Releases: macrocosm-os/folding
Taofold V2.2.0
Taofold V2.1.4
What's Changed
- Added tenacity retries to metagraph calls by @Sarkosos in #378
- Add monitoring of job creation by @Sarkosos in #377
- bug fixes
Full Changelog: V2.1.3...v2.1.4
TaoFold v2.1.3
This upgrade includes the following changes:
- Instantiate new miner in credibility functionality if it does not exist.
- Ensure that all lists are the same length in
get_energies
before iterating. - Add an extra check to restart rqlite if the monitoring function fails.
- Validators point to a new S3 bucket for storage and link generation for global job pool.
- Increase spec version to 2.1.3.
Taofold v2.1.1
This update includes the following changes from #363 :
- Fix the reward loop in
neurons/validator.py
and add error handling. - Refactor the
Protein()
class to use absolute paths for file handling. - Add a new parameter to the
run_step()
method infolding/validators/forward.py
to avoid unnecessary processing done by the validator. Carry over this parameter toneurons/validator.py
in theforward()
method. - Decrease the epoch length for weight setting from
300
blocks to150
. - Maintain the ordering of the events for rewarding in
get_energies()
fromfolding/validators/reward.py
Taofold v2.0.0
What's Changed
- Create MinerRegistry and Evaluation Pipelines by @mccrindlebrian in #334
- Validator rewrite for V0 of GJP release by @Sarkosos in #342
- Add documentation by @mccrindlebrian and @schampoux in #321 , #348 , #352
- Inactive job management by @mccrindlebrian in #349
- Credibility by @mccrindlebrian in #346
- Job Store rewrite by @Sarkosos in #347
- S3 Handler relocation by @mccrindlebrian in #351
- Reward sharing via GJP by @mccrindlebrian in #350
- Base miner rewrite by @schampoux in #344
Taofold V1.4.6
Taofold 1.4.6 is released 🎉
This update is mostly for decreasing volatility of V-trust in the network. We are increasing the epoch length to 300 blocks. We are increasing the threshold of error allowed in the state-checkpoint check to 2% and we are lowering the severity of the punishment of failing the state-checkpoint check from 0 rewards to 0.5 rewards.
Important note
In this update we are also adding new utility functions for miners in anticipation for the global job pool release.
- We have added a script called
start_read_node.sh
that will start a read node that connects to the global job pool.- This will make a local copy of the global job pool and update it automatically.
- We have scripts to allow you to query this database called
query_rqlite.py
Please get accustomed to these scripts as they will be very useful in the future.
Validators
Perform the usual update process
git pull
poetry install
Miners
No update required but please get accustomed to the new scripts.
Taofold V1.4.5
What's Changed
- Features/remove miner serving by @mccrindlebrian in #332
Full Changelog: v1.4.4...v1.4.5
TaoFold v1.4.4
Preventative measure for checkpoint file validation:
- Check miner energies based on position and velocity.
- If these diverge from submitted checkpoint file energies, reward is set to 0.
TaoFold v1.4.2
Major change:
- Any malicious keys should be set to 0 incentive.
- All submissions to a job should be unique. If they are not, all non-unique miners will receive 0 incentive.
TaoFold v1.4.1
This release includes:
- Additional tests for s3 functionality.
- Updates to
get_epistula_body
logic to ensure Global Job Pool updates are interpreted correctly. - Black formatting for the whole repository with an update to increase the line length parameter.
- validator blacklisting for maliciously acting coldkeys