Yuanzhao Zhai

A question about the approach

1

Hello! Thanks for your wonderful work. You predicted the next states of humans with the relational graph containing the robot. However, you also assumed that the robot is invisible to...

A small bug

1

I am reading your excellent code. I found a small bug that failing to load the imitation model. I suggest replacing model.load_state_dict_rl(torch.load(il_weight_file)) with policy.load_state_dict_rl(torch.load(il_weight_file)) in line 138 in train.py. Thanks!

A question about backpropagation

7

Hello! Thanks for your wonderful work. In the Optimization section of the paper, it said that "In order to optimize (2) via backpropagation, we need to compute a subgradient of...

Release of the Chinese Dataset.

1

Hello! Thanks for sharing the valuable experience. Do you plan to release the Chinese datasets?

大于三个作者的参考文献报错

1

感谢大佬分享，想请教一个问题： 使用\printbib 和 \printbibtabular[title={~}]时，参考文献只要大于三个，就会报如下错误： .10 \printbibtabular[title={~}] ./Tex/6_references.tex:10: Missing \cr inserted. \cr l.10 \printbibtabular[title={~}]

Zero win rate in SMAC scenario

2

Hello, thanks for your elegant code. When I directly run: python dicg_ce_smac_runner.py --map 3s_vs_5z python dicg_ce_smac_runner.py --map 6h_vs_8z the win rate is always zero. When I directly run: python dicg_ce_smac_runner.py...

Llama2 as actor using zero_stage3

1

Hello! Did anyone meet the following bug when using zero_stage3 for Lllama2? step3_rlhf_finetuning/rlhf_engine.py:61 in __init__ │ │ │ │ 58 │ │ self.num_total_iters = num_total_iters │ │ 59 │ │...

Hello! Have you implemented halfcheetah-jump and ant-angle tasks proposed in MOPO?

Hello! When I run the following command: "python -m eval_agent.main --agent_config openai --exp_config alfworld --split test --verbose" I got the results: All tasks done. Output saved to outputs/gpt-3.5-turbo/alfworld Average reward:...