Text Generation
PEFT
Safetensors
English
qwen2
qlora
governed-agent
proposal-only
research-only
szl-holdings
khipu
abstain-retrain
conversational
Instructions to use SZLHOLDINGS/KHIPU-R2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use SZLHOLDINGS/KHIPU-R2 with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit") model = PeftModel.from_pretrained(base_model, "SZLHOLDINGS/KHIPU-R2") - Notebooks
- Google Colab
- Kaggle
card: frontier credibility (datasets/Usage/Limitations) metadata-only 20261104T153547Z
Browse filesOutside Dev G — no weight changes; honest datasets YAML or undeclared note; short Usage/Limitations where audit MISSING.
README.md
CHANGED
|
@@ -49,6 +49,26 @@ szl:
|
|
| 49 |
|
| 50 |
# KHIPU-R2
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
A separate abstention-retraining experiment with adapter-scoped historical evaluation records.
|
| 53 |
|
| 54 |
**Artifact:** QLoRA adapter; separate merged checkpoint · **Stage:** Research · release blocked
|
|
@@ -173,6 +193,15 @@ verified. A license addition requires source-owner review of that coverage.
|
|
| 173 |
Canonical card authoring source: [khipu-r2/card/README.md](https://github.com/szl-holdings/szl-forge/blob/6f4ac90ff503265cec9c0436587f086e52d6124e/khipu-r2/card/README.md).
|
| 174 |
Doctrine v11 LOCKED 749/14/163. Λ = Conjecture 1, advisory, never a theorem.
|
| 175 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 176 |
<!-- SZL-CARD-TECHNICAL:v1:END -->
|
| 177 |
</details>
|
| 178 |
|
|
|
|
| 49 |
|
| 50 |
# KHIPU-R2
|
| 51 |
|
| 52 |
+
<!-- SZL-FRONTIER-CRED-20261104 -->
|
| 53 |
+
## Usage
|
| 54 |
+
|
| 55 |
+
PEFT/LoRA adapter — attach to the declared `base_model` at a pinned revision. Research-only unless a separate eval+serve gate says otherwise.
|
| 56 |
+
|
| 57 |
+
```python
|
| 58 |
+
from peft import PeftModel
|
| 59 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 60 |
+
base_id = "<BASE_FROM_CARD>"
|
| 61 |
+
base_rev = "<BASE_PIN>"
|
| 62 |
+
adapter = "SZLHOLDINGS/KHIPU-R2"
|
| 63 |
+
adapter_rev = "<PINNED_SHA>"
|
| 64 |
+
tok = AutoTokenizer.from_pretrained(base_id, revision=base_rev)
|
| 65 |
+
base = AutoModelForCausalLM.from_pretrained(base_id, revision=base_rev, device_map="auto")
|
| 66 |
+
model = PeftModel.from_pretrained(base, adapter, revision=adapter_rev)
|
| 67 |
+
```
|
| 68 |
+
|
| 69 |
+
Do not overwrite canonical serve tags (`chaski`, `SZL-Khipu-1.5B`, `szl1`).
|
| 70 |
+
|
| 71 |
+
|
| 72 |
A separate abstention-retraining experiment with adapter-scoped historical evaluation records.
|
| 73 |
|
| 74 |
**Artifact:** QLoRA adapter; separate merged checkpoint · **Stage:** Research · release blocked
|
|
|
|
| 193 |
Canonical card authoring source: [khipu-r2/card/README.md](https://github.com/szl-holdings/szl-forge/blob/6f4ac90ff503265cec9c0436587f086e52d6124e/khipu-r2/card/README.md).
|
| 194 |
Doctrine v11 LOCKED 749/14/163. Λ = Conjecture 1, advisory, never a theorem.
|
| 195 |
|
| 196 |
+
<!-- SZL-FRONTIER-CRED-20261104 -->
|
| 197 |
+
## Limitations
|
| 198 |
+
|
| 199 |
+
- Proposal / research artifact unless a separate signed eval+serve gate exists.
|
| 200 |
+
- Train loss ≠ evaluation; named-N gates may fail or be historical-only.
|
| 201 |
+
- Not authorized for autonomous execution or high-stakes deployment by this card alone.
|
| 202 |
+
- Pin revisions; floating `main` is not a serve contract.
|
| 203 |
+
- **datasets YAML undeclared:** no Hub `datasets:` id is asserted on this card. Local/receipt digests (jsonl SHA) are not Hub dataset ids; this wave does not invent them.
|
| 204 |
+
|
| 205 |
<!-- SZL-CARD-TECHNICAL:v1:END -->
|
| 206 |
</details>
|
| 207 |
|