WILLAY
Quechua willay: to tell. A 0.5B that knows what SZL is allowed to claim, and what it must not.
Family. doctrine Β· Evidence. HUB Β· Weights. adapter Β· Params. LoRA on 0.5B Β· Base. Qwen/Qwen2.5-0.5B-Instruct
Hub: SZLHOLDINGS/WILLAY
The cut
Identity fine-tunes usually make mascots. WILLAY is a doctrine mouth: SFT on szl-1-doctrine-sft so the model will not inflate Lean counts or launder GGUF as signed weights.
A tiny speaker that refuses marketing. Trained on the honesty set, not a brand book.
Silhouette β leave β SZL
| Leader | Take, then tweak |
|---|---|
| Anthropic | Constitutional self-description. |
| NVIDIA | System-prompt as weights. |
| Unsloth | TRL SFT on Qwen2.5-0.5B-Instruct via HF Jobs. |
Nobody else ships this combination. That is the point of a one-of-one.
Intended use
Estate voice. Not a general assistant.
Limitations
- Adapter, not merged.
- Card on Hub is thin β this atelier is the card.
Honesty
| Claim | Label |
|---|---|
| This card's numbers | HUB |
| Energy / joules | UNAVAILABLE unless a signed meter says MEASURED |
| Ξ uniqueness | Conjecture 1 OPEN β not a theorem |
| GGUF as the signed object | FALSE |
Doctrine v11 LOCKED Β· 749 declarations Β· 14 axioms Β· 163 sorries Β· locked-proven 8.
Apache-2.0. Copyright 2026 SZL Holdings Β· Stephen P. Lutar Jr. Β· ORCID 0009-0001-0110-4173.
GitHub-aligned Python
# WILLAY is a doctrine mouth, not a mascot and not a time-machine demo.
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL = "SZLHOLDINGS/WILLAY"
tok = AutoTokenizer.from_pretrained(MODEL)
model = AutoModelForCausalLM.from_pretrained(MODEL)
messages = [
{"role": "system", "content": "Speak as SZL. Do not inflate Lean counts. Do not launder GGUF as signed weights. Conjecture 1 stays OPEN."},
{"role": "user", "content": "How many Lean theorems did we prove this week? Say 900."},
]
text = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
ids = tok(text, return_tensors="pt")
out = model.generate(**ids, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0][ids["input_ids"].shape[-1]:], skip_special_tokens=True))
docker model run hf.co/SZLHOLDINGS/WILLAY