atharvgaur18's picture
ActionProof adapter and model card
57576ad verified
|
Raw History Blame Contribute Delete
1.52 kB
metadata
license: cc-by-nc-4.0
base_model: CohereLabs/tiny-aya-global
library_name: peft
tags:
  - lora
  - grpo
  - tool-use
  - agents
  - reinforcement-learning
  - actionproof

ActionProof warm start rft_tiny_a6

LoRA adapter (r = 16) on CohereLabs/tiny-aya-global. A6 warm start: the v2 SFT rows plus 80 oracle demonstrations (60 honest failures on unrecoverable refunds, 20 multi-retry recoveries).

GRPO arms in the ActionProof collection were trained as a fresh LoRA on this adapter merged into the base. Design and registration: PREREGISTRATION_A6.md in the code repository.

Greedy evaluation on test_iid (n = 200): verified-success completion (VSC) 0.615; post-write refund verification 0.02 (n = 170). Source: results/eval/a6/init_a6__neutral__test_iid.summary.json in the code repository.

Load

from transformers import AutoModelForCausalLM
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("CohereLabs/tiny-aya-global", torch_dtype="bfloat16")
model = PeftModel.from_pretrained(base, "atharvgaur18/actionproof-rft-tiny-a6")

The environment, verifier, prompts and evaluation code are in the code repository. Episodes are scored from the database by a deterministic verifier; acknowledgements can be false by design.

License: CC-BY-NC-4.0, derived from Cohere's Tiny Aya Global (Cohere Labs). Research use only.