Commit d2db1736 authored by williamzhangNU's avatar williamzhangNU
Browse files

minor udpate

parent 7363b15f
Loading
Loading
Loading
Loading
+5 −5
Changes for README.md: 5 added lines, 5 removed lines.
Original line number Diff line number Diff line
@@ -18,12 +18,12 @@ git submodule update --init --recursive


## TODO
1. Implement the validation in ray_trainer
2. Transfer the old ray_trainer file from RAGEN to VAGEN
1. Remove <box_start> and <box_end> in the rollout
2. Loss mask
3. Implement the validation in ray_trainer
4. Transfer the old ray_trainer file from RAGEN to VAGEN
    - Implement the metric for wandb
3. Add environment specific metrics for wandb logging
4. Remove <box_start> and <box_end> in the rollout
5. Loss mask
5. Add environment specific metrics for wandb logging

## NOTE
Does not support use_dynamic_bsz for now
 No newline at end of file
+0 −1
Changes for vagen/mllm_agent/rollout.py: 0 added lines, 1 removed line.
Original line number Diff line number Diff line
@@ -301,7 +301,6 @@ class QwenVLRolloutManger():
        
        start_step = max(0, step - window_size) if window_size is not None else 0
        end_step = step
        print(len(recording), f'start_step: {start_step}, end_step: {end_step}, step: {step}, window_size: {window_size},')
        assert len(recording) >= max(1, end_step + 1 - start_step), 'History length is not enough'
        history = recording[start_step: end_step + 1]