Fail to get MT10 upper bound

[niiceMing]

Description

we have successfully trained some algorithm on MT10. However,when we train a sac agent on MT1 to get "One SAC agent per task(upper bound)", it always fails due to critic loss is to high( >1e8), and the success rate is near 0%.
Is there any special config for MT1?

How to reproduce

we use the following config：

setup=metaworld
env=metaworld-mt1
agent=state_sac
experiment.num_eval_episodes=1
experiment.num_train_steps=2000000
setup.seed=10
replay_buffer.batch_size=1280
agent.multitask.num_envs=1
agent.multitask.should_use_disentangled_alpha=False
agent.encoder.type_to_select=identity
agent.multitask.should_use_multi_head_policy=False
agent.multitask.actor_cfg.should_condition_model_on_task_info=False
agent.multitask.actor_cfg.should_condition_encoder_on_task_info=True
agent.multitask.actor_cfg.should_concatenate_task_info_with_encoder=True

we change the default task_name in the function get_list_of_func_to_make_envs() ( src/mtenv/mtenv/envs/metaworld/env.py) to control the task uesd in MT1.

System information

• MTRL Version : latest
• Metaword Version : af8417bfc82a3e249b4b02156518d775f29eb289

.