- Adds Direct Optimization (DPO) style rewards by @opentaco on #99
- Changes print format on exception catch by @camfairchild on #135
- Brings back netuid and wandb to logged config by @p-ferreira on #137
- Adds DPO penalty update by @Eugene-hu on #138
- Adds original reward output to wandb logs by @isabella618033 on #139
- Reweights reward models by @Eugene-hu on #140
- Update stale documentation by @steffencruz on #129
Full Changelog: https://github.com/opentensor/validators/compare/v1.1.7...v1.2.0
- Make sure to serve axon first by @camfairchild in 14921d35c
- Adds scripts for releases on github by @camfairchild in #128
- Wandb config log changes @isabella618033 in #132
- Hotfix cutoff limit by @Eugene-hu in #126
- Diversity regularization by @isabella618033 in #124
- Adds new keywords for the task validator by @p-ferreira in #119
- Save historic embeddings on disk by @opentaco in #121
- Updates relevance mechanism by @Eugene-hu in #122
- HOTFIX: create and serve the axon at startup by @robertalanm in #120
- Adds subtensor to metagraph sync by @camfairchild in #79
- Fix wandb weights format logging by @p-ferreira in #88
- Adds netuid tag to wandb runs by @p-ferreira in #95
- Implements GPU cleaning for optmization by @Eugene-hu in #96
- Adds compatibility with bittensor 5.3.3 by @camfairchild in #107
- Adds historic diversity component by @isabella618033 in #111
- Improvements on diveristy model by @isabella618033 and @Eugene-hu in #111
- Prompt improvements by @mrseeker in #110 and @p-ferreira in #112
- Adds Task Validator Filter to reward pipeline by @p-ferreira in #112
- Fix for empty data retrieval from datasets by @p-ferreira in #113
- Deprecates pip usage by @p-ferreira in #114