============================================================ OPEN-ENDED EVAL: Base vs SFT (Multi-GPU) Base model: /workspace/rl4phyx/models/Qwen2.5-VL-3B-Instruct SFT model: /workspace/rl4phyx/RL4Phyx/SFT/checkpoints/sft_qwen25vl_3b_fullft/final Base GPUs: [0, 1, 2, 3] SFT GPUs: [4, 5, 6, 7] ============================================================ Loaded 1533 test samples Mechanics: 276 Electromagnetism: 275 Thermodynamics: 255 Waves/Acoustics: 253 Optics: 252 Modern Physics: 222 >>> Starting BASE model inference... [base][GPU 2] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [base][GPU 0] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [base][GPU 1] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [base][GPU 3] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored. The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored. Loading checkpoint shards: 0%| | 0/2 [00:00>> Starting SFT model inference... [sft][GPU 6] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [sft][GPU 4] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [sft][GPU 5] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [sft][GPU 7] Loading model... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.48, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored. The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored. Loading checkpoint shards: 0%| | 0/2 [00:00