Open cristiam86 opened 1 week ago
Should we then decide of a flat value @cristiam86 ?
Also, is this on creation of a new validator, or when installing the simulator?
This is in the backend when we create a new validator. We are randomizing the parameters (see: backend/node/create_nodes and backend/node/defaults.json) Making all these values random is wrong. We should aim for the optimal values where we can. More research is needed to understand which values can be random and which ones can be optimized
If I'm correct, this is only when using the init command genlayer init --numValidators 10
There's no randomization happening when creating a validator manually from the UI.
Do we still want to remove the randomization of config if that is the case?
For this issue we are always talking about the randomization of config, not the providers and models.
Sorry if I wasn't clear. What I mean is, it seems we only randomize the config when calling explicitly those endpoints:
And from what I understand, these are only used in tests and in debug_simulator.py (which will be removed).
The create_validator
endpoint (the one used when we create a validator via the frontend) does not randomize the config.
If I'm correct about the above, do we still need to change something?
Happy to discuss it if necessary.
@denishacquin I think it might be a good idea to wait for https://github.com/yeagerai/genlayer-simulator/issues/359, since all of this configuration will change after that refactor
@denishacquin @AgustinRamiroDiaz it seems to me that what Agustín did here: https://github.com/yeagerai/genlayer-simulator/pull/475/files#diff-c0c0177862bf06ec410fcbcd543e48045c6e9fc17c2120ff2e93e402b6037fb2R1 fixes this issue
Indeed, let's close this when we merge #475
When setting up heurist, it randomizes the configuration, but I don’t think max_tokens is a good parameter to randomise