Closed hide-on-bush-x closed 3 months ago
@hide-on-bush-x does this config help at all
weights_rate_limit
sounds like may help with the No attempt made. Perhaps it is too soon to set weights! - set_weights failed -
thing
Thanks for the input @teslashibe
@hide-on-bush-x be sure to ask in the developer discord. I don't see you active there and they are incredibly responsive. https://discord.gg/dkW2CaMD
@teslashibe will ask about this after we wrap up the docs, thx
Found out some answers
BrokenPipeError: [Errno 32] Broken pipe
means that the metagraph is out of sync ( Needs to reinstantiate the subtensor object or do metagraph.sync )
No attempt made. Perhaps it is too soon to set weights! - set_weights failed -
Is just a rate limit, configured in the hyperparameters.
More info about hyperparameters in #61
Hyperparams weights_rate_limit
update:
Attempted to set the rate limit via weights_rate_limit
on devnet
netuid 1
(a subnet I created / have the owner wallet for). Getting the following error when running btcli sudo set --param weights_rate_limit --value 5 --netuid 1 --subtensor.chain_endpoint ws://54.205.45.3:9945
:
raise InvalidScaleTypeValueException("Invalid byte for Compact")
Investigating further, will update here.
@grantdfoster can you share what's the status here? thank you!
@mudler this has been fixed with both #62 and #82... solved with setting weights_rate_limit
to a lower value and upgrading our devnet to the latest subtensor - this can be closed again!
Perfect! Thanks Grant!
set_weights
you can find this method defined in
masa/base/validator.py
Problems encountered
Tried to sort some of them already but didn't reach any good state where it works
Error during the query and score process: 'numpy.ndarray' object has no attribute 'to'
On this snippet you can see
self.metagraph.uids.to("cpu")
this is failing because some kind of type missmatch Tried to sort this out by doingtorch.from_numpy(self.metagraph.uids).to("cpu")
andtorch.from_numpy(self.metagraph.uids)
But this led toError during the query and score process: '<' not supported between instances of 'builtin_function_or_method' and 'int'
. Found no solution for this, I think thetorch.from_numpy
think is not the right wayMaybe the metagraph.uids is returning an unexpected value, not sure
Not waiting for finalization or inclusion. Assume successful
PR I found with this https://github.com/opentensor/bittensor/pull/1692/commits/559ec182bbda95f37de2ecd693aae6f430b6f437
Tried setting weights manually from python console resulting on aboves "error". Even tho it returns True when doing
subnet.W
it returns an empty arrayUseful information
Found this thread where they talk about this topic: https://discord.com/channels/799672011265015819/799672011814862902/1209127894195109939
in order to see the the weights, you need to instantiate the subnet on the following way
This way it returns the matrix and you can debug properly
After setting weights manually once I started getting more errors
like:
BrokenPipeError: [Errno 32] Broken pipe
orNo attempt made. Perhaps it is too soon to set weights! - set_weights failed -
Update
Comented out the first part of the set_weights method and seems like its being set, I am not really sure about it but maybe that works for now
This way we are skipping calling
process_weights_for_netuid
that was failing no matter wah