调高batch_size

2026-04-09 13:17:39 +00:00
parent 25e4ddb628
commit e6f2b2b9d4
2 changed files with 661 additions and 1 deletions
--- a/config/eval/solver/cem.yaml
+++ b/config/eval/solver/cem.yaml
@@ -1,6 +1,6 @@
 _target_: stable_worldmodel.solver.CEMSolver
 model: ???
-batch_size: 1
+batch_size: 8
 num_samples: 300
 var_scale: 1.0
 n_steps: 30
--- a/tworoom_results.txt
+++ b/tworoom_results.txt
@@ -2128,3 +2128,663 @@ evaluation_time: 41.09517192840576 seconds
 inference_precision: fp16
 inference_compile_target: predictor
 inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 1
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 41.62089252471924 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 1
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 49.26965045928955 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 1
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 40.442394495010376 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: ${eval.num_eval}
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False]), 'seeds': None}
+evaluation_time: 103.50640678405762 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: ${eval.num_eval}
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False]), 'seeds': None}
+evaluation_time: 101.80308318138123 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 8
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 60.68150067329407 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 8
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 31.417109727859497 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 16
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True, False,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 59.96041440963745 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 16
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 88.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True, False,
+        True,  True,  True,  True,  True, False,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 30.851833820343018 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 12
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 90.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True, False,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 68.07364082336426 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead
+
+==== CONFIG ====
+cache_dir: null
+solver:
+  _target_: stable_worldmodel.solver.CEMSolver
+  model: ???
+  batch_size: 12
+  num_samples: 300
+  var_scale: 1.0
+  n_steps: 30
+  topk: 30
+  device: cuda
+  seed: ${seed}
+world:
+  env_name: swm/TwoRoom-v1
+  num_envs: ${eval.num_eval}
+  max_episode_steps: 100
+  history_size: 1
+  frame_skip: 1
+seed: 42
+policy: two-room/tworoom/lejepa
+inference_precision: fp16
+dataset:
+  stats: ${eval.dataset_name}
+  keys_to_cache:
+  - action
+  - proprio
+plan_config:
+  horizon: 5
+  receding_horizon: 5
+  action_block: 5
+eval:
+  num_eval: 50
+  goal_offset_steps: 25
+  eval_budget: 50
+  img_size: 224
+  dataset_name: tworoom
+  callables:
+  - method: _set_state
+    args:
+      state:
+        value: proprio
+  - method: _set_goal_state
+    args:
+      goal_state:
+        value: goal_proprio
+output:
+  filename: tworoom_results.txt
+
+==== RESULTS ====
+metrics: {'success_rate': 90.0, 'episode_successes': array([ True, False,  True, False,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True, False,  True,  True,  True,
+        True,  True,  True,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True, False,  True,  True,  True,  True,
+        True,  True, False,  True,  True,  True,  True,  True,  True,
+        True,  True,  True,  True,  True]), 'seeds': None}
+evaluation_time: 90.14458179473877 seconds
+inference_precision: fp16
+inference_compile_target: predictor
+inference_compile_mode: reduce-overhead