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_a7
LoRA adapter (r = 16) on CohereLabs/tiny-aya-global. A7 warm start: the A6 SFT rows plus 20 store-credit workflow demonstrations.
GRPO arms in the ActionProof collection were trained as a fresh LoRA on this adapter merged into the base.
Design and registration: PREREGISTRATION_A7.md in the code repository.
Greedy evaluation on test_iid (n = 200): verified-success completion (VSC) 0.750; post-write refund verification 0.60 (n = 155). Source: results/eval/a7/a7_init__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-a7")
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.