dodox/dppo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update eval configs

Browse Source
This commit is contained in:
allenzren 2025-01-03 17:22:38 -05:00 committed by Allen Z. Ren
parent ace2bbdab9
commit 169a16dda7
27 changed files with 81 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
cfg/d3il/eval/avoid_m1/eval_diffusion_mlp.yaml
View File

@ -19,6 +19,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 25
render_num: 40
@ -47,7 +48,8 @@ env:
reset_within_step: False
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
#

4
cfg/furniture/eval/lamp_low/eval_diffusion_mlp.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 8
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/furniture/eval/lamp_low/eval_diffusion_unet.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 16
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/furniture/eval/one_leg_low/eval_diffusion_mlp.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 8
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/furniture/eval/one_leg_low/eval_diffusion_unet.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 16
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/furniture/eval/round_table_low/eval_diffusion_mlp.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 8
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/furniture/eval/round_table_low/eval_diffusion_unet.yaml
View File

@ -21,6 +21,7 @@ horizon_steps: 16
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: ${eval:'round(${env.max_episode_steps} / ${act_steps})'}
render_num: 0
@ -41,7 +42,8 @@ env:
sparse_reward: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/gym/eval/halfcheetah-v2/eval_diffusion_mlp.yaml
View File

@ -19,6 +19,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 250 # each episode can take maximum (max_episode_steps / act_steps, =250 right now) steps but may finish earlier in gym. We only count episodes finished within n_steps for evaluation.
render_num: 0
@ -40,7 +41,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
#

4
cfg/gym/eval/hopper-v2/eval_diffusion_mlp.yaml
View File

@ -19,6 +19,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 250 # each episode can take maximum (max_episode_steps / act_steps, =250 right now) steps but may finish earlier in gym. We only count episodes finished within n_steps for evaluation.
render_num: 0
@ -40,7 +41,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
#

4
cfg/gym/eval/kitchen-v0/eval_diffusion_mlp.yaml
View File

@ -19,6 +19,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 70
render_num: 0
@ -40,7 +41,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
#

4
cfg/gym/eval/walker2d-v2/eval_diffusion_mlp.yaml
View File

@ -19,6 +19,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 250 # each episode can take maximum (max_episode_steps / act_steps, =250 right now) steps but may finish earlier in gym. We only count episodes finished within n_steps for evaluation.
render_num: 0
@ -40,7 +41,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
#

4
cfg/robomimic/eval/can/eval_diffusion_mlp.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/can/eval_diffusion_mlp_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/can/eval_diffusion_unet.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 75 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/can/eval_diffusion_unet_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/lift/eval_diffusion_mlp.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/lift/eval_diffusion_mlp_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/lift/eval_diffusion_unet.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 75 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/lift/eval_diffusion_unet_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 300 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/square/eval_diffusion_mlp.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 400 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/square/eval_diffusion_mlp_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 400 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/square/eval_diffusion_unet.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 4
act_steps: 4
ft_denoising_steps: 10
n_steps: 100 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -44,7 +45,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/square/eval_diffusion_unet_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 4
act_steps: 4
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 400 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -58,7 +59,8 @@ shape_meta:
shape: [7]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

5
cfg/robomimic/eval/transport/eval_diffusion_mlp.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 8
act_steps: 8
ft_denoising_steps: 10
n_steps: 400 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -46,9 +47,9 @@ env:
max_episode_steps: ${env.max_episode_steps}
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/transport/eval_diffusion_mlp_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 8
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 200 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -62,7 +63,8 @@ shape_meta:
shape: [14]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/transport/eval_diffusion_unet.yaml
View File

@ -20,6 +20,7 @@ denoising_steps: 20
cond_steps: 1
horizon_steps: 16
act_steps: 8
ft_denoising_steps: 10
n_steps: 100 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -47,7 +48,8 @@ env:
reset_within_step: True
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3

4
cfg/robomimic/eval/transport/eval_diffusion_unet_img.yaml
View File

@ -23,6 +23,7 @@ horizon_steps: 16
act_steps: 8
use_ddim: True
ddim_steps: 5
ft_denoising_steps: 5
n_steps: 400 # each episode takes max_episode_steps / act_steps steps
render_num: 0
@ -62,7 +63,8 @@ shape_meta:
shape: [14]
model:
_target_: model.diffusion.diffusion.DiffusionModel
_target_: model.diffusion.diffusion_eval_ft.DiffusionEvalFT
ft_denoising_steps: ${ft_denoising_steps}
predict_epsilon: True
denoised_clip_value: 1.0
randn_clip_value: 3