SZL Holdings β€” governed, receipted, verifiable

Control before action. Evidence after.

Product Β· Proof Β· GitHub Β· HF org

WILLAY

Canonical family model β€” research fine-tune for SZL self-description / identity drafting on Qwen/Qwen2.5-0.5B-Instruct (root adapter + separate Unsloth adapter + derived merge).

Field Value
Status Historical qualification FAIL Β· deployment blocked in reviewed status
License Apache-2.0
Doctrine v11 LOCKED Β· Ξ› = Conjecture 1 (advisory)

What it does

Produces SZL-focused descriptive drafts. Objective β‰  proven doctrine adherence, refusal reliability, or factual accuracy.

What it does not do

  • Establish proof counts or artifact-signing claims from model text alone
  • Clear consent / privacy / lineage blocks in status.json / provenance.json
  • Interchange root adapter with adapter-unsloth/ (different bases/ranks)

Intended use

  • Research PEFT attach at pinned revision
  • Studying identity-SFT failure modes with honest FAIL receipts
  • Packaging / merge-receipt inspection

Out of scope

  • Production or autonomous identity speaker
  • Treating model output as evidence doctrine
  • Floating-base inference without immutable base pin
  • Laptop-lora sibling as canonical

Training data

Dataset Role
SZLHOLDINGS/doctrine-v10-v11 Doctrine adjacency
SZLHOLDINGS/szl-1-doctrine-sft Doctrine SFT adjacency
SZLHOLDINGS/why-we-lead Narrative adjacency

Exact train mix digests remain receipt-bound; Hub ids above are links, not a claim that every row entered gradients.

Evaluation (MEASURED or UNKNOWN)

Suite Result Class
Post-publication held-out (2026-09-15, rev 375a007e…) 0/5 passed MEASURED FAIL
Refusal vs baseline 0.0000 vs 0.4000 MEASURED FAIL
publicationAuthorization / productionAuthorization false MEASURED
Current Hub rev e04bd728… ≑ failed eval rev β€” UNKNOWN (not established byte-equivalent)
Energy / runtime improvement β€” UNAVAILABLE (explicitly not claimed)

Energy / BOM

Field Status
Energy UNAVAILABLE
BOM bom/model-bom.cdx.json when listed β€” not an energy meter
Merge receipt Present; unsigned; base named main@merge-time in places β€” reconstruction limits apply
DSSE Prefer real signature or honest UNSIGNED

Usage (research)

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_id = "Qwen/Qwen2.5-0.5B-Instruct"
base_rev = "<BASE_PIN>"
adapter = "SZLHOLDINGS/WILLAY"
adapter_rev = "<PINNED_SHA>"
tok = AutoTokenizer.from_pretrained(base_id, revision=base_rev)
base = AutoModelForCausalLM.from_pretrained(base_id, revision=base_rev, device_map="auto")
model = PeftModel.from_pretrained(base, adapter, revision=adapter_rev)

Select root vs Unsloth adapter vs merge explicitly. Do not overwrite canonical serve tags (chaski, SZL-Khipu-1.5B, szl1).

Source of truth

Limitations

  • FAIL receipts retained β€” do not launder into PASS
  • Base revision null in adapter configs β†’ reconstruction risk
  • Candidate enterprise / laptop-lora ids are not this SKU

Doctrine pin

  • Doctrine: v11 LOCKED
  • Ξ› (Lambda): Conjecture 1 β€” advisory, never a theorem
  • Evidence law: MEASURED or null / UNKNOWN / UNAVAILABLE β€” never fabricate energy, eval, or DSSE
  • Motto: Control before action. Evidence after.

License

Apache-2.0. See LICENSE in-repo when present.

Upgrade wave note

NVIDIA-style card upgrade draft Β· 2026-11-04T17:41Z Β· clean-room silhouette (DSSE receipts + MEASURED-or-null energy). Does not authorize train, serve, or autonomy. Does not overwrite sibling SKUs.

The estate

Surface Link
Company szlholdings.com
Sponsors Become a sponsor
GitHub szl-holdings
Receipts chain frontier-54-runs
Live dashboard SI command center
Downloads last month
168
Safetensors
Model size
0.5B params
Tensor type
F32
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ 1 Ask for provider support

Model tree for SZLHOLDINGS/WILLAY

Adapter
(1,043)
this model
Finetunes
1 model
Quantizations
1 model

Datasets used to train SZLHOLDINGS/WILLAY