Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
390 changes: 390 additions & 0 deletions weights/06_02_checkpoint/config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,390 @@
agent_index: null
controlled_exp:
train:
ent_coef:
values:
- 0.01
- 0.005
learning_rate:
values:
- 0.001
- 0.003
- 0.01
env:
action_type: discrete
collision_behavior: 1
compute_eval_metrics: false
control_mode: control_vehicles
dt: 0.3
dynamics_model: jerk
goal_on_lane: true
goal_radius: 2.0
goal_speed: 3.0
inactive_agent_threshold: 0.4
init_mode: create_all_valid
init_step: 0
map_dir: pufferlib/resources/drive/binaries/carla
max_agents_per_env: 150
max_waypoint_spacing: 60.0
min_agents_per_env: 1
min_waypoint_spacing: 20.0
num_agents: 2048
num_maps: 8
num_target_waypoints: 3
obs_dropout_boundary: 0.4
obs_dropout_lane: 0.5
obs_norm_goal_offset_m: 200.0
obs_norm_road_seg_length_m: 10.0
obs_norm_road_seg_width_m: 5.0
obs_norm_veh_length_m: 15.0
obs_norm_veh_width_m: 10.0
obs_norm_xy_offset_m: 200.0
obs_range_partner_m: 200.0
obs_range_road_behind_m: 40.0
obs_range_road_front_m: 200.0
obs_range_road_side_m: 50.0
obs_range_traffic_control_m: 100.0
obs_slots_boundary_n: 80
obs_slots_lane_n: 80
obs_slots_partners_n: 16
obs_slots_traffic_controls_n: 4
offroad_behavior: 1
partner_blindness_prob: 0.03
partner_blindness_trigger_prob: 0.05
phantom_braking_duration: 10
phantom_braking_prob: 0.02
phantom_braking_trigger_prob: 0.02
resample_frequency: 0
reward_ade: 0.0
reward_center_bias: 0.0
reward_collision: 1.5
reward_comfort: 0.05
reward_conditioning: true
reward_goal: 1.0
reward_lane_align: 0.025
reward_lane_center: 0.005
reward_offroad: 1.5
reward_overspeed: 0.05
reward_randomization: true
reward_reverse: 0.005
reward_stop_line: 1.0
reward_timestep: 2.5e-05
reward_vel_align: 1.0
reward_velocity: 0.0025
scenario_length: 1200
simulation_mode: gigaflow
spawn_initial_speed: 0.0
target_type: static
termination_mode: 1
traffic_light_behavior: 1
use_map_cache: 1
env_name: puffer_drive
eval:
behaviors_defaults:
clean: 'true'
enabled: 'false'
env:
control_mode: control_sdc_only
init_mode: create_all_valid
obs_slots_partners_n: 32
scenario_length: 201
simulation_mode: replay
eval:
num_scenarios: 50
render_max_steps: 200
render_num_scenarios: 2
interval: 250
mode: inline
render: 'true'
render_views:
- sim_state
- bev
behaviors_full_dir:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/nuplan_mini_train_bins
inherits: behaviors_defaults
type: behavior_class
behaviors_hard_stop:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/hard_stop
inherits: behaviors_defaults
type: behavior_class
behaviors_highway_straight:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/highway_straight
inherits: behaviors_defaults
type: behavior_class
behaviors_lane_change:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/lane_change
inherits: behaviors_defaults
type: behavior_class
behaviors_merge:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/merge
inherits: behaviors_defaults
type: behavior_class
behaviors_parked_cars:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/parked_cars
inherits: behaviors_defaults
type: behavior_class
behaviors_roundabout:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/roundabout
inherits: behaviors_defaults
type: behavior_class
behaviors_stopped_traffic:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/stopped_traffic
inherits: behaviors_defaults
type: behavior_class
behaviors_traffic_light_green:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/traffic_light_green
inherits: behaviors_defaults
type: behavior_class
behaviors_traffic_light_stop:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/traffic_light_stop
inherits: behaviors_defaults
type: behavior_class
behaviors_unprotected_left:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/unprotected_left
inherits: behaviors_defaults
type: behavior_class
behaviors_unprotected_right:
enabled: 0
env:
map_dir: /scratch/ev2237/data/nuplan/categories_v021/unprotected_right
inherits: behaviors_defaults
type: behavior_class
dnf_triage:
enabled: 'false'
env:
map_dir: pufferlib/resources/drive/binaries/carla/opendrive__Town10HD.bin
max_agents_per_env: 1
min_agents_per_env: 1
num_maps: 1
resample_frequency: 500
scenario_length: 500
simulation_mode: gigaflow
eval:
num_scenarios: 32
render_max_steps: 300
render_num_scenarios: 16
inherits: validation_defaults
render: 'true'
render_backend: triage_html
type: multi_scenario
validation_defaults:
clean: 'true'
enabled: 'true'
env:
collision_behavior: 1
eval_mode: 1
goal_speed: 3.0
num_agents: 1024
obs_dropout_boundary: 0.0
obs_dropout_lane: 0.0
obs_slots_boundary_n: 80
obs_slots_lane_n: 80
offroad_behavior: 1
reward_ade: 0.0
reward_collision: 3.0
reward_comfort: 0.05
reward_goal: 1.0
reward_lane_align: 0.025
reward_lane_center: 0.0038
reward_offroad: 3.0
reward_overspeed: 0.05
reward_randomization: false
reward_reverse: 0.005
reward_stop_line: 1.0
reward_timestep: 2.5e-05
reward_velocity: 0.0025
target_type: static
termination_mode: 0
traffic_light_behavior: 0
eval:
export_episode_csv: 'true'
num_scenarios: 250
verify_coverage: 'true'
interval: 250
mode: inline
validation_gigaflow:
enabled: 'true'
env:
map_dir: pufferlib/resources/drive/binaries/carla
max_agents_per_env: 40
min_agents_per_env: 40
num_agents: 1024
num_maps: 8
resample_frequency: 500
scenario_length: 500
simulation_mode: gigaflow
eval:
render_max_steps: 300
render_num_scenarios: 8
inherits: validation_defaults
render: false
render_backend: egl
render_views:
- sim_state
- bev
type: multi_scenario
validation_replay:
enabled: 0
env:
control_mode: control_sdc_only
map_dir: /scratch/ev2237/data/nuplan/nuplan_mini_train_bins
max_agents_per_env: 64
num_maps: 250
resample_frequency: 200
scenario_length: 200
simulation_mode: replay
eval:
render_max_steps: 200
render_num_scenarios: 5
inherits: validation_defaults
render: 'true'
render_backend: triage_html
type: multi_scenario
wosac:
clean: 'true'
enabled: 'false'
env:
control_mode: control_wosac
goal_radius: 2.0
init_mode: create_all_valid
init_step: 10
eval:
wosac_aggregate_results: 'true'
wosac_num_agents: 256
wosac_num_rollouts: 32
wosac_sanity_check: 'false'
interval: 500
mode: subprocess
render: 'false'
type: wosac
eval_simulation: null
fps: 15
gif_path: eval.gif
git:
commit_hash: aca6960c5c9368422a945fd62b81985378bc0031
load_id: null
load_model_path: null
local_rank: 0
max_runs: 200
max_suggestion_cost: 3600
mine:
num_episodes: 100
output_dir: ''
render: 'true'
score_threshold: -inf
neptune: false
neptune_name: pufferai
neptune_project: ablations
no_model_upload: {}
num_scenarios: 3
package: ocean
policy:
actor_hidden_size: 1024
actor_num_layers: 0
backbone_hidden_size: 1024
backbone_num_layers: 3
critic_hidden_size: 1024
critic_num_layers: 0
dropout: 0.0
encoder_gigaflow: true
input_size: 256
split_network: true
policy_name: Drive
render: 0
render_mode: auto
rnn:
hidden_size: 512
input_size: 512
rnn_name: null
run_name: null
save_frames: 0
sweep:
downsample: 10
goal: maximize
method: Protein
metric: score
tag: 2026-06-02_local_4gpu
tb: false
train:
adam_beta1: 0.9
adam_beta2: 0.999
adam_eps: 1.0e-08
adv_filter_ewma_beta: 0.25
adv_filter_threshold_scale: 0.01
adv_sampling_prio_alpha: 0.8499999999999999
adv_sampling_prio_beta0: 0.8499999999999999
amp: true
anneal_lr: true
batch_size: auto
bptt_horizon: 128
checkpoint_interval: 500
clip_coef: 0.2
compile: true
compile_fullgraph: false
compile_mode: default
cpu_offload: false
data_dir: experiments
device: cuda
ent_coef: 0.01
gae_lambda: 0.95
gamma: 0.999
learning_rate: 0.0005
max_grad_norm: 0.5
max_minibatch_size: 153600
minibatch_size: 153600
name: pufferai
normalize_rewards: false
obs_only: true
optimizer: adamw
ppo_granularity: auto
precision: bfloat16
project: ablations
render: false
render_interval: 1000
render_map: none
resume_state_path: null
seed: 0
show_grid: false
show_human_logs: true
show_lasers: false
torch_deterministic: false
total_timesteps: 10000000000
update_epochs: 3
use_rnn: false
vf_clip_coef: null
vf_coef: 0.5
vtrace_c_clip: 1
vtrace_rho_clip: 1
vec:
backend: Multiprocessing
batch_size: auto
num_envs: 20
num_workers: auto
seed: 42
zero_copy: true
video_path: videos
wandb: true
wandb_group: Nightly_MultiAgent
wandb_project: nightly-multi-agent
Binary file not shown.
Binary file added weights/06_02_checkpoint/trainer_state.pt
Binary file not shown.
Loading