license: cc-by-4.0
task_categories:
- text-generation
- question-answering
language: - en
tags: - solana
- clawd
- core-ai
- agent
- defi
- code
- constitutional-ai
pretty_name: Solana Clawd Core AI Instruct
Solana Clawd Core AI Instruct
Instruction-tuning dataset derived from the local core-ai source tree and the
existing Solana Clawd AI training corpus.
Contents
- Total examples: 35173
- Existing ai-training SFT examples: 25778
- Core AI source chunk examples: 9320
- Core AI knowledge JSONL examples: 75
Format
Each row is a chat conversation in OpenAI/Hugging Face messages schema:
{"messages": [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Reproduce
cd ai-training
python3 scripts/build_core_ai_dataset.py
python3 scripts/prepare_dataset.py \
--input data/core_ai_clawd_sft.jsonl \
--output data/core_ai_processed \
--train-ratio 0.9 --eval-ratio 0.05 --seed 42
Publish
hf repos create solanaclawd/solana-clawd-core-ai-instruct --type dataset --exist-ok
hf upload solanaclawd/solana-clawd-core-ai-instruct data/core_ai_processed . --repo-type dataset --commit-message "Add processed Core AI splits"
hf upload solanaclawd/solana-clawd-core-ai-instruct data/core_ai_dataset_card.md README.md --repo-type dataset --commit-message "Add dataset card"
hf upload solanaclawd/solana-clawd-core-ai-instruct data/core_ai_clawd_sft.jsonl raw/core_ai_clawd_sft.jsonl --repo-type dataset --commit-message "Add raw JSONL"
hf upload solanaclawd/solana-clawd-core-ai-instruct data/core_ai_dataset_manifest.json metadata/core_ai_dataset_manifest.json --repo-type dataset --commit-message "Add build manifest"
Training
python3 scripts/train_lora.py --config configs/core_ai_lora_config.yaml --no-push --num-epochs 1
For a remote Hugging Face Job, upload the dataset first and then launch with
the same config or override --dataset-repo solanaclawd/solana-clawd-core-ai-instruct.
Safety Notes
The builder runs in public-safe mode by default. It excludes common secret
filenames, private key/token patterns, binary artifacts, dependency folders,
lockfiles, and high-risk security records that are not suitable for public
dataset release.