SZLHOLDINGS/SZL-Khipu-1.5B-abstain

Upload khipu-f16.gguf with huggingface_hub

Upload folder using huggingface_hub

Upload eval_receipt.signed.json with huggingface_hub

Upload training_receipt.signed.json with huggingface_hub

Upload owner_pubkey.json with huggingface_hub

honesty: stamp NON-RELEASE / RESEARCH banner

Add per-repo CycloneDX BOM (canonical registry: datasets/SZLHOLDINGS/model-bom)

Add per-repo CycloneDX BOM (canonical registry: datasets/SZLHOLDINGS/model-bom)

fix: stop declaring a loadable PEFT adapter that has no weights

atelier: one-of-one card — cut, silhouette, GitHub Python. YAML preserved.

docs: stamp WEIGHTS UNAVAILABLE; successor is KHIPU-R2

docs: stamp WEIGHTS UNAVAILABLE; successor is KHIPU-R2

data: committed Khipu curriculum manifest.json (held-out never in gradients)

data: committed Khipu curriculum khipu.schema.json (held-out never in gradients)

data: committed Khipu curriculum adversarial.jsonl (held-out never in gradients)

data: committed Khipu curriculum eval.jsonl (held-out never in gradients)

data: committed Khipu curriculum train.abstain.jsonl (held-out never in gradients)

data: committed Khipu curriculum train.jsonl (held-out never in gradients)

chore: publication_eligible false until eval runs

chore: Apache-2.0 license

docs: adapter card base_model Qwen2.5-1.5B-Instruct publication_eligible false

feat: UV job script Unsloth QLoRA abstain retrain ABSTAIN_OVERSAMPLE=4

initial commit