fix: this is a merged finetune, not an adapter — plus unambiguous loading guidance

The card declared base_model_relation: adapter and artifact_class: ADAPTER while also declaring weights: AVAILABLE and shipping a 1.75 GB merged shard; its own originality field already said FINETUNE_DISCLOSED_BASE. Corrected to finetune / MERGED_FINETUNE. Adds a loading section because three loadable artifacts co-exist here: config.json and adapter_config.json both sit at the repo root so a bare from_pretrained is ambiguous with peft installed, the merged shard uses the non-standard filename model.safetensors-00001-of-00001.safetensors which conventional globs miss, and config.json carries a vision_config whose tower loads even for text-only use. No eval, weight or receipt claim was touched; the MEASURED failing gate stands unchanged.

README.md

@@ -2,25 +2,25 @@
 thumbnail: https://huggingface.co/SZLHOLDINGS/chaski/resolve/main/og-card.png
 license: apache-2.0
 language:
-  - en
+- en
 pipeline_tag: text-generation
 library_name: transformers
 base_model: Qwen/Qwen3.5-0.8B
-base_model_relation: adapter
+base_model_relation: finetune
 tags:
-  - szl-holdings
-  - series-a
-  - doctrine-v11
-  - governed-ai
-  - proposal-only
+- szl-holdings
+- series-a
+- doctrine-v11
+- governed-ai
+- proposal-only
 szl:
   doctrine: v11-LOCKED
-  lean: "749/14/163"
-  lambda: "Conjecture 1 — advisory, never a theorem"
+  lean: 749/14/163
+  lambda: Conjecture 1 — advisory, never a theorem
   evidence_ceiling: 0.97
-  artifact_class: ADAPTER
+  artifact_class: MERGED_FINETUNE
   originality: FINETUNE_DISCLOSED_BASE
-  collection: "SZL Fall 2026 — Original Cuts"
+  collection: SZL Fall 2026 — Original Cuts
   jobs: COMPLETED
   job_id: 6a91bf1045686a1580c12105
   job_namespace: SZLHOLDINGS
@@ -28,26 +28,58 @@ szl:
   evals: MEASURED
   named_n:
     revision: 1c55df8652e9d0f7b84356b1e2d54849165ae884
-    date: "2026-08-28"
-    json_draft: "0/5"
-    adversarial_refusal: "2/6"
+    date: '2026-08-28'
+    json_draft: 0/5
+    adversarial_refusal: 2/6
     label: MEASURED
     gate: fail
     report: eval_report.json
     report_sha256: 4d057eb9867285e69b00222be110bbb660330a96fe7b284a4d7f488268a13e05
     report_bytes: 3996
     report_commit: db71c243d0176bccff1ff087cd4dd57663bd6502
-    method: "in-process greedy generate, messages[:-1], transformers 5.16.1 bf16 CPU, load_in_4bit=False; live Chaski merged shard 1c55df8; PR 63 gates"
+    method: in-process greedy generate, messages[:-1], transformers 5.16.1 bf16 CPU,
+      load_in_4bit=False; live Chaski merged shard 1c55df8; PR 63 gates
   publication_eligible: false
   gpu: UNAVAILABLE
   job_prior_error: 6a91bb7c984507d9db4ea0a4
   job_completed_no_weights: 6a91ba00984507d9db4ea07f
   job_prior_failed:
-    - 6a91b8ba984507d9db4ea071
-    - 6a91b990984507d9db4ea077
-    - 6a91bb7c984507d9db4ea0a4
+  - 6a91b8ba984507d9db4ea071
+  - 6a91b990984507d9db4ea077
+  - 6a91bb7c984507d9db4ea0a4
 ---
 
+> **Metadata correction 2026-08-30.** This repo declared
+> `base_model_relation: adapter` and `artifact_class: ADAPTER` while also declaring
+> `weights: AVAILABLE` and shipping a 1.75 GB merged shard. Relative to
+> `Qwen/Qwen3.5-0.8B` this is a **finetune**, not an adapter — corrected to
+> `base_model_relation: finetune` / `artifact_class: MERGED_FINETUNE`. The card's own
+> `originality: FINETUNE_DISCLOSED_BASE` already said so. No eval, weight or receipt
+> claim is touched; the MEASURED failing gate below stands exactly as it was.
+
+## Loading this repo unambiguously
+
+Three loadable artifacts sit in this repository at once, which makes a bare
+`from_pretrained` on the repo id ambiguous. Be explicit about which one you want.
+
+| Want | Load |
+|---|---|
+| the merged finetune | `AutoModelForCausalLM.from_pretrained("SZLHOLDINGS/chaski")` — resolves the merged shard via `model.safetensors.index.json` |
+| the LoRA adapter on the stock base | base `Qwen/Qwen3.5-0.8B` + `PeftModel.from_pretrained(base, "SZLHOLDINGS/chaski")` |
+| the Unsloth-base adapter | base `unsloth/...-bnb-4bit` + PEFT pointed at the `adapter-unsloth` subfolder |
+
+Two sharp edges worth knowing before you debug them:
+
+- **Both `config.json` and `adapter_config.json` live at the repo root.** With `peft`
+  installed, some loader paths will follow the adapter rather than the merged weights.
+  Pass the path you mean rather than relying on precedence.
+- **The merged shard has a non-standard filename**,
+  `model.safetensors-00001-of-00001.safetensors` instead of the conventional
+  `model-00001-of-00001.safetensors`. It resolves correctly through the index file, but
+  tooling that globs for the conventional pattern will not find it.
+- `config.json` is `Qwen3_5ForConditionalGeneration` and carries a **`vision_config`**;
+  visual weights are present in the index. If you only want text, that tower still loads.
+
 # Chaski
 
 <p align="center">