YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Capcomp (Bittensor SN103) — certified adapter pool

The frozen pool every recipe composes from. Pinned by source_snapshot_sha256 = sha256:e88c0493a3aa66f75cd7ebafc34c4897c394787e026d350cdc169ea945bfec7e, against base model Qwen/Qwen3-8B@b968826d9c46dd6066d109eabc6255188de91218.

Published so that miners can build against the same weights the engine merges, and so that validators can reconstruct and measure candidates themselves rather than taking the operator's scores on trust.

The pool

adapter capability licence weights sha256
action-planner-v1 tool_calling apache-2.0 here f59786214371…
chained-reasoning-v1 fault_reasoning apache-2.0 here 6cd03b42d84b…
code-generation-v1 python_code_generation apache-2.0 here 3971d8091491…
constrained-selection-v1 resource_decision apache-2.0 here a71038a803f7…
cot-oracle-ablation-stride5-3layers-v1 unmeasured apache-2.0 here 9e4e4cc7f277…
creative-writing-v1 distractor_creative_writing · distractor apache-2.0 here 91baca52147c…
document-extraction-v1 structured_log_extraction apache-2.0 here 086c4b85d3ae…
embedded-engineering-v1 technical_domain_language apache-2.0 here a108ab310db6…
general-instruction-v1 verbose_self_explanation · distractor apache-2.0 here acf32d10aa72…
industrial-ifc-v1 industrial_asset_schema apache-2.0 here 95d0500f2aeb…
legal-citation-v1 distractor_legal_citation · distractor apache-2.0 here 3982b28827f8…
metabollm-qwen3-8b-v1 unmeasured apache-2.0 here 1ed3ae09714d…
parallel-reasoning-v1 general_reasoning apache-2.0 here cec095c2dc97…
q3-8b-z-plus-arkts-sft-lora-mpj-0430-v1 unmeasured unstated source 117d167d8d6f…
qwen-devops-foundation-lora-v1 unmeasured apache-2.0 here c5a273abbe21…
qwen3-8b-cpt-adapter-v1 unmeasured apache-2.0 here c7e4e7372976…
qwen3-8b-hip-adapter-v1 unmeasured apache-2.0 here a7d68c7746c3…
qwen3-8b-knowledge-test-v1 unmeasured apache-2.0 here 7291ea0a1b27…
qwen3-8b-lora-bioasq-resplit-v1 unmeasured apache-2.0 here 19ad1322e870…
qwen3-8b-lora-narrativeqa-v1 unmeasured apache-2.0 here 466203545fb3…
qwen3-8b-orchestrator-lora-successful--v1 unmeasured apache-2.0 here f1476fc1bffa…
qwen3-8b-orchestrator-lora-with-failur-v1 unmeasured apache-2.0 here fa69c1aa86db…
qwen3-8b-persistent-knapsack-lora-seed-v1 unmeasured apache-2.0 here 3a330078ca9c…
qwen3-8b-stateless-knapsack-lora-seed7-v1 unmeasured apache-2.0 here eed6808f3b06…
qwen3-8b-terminal-agent-fix-git-overfi-v1 unmeasured unstated source c735ac90c646…
qwen3-8b-test-v1 unmeasured apache-2.0 here 5b270a66b9bb…
reasoning-gen-8b-lora-v1 unmeasured apache-2.0 here 45e6e36decae…
structured-explanation-v1 structured_output apache-2.0 here 242cec8c5fa3…
writing-rlvr-qwen3-8b-lora-v1 unmeasured unstated source 92661027d725…
writing-rlvr-v2-qwen3-8b-lora-v1 unmeasured unstated source d4dee610b30e…

26 of 30 adapters are mirrored here under a licence that permits it, with attribution to the original author in each directory.

4 are listed but not mirrored. Their source repositories declare no licence, and absent a grant we do not republish someone's weights. Fetch those four from the repositories linked above — the same places we took them from — and check them against the digests in pool.json. The pool is complete either way; only the hosting differs.

Using it

git clone <repository-url> && cd capability-composition-subnet
pip install -e .

capcomp pool                 # ids, capabilities, distractors, the snapshot digest

Distractors are in the pool on purpose. Recognising that a plausible-looking adapter actively hurts is part of the composition problem, so they are marked rather than hidden.

Checking what you downloaded

pool.json carries a sha256 for every adapter's weights, including the four hosted elsewhere. An adapter whose digest does not match is not the one the engine merges, and a recipe built on it will not reproduce.

Attribution for every adapter — original repository, revision and licence — is in pool.json and in each mirrored directory's SOURCE.md.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support