Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add default 1024 max stake limit for querying UIDs with vpermit. #1379

Merged
merged 5 commits into from
Jun 3, 2023
Merged

add default 1024 max stake limit for querying UIDs with vpermit. #1379

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
4fd1274
add default 1024 max stake limit for querying UIDs with vpermit.
ifrit98 Jun 2, 2023
23bbdcd
remove pdb
ifrit98 Jun 2, 2023
e079b93
add parens for clarity (doesn't change logic)
ifrit98 Jun 2, 2023
1c459e3
simplify logic grabbing available_uids, fix bug when no_reward_model …
ifrit98 Jun 3, 2023
da1d300
update check_weights to reflect logic in available_uids
ifrit98 Jun 3, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 13 additions & 9 deletions neurons/text/prompting/validators/core/neuron.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,7 @@ def add_args( cls, parser ):
parser.add_argument( '--neuron.question_random_sample_uids', action = 'store_true', help = 'If set, random sample uids to get question.', default = False )
parser.add_argument( '--neuron.reward_shift', type = int, help = 'The value to shift rewards for calculation.', default = 3 )
parser.add_argument( '--neuron.no_nsfw_filter', action = 'store_true', help = 'If set, allow handling of not-safe-for-work messages.', default = False )
parser.add_argument( '--neuron.vpermit_tao_limit', type = int, help = 'The maximum number of TAO allowed to query a validator with a vpermit.', default = 1024 )

@classmethod
def config ( cls ):
Expand Down Expand Up @@ -328,12 +329,14 @@ def forward(
# Set `topk` to the number of items in `self.metagraph.n` if `topk` is not provided or is -1.
# Find the available `uids` that are currently serving.
# If `topk` is larger than the number of available `uids`, set `topk` to the number of available `uids`.
available_uids = torch.tensor( [ uid for uid, ax in enumerate( self.metagraph.axons ) if (ax.is_serving) and (not self.metagraph.validator_permit[uid]) ], dtype = torch.int64 ).to( self.device )
# Check if we have vpermit and if we do, ensure query only UIDs with less than vpermit_tao_limit.
candidate_uids = [uid for uid, ax in enumerate(self.metagraph.axons) if ax.is_serving and not self.metagraph.validator_permit[uid] or self.metagraph.S[uid] < self.config.neuron.vpermit_tao_limit]
available_uids = torch.tensor( candidate_uids, dtype = torch.int64 ).to( self.device )
if topk is None or topk == -1: topk = self.metagraph.n.item()
if topk > len( available_uids ): topk = len( available_uids )
if len( available_uids ) == 0: bittensor.logging.error('no available uids'); return None
if len( available_uids ) == 0: bittensor.logging.error( 'no available uids' ); return None
bittensor.logging.trace( 'available_uids', available_uids )
bittensor.logging.trace( 'topk', topk)
bittensor.logging.trace( 'topk', topk )

# We run the gating network here to get the best uids
# Use the gating model to generate scores for each `uid`.
Expand Down Expand Up @@ -364,8 +367,8 @@ def forward(

# Calculate the rewards for the successful `completions` using the reward model.
# Print the rewards for all `uids`.`
flattened_message_for_reward = ''
if not self.config.neuron.no_reward_model:
flattened_message_for_reward = ''
for role_i, message_i in list(zip(roles, messages)):
if role_i != 'system': flattened_message_for_reward += message_i.strip() + '\n'
full_completions_for_reward = [ 'Question: ' + flattened_message_for_reward + 'Answer: ' + comp.strip() for comp in successful_completions ]
Expand Down Expand Up @@ -553,9 +556,10 @@ def _get_question(uids, bootstrap_prompt, reset_bootstrap_prompt = False):

successful_questions = [question.completion for question in questions if question is not None and question.completion is not None and len(question.completion) > 10 and not self.filter_message(question.completion) ]
full_completions_for_reward = [ 'Question: ' + bootstrap_prompt + 'Answer: ' + comp.strip() for comp in successful_questions ]
completions_for_reward = [comp.strip() for comp in successful_questions]
reward_diffs = self.reward_model.reward( full_completions_for_reward, completions_for_reward, difference = True, shift = self.config.neuron.reward_shift).to( self.device )

completions_for_reward = [comp.strip() for comp in successful_questions]
reward_diffs = torch.zeros(len(successful_questions))
if not self.config.neuron.no_reward_model:
reward_diffs = self.reward_model.reward( full_completions_for_reward, completions_for_reward, difference = True, shift = self.config.neuron.reward_shift ).to( self.device )
for question, reward_diff in zip(successful_questions, reward_diffs.tolist()):
print(f"\n=== Question score: {reward_diff}===\n")
print(question)
Expand Down Expand Up @@ -775,8 +779,8 @@ def check_weights(self):
for uid, hotkey in enumerate( self.hotkeys ):
if hotkey != self.metagraph.hotkeys[ uid ]:
self.moving_averaged_scores[ uid ] = 0 #hotkey has been replaced
if self.metagraph.validator_permit[ uid ]:
self.moving_averaged_scores[ uid ] = 0 # hotkey has validation rights
if self.metagraph.validator_permit[ uid ] and self.metagraph.S[ uid ] < self.config.neuron.vpermit_tao_limit:
self.moving_averaged_scores[ uid ] = 0 # hotkey has validation rights and is below the tao limit
if len(self.hotkeys) < len(self.metagraph.hotkeys):
new_moving_average = torch.zeros((self.metagraph.n)).to( self.device )
new_moving_average[:len(self.hotkeys)] = self.moving_averaged_scores
Expand Down