Release v3.0.3

Announcing Release 3.0.3

Two technical improvements are added with this release.

Changes

• Fixing truncation = False
• Adding system prompt
• Reset validator state

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v3.0.2

Announcing Release 3.0.2

Following feedback from the SN37 community, a few more bug are squashed with this release.

Changes

• Decreased eval_block_delay
• Fixed bug that kept evaluating models from inactive competitions.
• Fixed huge rain of useless logs.
• Validator state reset.

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v3.0.1

Announcing Release 3.0.1

Coming in hot behind release 3.0.0, improving the padding logic when using the model's own tokenizer.

Release v3.0.0

Announcing Release 3.0.0

The new DISTILLED_REASONING_3B competition is scheduled with this release.

Changes

At block 5_158_632 (~midnight GMT+0 on Wednesday, March 19, 2025):

• The current INSTRUCT_8B competition will be sunset.
• The new DISTILLED_REASONING_3B competition will start.

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v2.10.2

Announcing Release 2.10.2

Following feedback from the SN37 community, one more bug is squashed with this release.

Changes

• Removed unsupported logging exc_info argument.

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v2.10.1

Announcing Release 2.10.1

Following feedback from the SN37 community, two bugs are squashed with this release.

Changes

• Fixed regex for the MMLU parser
• Removed deprecated argument in check_for_reasonable_output() function call.
• Other small improvements.

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v2.10.0

Announcing Release 2.10.0

Following the drop in synthetic mmlu samples generated by SN1 (due to the increase in organic queries), our synthetic_mmlu and if_eval tasks were not fed enough samples to function properly. This is a hotfix that stabilises the subnet:

Changes

• New task weights: synthetic_mmlu 30%, word_sorting 5%, fineweb 35%, if_eval_v2 30%.
• Validators now sample 10 configs of the macrobench dataset and then samples the individual text samples from this subset.
• Common errors in subprocess handling addressed. Mainly the too many open files error.

Notes to validators

Please make sure to increase the number of open files limit before running the validator, then update the dependencies:

$ ulimit -n 64000
$ python -m pip install -e .

Release v2.9.2

Announcing Release 2.9.2

This is a hotfix that applies the following changes:

Change

Reduces min subnet 1 validator stake from 100k to 10k to increase the rate of synthetic benchmarks.

This change is necessary because of the subnet stake changes due to DTAO.

Release v2.9.1

Announcing release v2.9.1

Here are the main changes:

Changes

• A small fix to model upload script and related mining utility functions
• Bumped taoverse dependency to version 1.4.1

NOTES TO VALIDATORS

-Please also make sure to rerun pip install to ensure updated dependencies.
python -m pip install -e

Release 2.9.0

Announcing Release 2.9.0

This is the first release that support dTao migration with bittensor==9.0.0:

Changes

• bumped bittensor dependency version to 9.0.0 to support dTao release.
• Bumped taoverse dependency version to 1.4.0.

Validators should update as soon as they can. Note that due to requirement updates you will need to rerun

python -m pip install -e .