Skip to content
Open
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
11 changes: 11 additions & 0 deletions build.sh
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,15 @@ else
echo "Error: environment '$ENV' not found" && exit 1
fi

case "$ENV" in
osrs_*)
python3 ocean/osrs/scripts/osrs_asset_manifest.py generate-c-header \
ocean/osrs/asset_manifest.json \
--output ocean/osrs/osrs_assets_generated.h
bash ocean/osrs/scripts/setup-data.sh
;;
esac

OUTPUT_NAME=${OUTPUT_NAME:-$ENV}
SRC_FILE=${SRC_FILE:-$SRC_DIR/$ENV.c}

Expand Down Expand Up @@ -267,6 +276,7 @@ if [ "$MODE" = "native" ]; then
-I$CUDA_HOME/include $NCCL_IFLAG -I$RAYLIB_NAME/include \
"${ENV_COMPILE_FLAGS[@]}" \
-DENV_NAME=$ENV \
-DPUFFER_ENV_NAME="\"$ENV\"" \
-DPUFFERLIB_BUILD_MAIN \
-Xcompiler=-DPLATFORM_DESKTOP \
-Xcompiler=-fopenmp \
Expand All @@ -287,6 +297,7 @@ elif [ "$MODE" = "profile" ]; then
-I$CUDA_HOME/include $NCCL_IFLAG -I$RAYLIB_NAME/include \
"${ENV_COMPILE_FLAGS[@]}" \
-DENV_NAME=$ENV \
-DPUFFER_ENV_NAME="\"$ENV\"" \
-Xcompiler=-DPLATFORM_DESKTOP \
$PRECISION \
-Xcompiler=-fopenmp \
Expand Down
142 changes: 142 additions & 0 deletions config/osrs_colosseum.ini
Original file line number Diff line number Diff line change
@@ -0,0 +1,142 @@
# OSRS Fortis Colosseum encounter.

[base]
env_name = osrs_colosseum
async = 0

[env]
start_wave = 1
damage_reward_coeff = 0.01
wave_clear_bonus = 0.9480448239932152
win_bonus = 0.6761689252871952
death_penalty_coeff = 1.0
timeout_penalty = 0.42966922179495604
farm_safe_damage_cap = 1
farm_cap_waves = 4
boss_damage_reward_coeff = 0.0
boss_phase_bonus = 0.930438578582552
argmax_gear_reward_coeff = 0.003360076095091039
offensive_boost_reward_coeff = 0.01150336488068926
loadout_profile_mode = 2
beginner_loadout_fraction = 0.5
step_out_forecast_obs_enabled = 1
threat_field_obs_enabled = 1
forecast_horizon = 4
mask_inventory_heads = 0
prayer_oracle_mode = 0
late_start_state_mode = 2
late_start_supply_fraction_per_wave = 0.07
bis_gear_oracle_mode = 0
invuln_mode = 0
episode_max_ticks_override = 0
remove_brews = 0
prayer_switch_fail_prob = 0.0
player_damage_received_scale = 1.0
damage_scale_anneal_start = 1.0
damage_scale_anneal_ticks = 0
classic_curriculum_mode = 1
curriculum_num_tiers = 3
curriculum_wave_1 = 4
curriculum_frac_1 = 0.07367961709476761
curriculum_wave_2 = 8
curriculum_frac_2 = 0.03234710556986353
curriculum_wave_3 = 12
curriculum_frac_3 = 0.009776253765422502

[vec]
total_agents = 8192
num_buffers = 8
num_threads = 32
action_mask_size = 452
gpu_env = 0

[policy]
hidden_size = 2048
num_layers = 2
expansion_factor = 1

[train]
total_timesteps = 143245903
horizon = 16
min_lr_ratio = 0.32764059472071716
learning_rate = 0.0003724286337024778
momentum = 0.999
ent_coef = 1e-07
gamma = 0.9999997409988988
gae_lambda = 0.41662511987289663
vtrace_rho_clip = 2.0739707238638987
vtrace_c_clip = 3.444175403341336
prio_alpha = 0.2617270145386202
prio_beta0 = 0.0
clip_coef = 0.07092591964303953
vf_coef = 5.0
vf_clip_coef = 0.0001
max_grad_norm = 2.5111866624698655
replay_ratio = 2.0
minibatch_size = 4096

[sweep]
max_suggestion_cost = 1200
max_runs = 400
gpus = 1
early_stop_quantile = 0.25

[sweep.train.total_timesteps]
distribution = log_normal
min = 10_000_000
max = 2_000_000_000
scale = time

[sweep.env.damage_reward_coeff]
distribution = uniform
min = 0.0
max = 0.010
scale = auto

[sweep.env.win_bonus]
distribution = uniform
min = 0.0
max = 1.0
scale = auto

[sweep.env.boss_phase_bonus]
distribution = uniform
min = 0.0
max = 1.0
scale = auto

[sweep.env.death_penalty_coeff]
distribution = uniform
min = 0.0
max = 1.0
scale = auto

[sweep.env.timeout_penalty]
distribution = uniform
min = 0.0
max = 1.0
scale = auto

[sweep.env.argmax_gear_reward_coeff]
distribution = uniform
min = 0.0
max = 0.05
scale = auto

[sweep.env.offensive_boost_reward_coeff]
distribution = uniform
min = 0.0
max = 0.05
scale = auto

[sweep.train.vf_coef]
distribution = uniform
min = 0.1
max = 8.0
scale = auto

[sweep.train.vf_clip_coef]
distribution = uniform
min = 0.0001
max = 1.0
scale = auto
Loading
Loading