diff --git "a/exp2a_swap_analysis/logs/nvila/roborefer.log" "b/exp2a_swap_analysis/logs/nvila/roborefer.log" new file mode 100644--- /dev/null +++ "b/exp2a_swap_analysis/logs/nvila/roborefer.log" @@ -0,0 +1,1253 @@ +2026-02-13 22:47:40,966 - INFO - +=== Loading & Creating Swap Pairs === +2026-02-13 22:47:43,784 - INFO - Swap pair creation stats: +2026-02-13 22:47:43,785 - INFO - left: 616/616 +2026-02-13 22:47:43,785 - INFO - right: 620/620 +2026-02-13 22:47:43,785 - INFO - above: 596/596 +2026-02-13 22:47:43,785 - INFO - under: 602/602 +2026-02-13 22:47:43,785 - INFO - far: 594/594 +2026-02-13 22:47:43,785 - INFO - close: 612/612 +2026-02-13 22:47:43,785 - INFO - Total pairs: 3640 +2026-02-13 22:47:44,012 - INFO - PyTorch version 2.3.0 available. +2026-02-13 22:47:44,567 - INFO - Loading HF dataset: FlagEval/EmbSpatial-Bench +2026-02-13 22:48:00,016 - INFO - Built bbox cache: 3640 entries +2026-02-13 22:48:00,031 - INFO - Cross-group quads: 0/1206 (ambiguous=0, no_bbox=1206) +2026-02-13 22:48:00,032 - INFO - +============================================================ +2026-02-13 22:48:00,032 - INFO - Processing nvila - roborefer +2026-02-13 22:48:00,032 - INFO - Model path: /data/shared/Qwen/mydisk/RoboRefer_model +2026-02-13 22:48:00,032 - INFO - ============================================================ +[2026-02-13 22:48:03,779] [INFO] [real_accelerator.py:110:get_accelerator] Setting ds_accelerator to cuda (auto detect) +2026-02-13 22:48:08,615 - INFO - We will use 90% of the memory on device 0 for storing the model, and 10% for the buffer to avoid OOM. You can set `max_memory` in to a higher value to use more memory (at your own risk). +You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`. +2026-02-13 22:48:42,438 - INFO - Loaded RoboRefer from /data/shared/Qwen/mydisk/RoboRefer_model +2026-02-13 22:48:42,439 - INFO - Model has 28 layers. Extracting ALL. +2026-02-13 22:48:42,440 - INFO - +--- Phase A: Extracting swap pair features --- + Swap pairs: 0%| | 0/1200 [00:00