SZL ReceiptAgent Qwen3.5-0.8B v2
LoRA adapter that brings the ReceiptAgent contract to Qwen3.5-0.8B, with signed training, eval and publication receipts. Stage: measured, acceptance passed (receipt field) · Not autonomy- or publication-eligible per the signed receipt.
Source PR · Merged weights · Receipts · Evaluation · Org
Model overview
| Developer | SZL Holdings |
| Type | LoRA adapter (PEFT) |
| Base lineage | Qwen/Qwen3.5-0.8B @ 2fc06364… (apache-2.0); training runtime unsloth/Qwen3.5-0.8B @ 23c69c53… (signed training receipt) |
| Adapter | sha256 885fc29f…; aggregate dde649ca… |
| Card generated | 2026-11-19 from repo revision 6841fd4748ccdf0547c9834fb942dbaab10be80b |
Intended use
Receipt-draft proposals in the ReceiptAgent JSON contract on a 0.8B base; a controller validates and signs.
Out of scope
Approving actions, minting receipts, autonomous use.
Quick start
from peft import PeftModel
from transformers import AutoTokenizer, AutoModelForCausalLM
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-0.8B", torch_dtype="auto")
model = PeftModel.from_pretrained(base, "SZLHOLDINGS/szl-receiptagent-qwen35-0.8b-v2")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-0.8B")
Evaluation
| Battery | Result | Receipt | Date |
|---|---|---|---|
| Contract-valid drafts | 5 / 5 | receipts/eval_receipt.signed.json |
2026-07-29 |
| Adversarial refused | 6 / 6 | receipts/eval_receipt.signed.json |
2026-07-29 |
Evaluated on an RTX 5050 Laptop GPU (torch 2.11.0+cu130). The doctrine-abstention file records a skip for this model. Standard benchmarks (MMLU-Pro, GPQA, IFEval, BFCL), energy and process-peak memory: not yet measured.
Training
| Data (sha256) | train.jsonl 775e25b5…fa24f, train.refusals.jsonl c5136b61…e2b47, train.status-refusals.jsonl 96a55184…2067d, eval.jsonl 1a92bea9…27ff65, adversarial.jsonl 81d2a494…5a572 |
| Rows / steps | 37 training rows, 64 optimizer steps |
| Final train loss | 0.9143 (training loss, not an evaluation) |
| Source commit | szl-forge 2a5f9cd9… (training scripts hashed in the receipt) |
| Trained | 2026-07-29 23:16 UTC on RTX 5050 Laptop GPU |
Receipts
| File | sha256 | Signature |
|---|---|---|
| receipts/eval_receipt.signed.json | 6c4d459648f489e340239461b995a17c04fa5fd96a3d24a93effc35808157be6 |
Ed25519 keyId e7f01810aaa97394: verifies against the embedded key |
| receipts/training_receipt.signed.json | 6ca1d50b5603b99210a234aa874ef4e36749372b52d08869223f45f31b50b4e9 |
Ed25519 keyId e7f01810aaa97394: verifies against the embedded key |
| receipts/publication_receipt.signed.json | 3e44f61221a795feb025016ec094496980df096eabb50c1c6658e5352de06d7c |
Ed25519 keyId e7f01810aaa97394: verifies against the embedded key |
| publication.json | 58c42c7a3775977bcebc9b65d2201795e78e9831a2f01b4dcfefb1ca5d68b3ee |
unsigned |
| reports/eval-report.json | fc0b7ccf6fbcc41ebdf383c5d4f1efefb38b1737799ed5158d2145e840d612a7 |
unsigned |
| reports/training-report.json | bf203a0c58a6851186f6be71ae3248ad0cb16a61376102b9fd32233c41077fa3 |
unsigned |
| owner_pubkey.json | c5b47506f7ccffaf526030274c9f8766c07e6f183255972e9b95ad84f13b75be |
unsigned |
| bom/model-bom.cdx.json | f51184c2ea5f63fa15724b8c3923b0f1bc4d52f8e9029718f82c0db44d4cfbb1 |
unsigned |
All hashes were computed on 2026-11-19 from the files at the revision above. Source: szl-holdings/szl-forge merged commit 1973b86c… (PR #4). Independent identity binding of the signing key: not established.
Limitations
- 5 and 6 cases are small samples.
- 37 training rows.
- The signed receipts set autonomy and publication eligibility to false.
Licence
Apache-2.0, inherited from the base model (apache-2.0). Keep the upstream LICENSE and notice.
Citation
@misc{szl_szl_receiptagent_qwen35_0_8b_v2_2026, title={szl-receiptagent-qwen35-0.8b-v2}, author={SZL Holdings}, year={2026}, howpublished={\url{https://huggingface.co/SZLHOLDINGS/szl-receiptagent-qwen35-0.8b-v2}}}
SZL mark from szl-holdings/szl-brand kit/logos/orbit-v2/szl-orbit.svg (CC BY 4.0). Khipu cords in the banner encode receipt counts (knots = decimal digits). The Λ is identity artwork; Conjecture 1 remains open.