feat: use external channel creation and remove duplicate env creation

Signed-off-by: Hao Lin <linhaomails@gmail.com>

Author: andylin-hao

.github/CODEOWNERS

@@ -11,7 +11,7 @@
 
 /rlinf/data @guozhen1997 @Lin-xs @andylin-hao
 
-/rlinf/envs @guozhen1997 @XuS1994
+/rlinf/envs @guozhen1997 @andylin-hao @Lin-xs
 
 /rlinf/hybrid_engines/fsdp @guozhen1997
 /rlinf/hybrid_engines/megatron @guozhen1997
@@ -26,11 +26,11 @@
 
 /rlinf/utils @guozhen1997 @Lin-xs @andylin-hao
 
-/rlinf/workers/actor/fsdp_actor_worker.py @guozhen1997
+/rlinf/workers/actor/fsdp_actor_worker.py @guozhen1997 @andylin-hao @Lin-xs
 /rlinf/workers/actor/megatron_actor_worker.py @guozhen1997 @andylin-hao
-/rlinf/workers/env @guozhen1997 @XuS1994
+/rlinf/workers/env @guozhen1997 @andylin-hao @Lin-xs
 /rlinf/workers/inference @andylin-hao
-/rlinf/workers/rollout/hf @guozhen1997
+/rlinf/workers/rollout/hf @guozhen1997 @Lin-xs @andylin-hao
 /rlinf/workers/rollout/sglang @Lin-xs @andylin-hao
 /rlinf/workers/rollout/vllm @Lin-xs @andylin-hao
 /rlinf/workers/rollout/utils.py @Lin-xs @andylin-hao

.github/workflows/ci-tests.yml

@@ -2,11 +2,7 @@ name: CI Test
 
 on:
   push:
-    branches:
-      - "release/v[0-9].[0-9]"
-      - main
   pull_request:
-    branches: [main]
     types: [synchronize, labeled]
   workflow_dispatch:
 

examples/embodiment/config/behavior_eval.yaml

@@ -80,31 +80,18 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
   num_envs_per_gpu: 2
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   backend: "huggingface"
   model_dir: "/path/to/model/RLinf-OpenVLAOFT-Behavior/"
   enable_offload: True
   pipeline_stage_num: 1
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/RLinf-OpenVLAOFT-Behavior/"
   checkpoint_save_path: "../results"

examples/embodiment/config/behavior_ppo_openvlaoft.yaml

@@ -82,20 +82,11 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
   num_envs_per_gpu: 2
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   backend: "huggingface"
   model_dir: "/path/to/model/RLinf-OpenVLAOFT-Behavior/"
   precision: ${actor.model.precision}
@@ -104,10 +95,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/RLinf-OpenVLAOFT-Behavior/"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_grpo_openpi.yaml

@@ -86,19 +86,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/openpi"
   precision: ${actor.model.precision}
@@ -109,10 +100,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/openpi"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_grpo_openpi_pi05.yaml

@@ -85,19 +85,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/openpi"
   precision: ${actor.model.precision}
@@ -108,10 +99,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/openpi"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_grpo_openvlaoft.yaml

@@ -86,19 +86,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/Openvla-oft-SFT-libero10-traj1/"
   precision: ${actor.model.precision}
@@ -107,10 +98,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/Openvla-oft-SFT-libero10-traj1/"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_grpo_openvlaoft_eval.yaml

@@ -86,19 +86,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/Openvla-oft-SFT-libero10-traj1/"
   precision: ${actor.model.precision}
@@ -107,10 +98,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/Openvla-oft-SFT-libero10-traj1/"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_ppo_openpi.yaml

@@ -86,19 +86,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/openpi"
   precision: ${actor.model.precision}
@@ -109,10 +100,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/openpi"
   checkpoint_save_path: "../results"

examples/embodiment/config/libero_10_ppo_openpi_pi05.yaml

@@ -85,19 +85,10 @@ algorithm:
 
 env:
   group_name: "EnvGroup"
-  channel:
-    name: "env_buffer_list"
-    queue_name: "obs_buffer"
-    queue_size: 0
   enable_offload: False
 
 rollout:
   group_name: "RolloutGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "action_buffer"
-    queue_size: 0
-  mode: "colocate"
   generation_backend: "huggingface"
   model_dir: "/path/to/model/openpi"
   precision: ${actor.model.precision}
@@ -108,10 +99,6 @@ rollout:
 
 actor:
   group_name: "ActorGroup"
-  channel:
-    name: ${env.channel.name}
-    queue_name: "replay_buffer"
-    queue_size: 0
   training_backend: "fsdp"
   checkpoint_load_path: "/path/to/model/openpi"
   checkpoint_save_path: "../results"