Third-party provenance
Project code uses Apache 2.0. External inputs retain their own licenses and attribution requirements.
| Material | Source and treatment |
|---|---|
| CometBFT 0.38.26 and ABCI definitions | CometBFT, derived from Tendermint; Apache 2.0 license. The installer builds the engine separately. The minimal ABCI definitions and generated bindings preserve protocol compatibility. |
| Tiny Shakespeare | docs/eval/data/input.txt and its wheel copy match Karpathy's char-rnn corpus, SHA-256 86c4e6aa9db7c042ec79f339dcb96d42b0075e16b8fc2e86bf0ca57e2dc565ed. The upstream README identifies its license as MIT. The MIT terms are included. This distribution attributes the corpus preparation to Andrej Karpathy and the underlying text to William Shakespeare. No private user corpus is included. |
| SmolLM2-135M-Instruct | HuggingFaceTB model, revision 12fd25f77366fa6b3b4b768ec3050bf629380bac, Apache-2.0. Downloaded separately, SHA-256 checked and mirrored with attribution; pretrained weights are not inside the Python wheel. The trainable residual adapter is initialized by this repository. |
| SmolLM2-1.7B-Instruct | HuggingFaceTB model, revision 31b70e2e869a7173562077fd711b654946d38674, Apache-2.0. Downloaded separately for operated GPU research. Published modified full-model checkpoints preserve the upstream model card, license, source revision and experiment scope; weights are not inside the Python wheel. |
| BGE question reranker | BAAI/bge-reranker-v2-m3, revision 953dc6f6f85a1b2dbfca4c34a2796e7dde08d41e, Apache-2.0. Downloaded separately for question-family selection; no answer text enters its inputs. The published three-file checkpoint preserves all 393 original tensors exactly and includes the upstream model card. Model bytes are not in the Python wheel. |
| Smol-SmolTalk | HuggingFaceTB dataset, revision f73fe857d519ff6ac5af2ea67c4d3834da7b8bcc, Apache-2.0. The initial public dataset derives from 512 training records using the pinned model chat template. Preserve upstream dataset attribution and the execution manifest. |
| Python dependencies | Installed separately. LLM versions are pinned in docs/llm-requirements.txt; the earlier reference retains docs/demo-requirements.txt. Dependency versions are part of the execution compatibility assessment. |
Preserve provenance and required notices when adding code, data, models, or assets. Historical materials retain their original attribution and do not establish production readiness.
Website assets, fonts, npm dependencies and IEEEtran manuscript files are outside the current distribution. Their original notices remain with the archived publishing materials and in the historical repository revision; archiving does not change their licenses.