IPs (aka metallb), ports, and GPU directly through the Akash provider's Helm chart variables. This way, providers would only need to update the provider.yaml file and avoid having to modify the bid price script.
Additionally, I need to test the AKASH_BID_PRICE_SCRIPT_PROCESS_TIMEOUT parameter. Its default value is set at 5 seconds, which seems a bit too short. I'm considering adjusting it to 10 seconds, especially since the API query for acquiring the AKT price from the Osmosis/CoinGecko API every 60 minutes can sometimes exceed the default 5-second timeout.
The text was updated successfully, but these errors were encountered:
@chainzero has experimented with the provider per GPU model bid/pricing in a revised scripting logic I created and seems to work nicely.
He modified the default bid script with per GPU model pricing here:
It would be extremely beneficial for the bid script to have the capability to identify the specific GPU that will be allocated when a deployment opts (in SDL) for the "DEFAULT" (
*
) card model setting.In the absence of this functionality, I suggest we configure the
TARGET_GPU_DEFAULT
to the highest-value GPU available from providers. Although this method is not ideal, it will serve as an adequate interim solution.I'll explore the option of configuring pricing variables for CPU, memory, storage, various storage types (pers.storage aka ceph), leased IPs (aka metallb), ports, and GPU directly through the Akash provider's Helm chart variables. This way, providers would only need to update the
provider.yaml
file and avoid having to modify the bid price script.Additionally, I need to test the
AKASH_BID_PRICE_SCRIPT_PROCESS_TIMEOUT
parameter. Its default value is set at5 seconds
, which seems a bit too short. I'm considering adjusting it to10 seconds
, especially since the API query for acquiring the AKT price from the Osmosis/CoinGecko API every60 minutes
can sometimes exceed the default 5-second timeout.The text was updated successfully, but these errors were encountered: