Below are some common errors that our users have run into, we suggest checking this list out before posting in the Discord server to save you time, as most of these are easily fixable.
pip3 install -r requirements.txt, an error appears:
ImportError: No module named 'Cython'
this is a common error due to pyarrow's requirement of cython. Simply install cython module:
$ pip install cython
When creating a bittensor wallet and inputting a password for the wallet:
You can use the mnemonic to recreate the key in case it gets lost. The command to use to regenerate the key using this mnemonic is:bittensor-cli regen --mnemonic urge oppose source index dirt boy bulb axis rose baby wait swearSpecify password for key encryption:Password not strong enough. Try increasing the length of the password or the password complexity
As the error message states, your password needs to be strong enough. For security, Bittensor does not have a predefined set of rules (i.e. you need alphanumeric characters, special symbols, special length, etc.). Instead, Bittensor uses so-called "entropy bits". This defines how much variety does your password have. For example, '01111010010011' is long enough, but has only 2 entropy bits: that's how many bits you need to store its alphabet. However, a password that uses plenty of characters has more entropy.
This is because you are likely running the miner with default settings. The default settings are tuned so that the miner can run on very minimal setups (1 CPU, 4GB RAM). You can increase the miner's batch size using the
--miner.batch_size_train flag. If you're using a GPU it's recommended that you slowly increment the batch size until you see a healthy but full utilization of your GPU, as it is very easy to run out of memory. If you still have more room for utilization after increasing the batch_size, you can also increase the size of the model by looking into the
No big deal! Just restart your miner by re-running the miner command you used to run your miner previously.
Each time you run the miner, it creates a new directory under
gpt2_genesis is the name of your miner, this can be different according to whatever miner you're running). The miner stores the model that it is training there, and each model is approx. 512MB to 1.5GB in size, depending on your training parameters. The idea is that you can reload these compiled
.torch files to a regular torch model and start using it for inference (see
This happens, and is totally normal. There are many reasons why your rank may drop, including: your internet connectivity (slow, choppy internet), your model performing poorly, or perhaps even because you're behind on the bittensor version (a quick git pull master will fix this).
This is impossible, plain and simple. Chances are you most likely linked more than one miner with the same hotkey, and each time you check your wallet, the fetch mechanism is fetching the balance of one of the miners and you are seeing that balance, not the one you are expecting. It's best to triple check your miners and their hotkeys to make sure you haven't allocated the same hotkey to more than one miner.