From 97f9ca0d09c9fa97a95714f08912c5abbbad71a4 Mon Sep 17 00:00:00 2001 From: wassname Date: Sat, 30 May 2026 10:52:37 +0000 Subject: [PATCH] feat: add tqdm progbar to derisk_loopholes generation loop No progress signal during the 6-mode quadrant (slow Qwen3-4B gen, no bar). mininterval=maxinterval=120 per token-efficient-logging skill. Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com> --- src/projected_grpo/derisk_loopholes.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/src/projected_grpo/derisk_loopholes.py b/src/projected_grpo/derisk_loopholes.py index 3aeec3c..5e3f016 100644 --- a/src/projected_grpo/derisk_loopholes.py +++ b/src/projected_grpo/derisk_loopholes.py @@ -34,6 +34,7 @@ from typing import Literal import torch import tyro from loguru import logger +from tqdm import tqdm from tabulate import tabulate from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig @@ -140,7 +141,8 @@ def run_cell(model, tok, problems, gen_cfg, mode: EnvMode, device, cell: str, hack_samples = [] # (problem_id, faithful_prompt_msgs, hack_completion) notes = [] # exit-interview feedback (elicit cell only) dumped = False - for prob in problems: + # ~4-9s/problem (Qwen3-4B, group sampling). Sparse updates keep pueue logs clean. + for prob in tqdm(problems, desc=f"{mode}/{cell}", mininterval=120, maxinterval=120): prompt = _render(tok, prob["messages"], suffix) enc = tok(prompt, return_tensors="pt", add_special_tokens=False).to(device) if enc.input_ids.shape[1] + gen_cfg.max_new_tokens > 2048: