Official agent-miner image (bittensor==9.9.0) can't talk to the live chain — NetUid encoding mismatch

[davion-knight]

Summary

The published redteamsubnet61/agent-miner:3.1.8-260713 image (currently the latest tag) pins bittensor==9.9.0 via redteam_core@4.7.4's requirements.txt. That version can't correctly encode netuid for the current live subtensor chain's runtime metadata, so the miner crashes on startup, every time, for any miner:

ValueError: Invalid type for data: 61 of type <class 'int'>, type_def: Composite(TypeDefComposite { fields: [Field { name: None, ty: UntrackedSymbol { id: 42, marker: PhantomData<fn() -> core::any::TypeId> }, type_name: Some("u16"), docs: [] }] })

Reproduced directly via:

import bittensor as bt
st = bt.Subtensor(network="finney")
st.neurons_lite(netuid=61)  # <- raises the above

bittensor==9.9.0's neurons_lite() passes netuid as a bare int (params=[netuid]), but the live chain's runtime metadata now expects it as a Composite-wrapped NetUid type, not a raw u16. This is a client/chain version-skew bug, not a config issue — happens with a completely default .env/wallet setup.

This is very likely why the dependabot PRs attempting to bump bittensor (#111 → ~=10.2.0, #102 → ~=10.1.0, #96 → ~=10.0.1, all in RedTeamSubnet/RedTeam) were closed unmerged — see the fix section below for why a naive version bump doesn't work cleanly.

Verified fix

Confirmed working end-to-end (metagraph syncs, agent correctly self-identifies its UID, reaches full startup) by upgrading three packages together and patching four renamed API calls:

1. bittensor==10.5.0 (latest at time of writing) — required to encode NetUid correctly for the current chain
2. Its natural transitive deps: async-substrate-interface==2.2.1, bittensor-drand==2.0.0
3. A scalecodec/cyscale namespace clash this produces, fixed by uninstalling both and reinstalling only cyscale fresh
4. bittensor 10.x renamed core classes lowercase→PascalCase with identical constructor kwargs (config=, wallet=, port=), so src/miner/_base.py needs:
   • bt.wallet( → bt.Wallet(
   • bt.subtensor( → bt.Subtensor(
   • bt.dendrite( → bt.Dendrite(
   • bt.axon( → bt.Axon(

This is likely the real reason the earlier bittensor-bump PRs stalled — a plain requirements.txt version bump alone isn't sufficient; it also needs the four renamed calls in _base.py (and presumably similar renames in the validator repo) fixed, plus the scalecodec/cyscale conflict resolved.

Local patch Dockerfile (extends the published image, no source changes needed to verify):

FROM redteamsubnet61/agent-miner:3.1.8-260713

RUN pip install --no-cache-dir "bittensor==10.5.0" && \
    pip uninstall -y scalecodec cyscale && \
    pip install --no-cache-dir cyscale

RUN sed -i \
    -e 's/bt\.wallet(/bt.Wallet(/' \
    -e 's/bt\.subtensor(/bt.Subtensor(/' \
    -e 's/bt\.dendrite(/bt.Dendrite(/' \
    -e 's/bt\.axon(/bt.Axon(/' \
    /app/agent-miner/miner/_base.py

With this, the agent gets past wallet/subtensor/dendrite/metagraph/axon setup and correctly reaches Running miner on uid: <N> — full startup, no crash.

Impact

Every miner running the current official image is affected — the container will crash-loop on startup regardless of wallet/config correctness.

Request

Could the bittensor pin in redteam_core's requirements.txt (and wherever the validator repo pins it) be bumped, along with the four renamed calls in _base.py/equivalent validator code? Happy to open a PR with this if useful.