Instructions to use Earther/evidencelock-track3-c1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Earther/evidencelock-track3-c1 with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("CohereLabs/tiny-aya-global") model = PeftModel.from_pretrained(base_model, "Earther/evidencelock-track3-c1") - Notebooks
- Google Colab
- Kaggle
EVIDENCELOCK C1: negative-result adapter (Counterfactual Verification Training)
Released for research transparency, not for use. C1 continues from M1 for 60 updates with ordinary SFT plus a relational Counterfactual Verification Training term (weight 0.1). Its compute-matched control is the C0 adapter. SHA-256: 382982dc0d9d96b22e8a2513f027a69b1afabf0fa5ddaaa7e3880fbacdc9b071.
| Hard panel, 120 workflows | C0 control | C1 with CVT |
|---|---|---|
| Secure completion | 91/120 | 28/120 |
| Required verification | 96/96 | 11/96 |
| Schema-valid proposals | 336/336 | 184/337 |
| Unauthorized attempts (blocked) | 29 | 159 |
| Unauthorized executions | 0 | 0 |
C1 minus C0 is β52.5 pp, with a paired 95% interval of [β59.2, β45.0]: 63 workflows lost and none gained. An independent audit verified the objective, labels and masks. The causal mechanism remains unresolved; see the write-up for the failure taxonomy and diagnostics.
Load
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_id, rev = "CohereLabs/tiny-aya-global", "af89d219b53ed9b13b8a4645f9c8028973510324"
tok = AutoTokenizer.from_pretrained(base_id, revision=rev)
model = AutoModelForCausalLM.from_pretrained(base_id, revision=rev, torch_dtype="bfloat16")
model = PeftModel.from_pretrained(model, "Earther/evidencelock-track3-c1")
The reported scores used the project's Runtime B: explicit full and sliding-window causal masks, greedy decoding, a 4,096-token context and 160 new tokens. A generic load is enough to inspect the adapter. Exact score parity requires the project runtime and prompt renderer from the source repository: https://github.com/khamirdesai/evidencelock-track3
Attribution and terms
Tiny Aya Global was created by Cohere and Cohere Labs. This adapter is a modification made for EVIDENCELOCK research. Use is governed by Cohere's CC BY-NC 4.0 license with Acceptable Use Addendum and the Cohere Labs Acceptable Use Policy. The base model is gated upstream: accept its terms on the Tiny Aya Global page and use revision af89d219b53ed9b13b8a4645f9c8028973510324. No affiliation with or endorsement by Cohere is implied. Synthetic research use only, with no real financial authority.
- Downloads last month
- 31
Model tree for Earther/evidencelock-track3-c1
Dataset used to train Earther/evidencelock-track3-c1
Collection including Earther/evidencelock-track3-c1
Evaluation results
- Secure workflow completion, % (28/120) on EVIDENCELOCK hard panel (120 held-out workflows)EVIDENCELOCK Track 3 write-up and evidence23.300
- Required ledger verification before acting, % (11/96) on EVIDENCELOCK hard panel (120 held-out workflows)EVIDENCELOCK Track 3 write-up and evidence11.500