pokeagent

alakazam/alakazam/ckpts/config.txt

@@ -0,0 +1,134 @@
+import amago.agent
+import amago.experiment as amago2
+import amago.nets.actor_critic as amago3
+import amago.nets.traj_encoders as amago4
+import amago.nets.transformer as amago5
+
+# Parameters for Agent:
+# ==============================================================================
+# None.
+
+# Parameters for ClippedSlidingSinkAttention:
+# ==============================================================================
+ClippedSlidingSinkAttention.logit_clip = 50
+ClippedSlidingSinkAttention.sink_bias = 0.0
+ClippedSlidingSinkAttention.sink_size = 5
+ClippedSlidingSinkAttention.window_size = 96
+
+# Parameters for Discrete:
+# ==============================================================================
+Discrete.clip_prob_high = 0.99
+Discrete.clip_prob_low = 0.001
+
+# Parameters for Experiment:
+# ==============================================================================
+Experiment.batches_per_update = 1
+Experiment.force_reset_train_envs_every = None
+Experiment.has_dset_edit_rights = True
+Experiment.mixed_precision = 'no'
+Experiment.padded_sampling = 'none'
+Experiment.sample_actions = True
+Experiment.save_trajs_as = 'npz'
+Experiment.traj_save_len = 10000000000.0
+Experiment.wandb_group_name = None
+
+# Parameters for MetamonAMAGOExperiment:
+# ==============================================================================
+MetamonAMAGOExperiment.agent_type = @agent.MultiTaskAgent
+MetamonAMAGOExperiment.critic_loss_weight = 12.5
+MetamonAMAGOExperiment.grad_clip = 1.5
+MetamonAMAGOExperiment.l2_coeff = 0.0001
+MetamonAMAGOExperiment.learning_rate = 0.000125
+MetamonAMAGOExperiment.lr_warmup_steps = 1250
+MetamonAMAGOExperiment.max_seq_len = 200
+MetamonAMAGOExperiment.traj_encoder_type = @traj_encoders.TformerTrajEncoder
+MetamonAMAGOExperiment.tstep_encoder_type = @MetamonTstepEncoder
+
+# Parameters for MetamonMaskedResidualActor:
+# ==============================================================================
+MetamonMaskedResidualActor.activation = 'leaky_relu'
+MetamonMaskedResidualActor.continuous_dist_type = None
+MetamonMaskedResidualActor.dropout_p = 0.0
+MetamonMaskedResidualActor.feature_dim = 256
+MetamonMaskedResidualActor.mask_illegal_actions = True
+MetamonMaskedResidualActor.normalization = 'layer'
+MetamonMaskedResidualActor.residual_blocks = 2
+MetamonMaskedResidualActor.residual_ff_dim = 512
+
+# Parameters for MetamonTstepEncoder:
+# ==============================================================================
+MetamonTstepEncoder.d_model = 108
+MetamonTstepEncoder.dropout = 0.05
+MetamonTstepEncoder.extra_emb_dim = 18
+MetamonTstepEncoder.n_heads = 6
+MetamonTstepEncoder.n_layers = 4
+MetamonTstepEncoder.numerical_tokens = 6
+MetamonTstepEncoder.scratch_tokens = 6
+MetamonTstepEncoder.token_mask_aug = False
+
+# Parameters for Multigammas:
+# ==============================================================================
+Multigammas.continuous = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+Multigammas.discrete = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+
+# Parameters for MultiTaskAgent:
+# ==============================================================================
+MultiTaskAgent.actor_type = @MetamonMaskedResidualActor
+MultiTaskAgent.critic_type = @actor_critic.NCriticsTwoHot
+MultiTaskAgent.fake_filter = False
+MultiTaskAgent.fbc_filter_func = @agent.leaky_relu_filter
+MultiTaskAgent.gamma = 0.999
+MultiTaskAgent.num_actions_for_value_in_actor_loss = 4
+MultiTaskAgent.num_actions_for_value_in_critic_loss = 4
+MultiTaskAgent.num_critics = 6
+MultiTaskAgent.num_critics_td = 2
+MultiTaskAgent.offline_coeff = 1.0
+MultiTaskAgent.online_coeff = 0.0
+MultiTaskAgent.pass_obs_keys_to_actor = ['illegal_actions']
+MultiTaskAgent.popart = True
+MultiTaskAgent.reward_multiplier = 10.0
+MultiTaskAgent.tau = 0.004
+MultiTaskAgent.use_multigamma = True
+MultiTaskAgent.use_target_actor = True
+
+# Parameters for NCriticsTwoHot:
+# ==============================================================================
+NCriticsTwoHot.activation = 'leaky_relu'
+NCriticsTwoHot.d_hidden = 512
+NCriticsTwoHot.dropout_p = 0.0
+NCriticsTwoHot.max_return = 2100
+NCriticsTwoHot.min_return = -100
+NCriticsTwoHot.n_layers = 2
+NCriticsTwoHot.output_bins = 96
+NCriticsTwoHot.use_symlog = False
+
+# Parameters for PopArtLayer:
+# ==============================================================================
+PopArtLayer.beta = 0.0005
+PopArtLayer.init_nu = 100.0
+
+# Parameters for ResidualActor:
+# ==============================================================================
+# None.
+
+# Parameters for TformerTrajEncoder:
+# ==============================================================================
+TformerTrajEncoder.activation = 'leaky_relu'
+TformerTrajEncoder.attention_type = @transformer.ClippedSlidingSinkAttention
+TformerTrajEncoder.d_ff = 3072
+TformerTrajEncoder.d_model = 768
+TformerTrajEncoder.dropout_attn = 0.0
+TformerTrajEncoder.dropout_emb = 0.05
+TformerTrajEncoder.dropout_ff = 0.05
+TformerTrajEncoder.dropout_qkv = 0.0
+TformerTrajEncoder.head_scaling = True
+TformerTrajEncoder.n_heads = 12
+TformerTrajEncoder.n_layers = 6
+TformerTrajEncoder.norm = 'layer'
+TformerTrajEncoder.normformer_norms = True
+TformerTrajEncoder.pos_emb = 'fixed'
+TformerTrajEncoder.sigma_reparam = False
+
+# Parameters for TransformerTurnEmbedding:
+# ==============================================================================
+# None.

alakazam/ckpts/config.txt

@@ -0,0 +1,132 @@
+import amago.agent
+import amago.experiment as amago2
+import amago.nets.actor_critic as amago3
+import amago.nets.traj_encoders as amago4
+import amago.nets.transformer as amago5
+
+# Parameters for Agent:
+# ==============================================================================
+# None.
+
+# Parameters for ClippedSlidingSinkAttention:
+# ==============================================================================
+ClippedSlidingSinkAttention.logit_clip = 50
+ClippedSlidingSinkAttention.sink_bias = 0.0
+ClippedSlidingSinkAttention.sink_size = 5
+ClippedSlidingSinkAttention.window_size = 96
+
+# Parameters for Discrete:
+# ==============================================================================
+Discrete.clip_prob_high = 0.99
+Discrete.clip_prob_low = 0.001
+
+# Parameters for Experiment:
+# ==============================================================================
+Experiment.always_load_latest = False
+Experiment.always_save_latest = True
+Experiment.has_dset_edit_rights = True
+Experiment.save_trajs_as = 'npz'
+Experiment.stagger_traj_file_lengths = True
+Experiment.traj_save_len = 10000000000.0
+Experiment.wandb_group_name = None
+
+# Parameters for MetamonAMAGOExperiment:
+# ==============================================================================
+MetamonAMAGOExperiment.agent_type = @agent.MultiTaskAgent
+MetamonAMAGOExperiment.critic_loss_weight = 13.5
+MetamonAMAGOExperiment.grad_clip = 1.5
+MetamonAMAGOExperiment.l2_coeff = 0.0001
+MetamonAMAGOExperiment.learning_rate = 0.000125
+MetamonAMAGOExperiment.lr_warmup_steps = 1500
+MetamonAMAGOExperiment.max_seq_len = 200
+MetamonAMAGOExperiment.traj_encoder_type = @traj_encoders.TformerTrajEncoder
+MetamonAMAGOExperiment.tstep_encoder_type = @MetamonTstepEncoder
+
+# Parameters for MetamonMaskedResidualActor:
+# ==============================================================================
+MetamonMaskedResidualActor.activation = 'leaky_relu'
+MetamonMaskedResidualActor.continuous_dist_type = None
+MetamonMaskedResidualActor.dropout_p = 0.0
+MetamonMaskedResidualActor.feature_dim = 256
+MetamonMaskedResidualActor.mask_illegal_actions = True
+MetamonMaskedResidualActor.normalization = 'layer'
+MetamonMaskedResidualActor.residual_blocks = 2
+MetamonMaskedResidualActor.residual_ff_dim = 512
+
+# Parameters for MetamonTstepEncoder:
+# ==============================================================================
+MetamonTstepEncoder.d_model = 108
+MetamonTstepEncoder.dropout = 0.05
+MetamonTstepEncoder.extra_emb_dim = 18
+MetamonTstepEncoder.n_heads = 6
+MetamonTstepEncoder.n_layers = 4
+MetamonTstepEncoder.numerical_tokens = 6
+MetamonTstepEncoder.scratch_tokens = 6
+MetamonTstepEncoder.token_mask_aug = False
+
+# Parameters for Multigammas:
+# ==============================================================================
+Multigammas.continuous = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+Multigammas.discrete = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+
+# Parameters for MultiTaskAgent:
+# ==============================================================================
+MultiTaskAgent.actor_type = @MetamonMaskedResidualActor
+MultiTaskAgent.critic_type = @actor_critic.NCriticsTwoHot
+MultiTaskAgent.fake_filter = False
+MultiTaskAgent.fbc_filter_func = @agent.leaky_relu_filter
+MultiTaskAgent.gamma = 0.999
+MultiTaskAgent.num_actions_for_value_in_actor_loss = 4
+MultiTaskAgent.num_actions_for_value_in_critic_loss = 4
+MultiTaskAgent.num_critics = 6
+MultiTaskAgent.num_critics_td = 2
+MultiTaskAgent.offline_coeff = 1.0
+MultiTaskAgent.online_coeff = 0.1
+MultiTaskAgent.pass_obs_keys_to_actor = ['illegal_actions']
+MultiTaskAgent.popart = True
+MultiTaskAgent.reward_multiplier = 10.0
+MultiTaskAgent.tau = 0.008
+MultiTaskAgent.use_multigamma = True
+MultiTaskAgent.use_target_actor = True
+
+# Parameters for NCriticsTwoHot:
+# ==============================================================================
+NCriticsTwoHot.activation = 'leaky_relu'
+NCriticsTwoHot.d_hidden = 512
+NCriticsTwoHot.dropout_p = 0.0
+NCriticsTwoHot.max_return = 2100
+NCriticsTwoHot.min_return = -100
+NCriticsTwoHot.n_layers = 2
+NCriticsTwoHot.output_bins = 96
+NCriticsTwoHot.use_symlog = False
+
+# Parameters for PopArtLayer:
+# ==============================================================================
+PopArtLayer.beta = 0.0005
+PopArtLayer.init_nu = 100.0
+
+# Parameters for ResidualActor:
+# ==============================================================================
+# None.
+
+# Parameters for TformerTrajEncoder:
+# ==============================================================================
+TformerTrajEncoder.activation = 'leaky_relu'
+TformerTrajEncoder.attention_type = @transformer.ClippedSlidingSinkAttention
+TformerTrajEncoder.d_ff = 3072
+TformerTrajEncoder.d_model = 768
+TformerTrajEncoder.dropout_attn = 0.0
+TformerTrajEncoder.dropout_emb = 0.05
+TformerTrajEncoder.dropout_ff = 0.05
+TformerTrajEncoder.dropout_qkv = 0.0
+TformerTrajEncoder.head_scaling = True
+TformerTrajEncoder.n_heads = 12
+TformerTrajEncoder.n_layers = 6
+TformerTrajEncoder.norm = 'layer'
+TformerTrajEncoder.normformer_norms = True
+TformerTrajEncoder.pos_emb = 'fixed'
+TformerTrajEncoder.sigma_reparam = False
+
+# Parameters for TransformerTurnEmbedding:
+# ==============================================================================
+# None.

alakazam/ckpts/latest/policy.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b9d7b7dbb389dc8fb66e92392384feb2a11b302b6e1130bfe32790cd1c177c
+size 233767193

alakazam/ckpts/policy_weights/policy_epoch_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23066ec5e029b29b95d8c2a41c18e3793a69a765bfd630814b5eaaf5b290bf65
+size 233772873

alakazam/ckpts/policy_weights/policy_epoch_10.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb1dee5aa36e76bacdf1bb72c5428fa4c08865756573a27192f632df6b5634a3
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_12.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ed66cb98aa85736f05b94c939dc22e1885570d33326bf413bf73b938d26bca5
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_14.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de4aa18f93025d31bba32170207c1c09c712dd9b4ca0bd411bb23b3297900e01
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_16.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a07b872b14ff15a303a37a05beef6ffd72e56871056b465bfd012260b579b0f8
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_18.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cffb486f4256c6a82be7df6ac68e275f4a121d2516e37a6f5ac6bbdd0608d56
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92421ef958cb85f73f43172431d7d648103efde6bddb6cc128258c284f874bd7
+size 233772873

alakazam/ckpts/policy_weights/policy_epoch_20.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fe5a8dc3a349176445ae8a82ab605180de6bdfd431a6c692eb38d4a16a74b8b
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_22.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbc1c4c1091337fa7eeb24adff88405d0794bc319ad536335922a2327fa44a68
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_24.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d85d042459182e9510e5c1d806fc7bec25510d76af846d87b6c6612e8d55c4d
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_26.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:134397c7107ce1278bf2021e571aa01340bb59270c17b7f4b052e79aee903e88
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_28.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:864d54aeeb99b5a78c9c2b447a4d20427c1f89ae0bff57476fe7ec14e9cc331d
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_30.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0549977f771affff1b8509e44f24bf48990fea824ae4d287090442a06634a637
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_32.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04fd9a00c67e19b3aceff2248a6e5ba12aba59d56db9731ee40b6d03f5976a84
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_34.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35e3edcc5115e7e0911cf63171a0daa7f103ef5efb1235b19e09d8cc403921e1
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_36.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68e79bf875d6ed5c0303504509f35d451c95bd1155f38b7dbe541c96dc8b4227
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_38.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f7c199ee69f72250c88d4d5ff69efd2931bbfad19fd3304bf7669cccbe0060b
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3af3e365fdc6db33abd761b7c192d84f4dad32adf0ab7979229c799d64a0fbce
+size 233772873

alakazam/ckpts/policy_weights/policy_epoch_40.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d14a7aecd49e3a201d9354f17c7d0b3d5161b8d1546168b31d878a0818b7e82
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_42.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:468d24a4ae25a6d5a1aa58e9234403259c8d4e544727da5f1f7c3083f8c6af66
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_44.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba5a20cd3df727a11681e4a554ce8adc9e6cf5f81d38f0a59a3e661a987ec7f2
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_46.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc4ae28f50214281a494bebce96b85d3b58b20b07b7d742e9bb72f3f4f983c50
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_48.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d8f9eff4f41fc552c5e154fa0659bcf6b4038706f166d34d35370be614df3cd
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_50.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af3a6fe709abf8c057c43c8e09672c9614a29808ef7cdf3277ed98a2502cd933
+size 233773103

alakazam/ckpts/policy_weights/policy_epoch_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cc117703af7788d33cb011bbffd621ac26467298020e39262e22511dfba9751
+size 233772873

alakazam/ckpts/policy_weights/policy_epoch_8.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466bb352d16d9191fb32271e4504f43ba8d9aa7115c279006818984f7f1533e0
+size 233772873

alakazam2/ckpts/config.txt

@@ -0,0 +1,135 @@
+import amago.agent
+import amago.experiment as amago2
+import amago.nets.actor_critic as amago3
+import amago.nets.traj_encoders as amago4
+import amago.nets.transformer as amago5
+
+# Parameters for Agent:
+# ==============================================================================
+# None.
+
+# Parameters for Discrete:
+# ==============================================================================
+Discrete.clip_prob_high = 0.99
+Discrete.clip_prob_low = 0.001
+
+# Parameters for Experiment:
+# ==============================================================================
+Experiment.always_load_latest = False
+Experiment.always_save_latest = True
+Experiment.has_dset_edit_rights = True
+Experiment.save_trajs_as = 'npz'
+Experiment.stagger_traj_file_lengths = True
+Experiment.traj_save_len = 10000000000.0
+Experiment.wandb_group_name = None
+
+# Parameters for FlashAttention:
+# ==============================================================================
+FlashAttention.window_size = (96, 0)
+
+# Parameters for MetamonAMAGOExperiment:
+# ==============================================================================
+MetamonAMAGOExperiment.agent_type = @agent.MultiTaskAgent
+MetamonAMAGOExperiment.critic_loss_weight = 13.5
+MetamonAMAGOExperiment.grad_clip = 1.5
+MetamonAMAGOExperiment.l2_coeff = 0.0001
+MetamonAMAGOExperiment.learning_rate = 0.000125
+MetamonAMAGOExperiment.lr_warmup_steps = 1500
+MetamonAMAGOExperiment.max_seq_len = 200
+MetamonAMAGOExperiment.traj_encoder_type = @traj_encoders.TformerTrajEncoder
+MetamonAMAGOExperiment.tstep_encoder_type = @MetamonPerceiverTstepEncoder
+
+# Parameters for MetamonMaskedResidualActor:
+# ==============================================================================
+MetamonMaskedResidualActor.activation = 'leaky_relu'
+MetamonMaskedResidualActor.continuous_dist_type = None
+MetamonMaskedResidualActor.dropout_p = 0.0
+MetamonMaskedResidualActor.feature_dim = 400
+MetamonMaskedResidualActor.mask_illegal_actions = True
+MetamonMaskedResidualActor.normalization = 'layer'
+MetamonMaskedResidualActor.residual_blocks = 2
+MetamonMaskedResidualActor.residual_ff_dim = 512
+
+# Parameters for MetamonPerceiverTstepEncoder:
+# ==============================================================================
+MetamonPerceiverTstepEncoder.d_model = 108
+MetamonPerceiverTstepEncoder.dropout = 0.05
+MetamonPerceiverTstepEncoder.extra_emb_dim = 18
+MetamonPerceiverTstepEncoder.latent_tokens = 8
+MetamonPerceiverTstepEncoder.max_tokens_per_turn = 128
+MetamonPerceiverTstepEncoder.n_heads = 6
+MetamonPerceiverTstepEncoder.n_layers = 5
+MetamonPerceiverTstepEncoder.numerical_tokens = 6
+MetamonPerceiverTstepEncoder.token_mask_aug = False
+
+# Parameters for Multigammas:
+# ==============================================================================
+Multigammas.continuous = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+Multigammas.discrete = [0.1, 0.9, 0.95, 0.97, 0.99, 0.995]
+
+# Parameters for MultiTaskAgent:
+# ==============================================================================
+MultiTaskAgent.actor_type = @MetamonMaskedResidualActor
+MultiTaskAgent.critic_type = @actor_critic.NCriticsTwoHot
+MultiTaskAgent.fake_filter = False
+MultiTaskAgent.fbc_filter_func = @agent.leaky_relu_filter
+MultiTaskAgent.gamma = 0.999
+MultiTaskAgent.num_actions_for_value_in_actor_loss = 4
+MultiTaskAgent.num_actions_for_value_in_critic_loss = 4
+MultiTaskAgent.num_critics = 6
+MultiTaskAgent.num_critics_td = 2
+MultiTaskAgent.offline_coeff = 1.0
+MultiTaskAgent.online_coeff = 0.1
+MultiTaskAgent.pass_obs_keys_to_actor = ['illegal_actions']
+MultiTaskAgent.popart = True
+MultiTaskAgent.reward_multiplier = 10.0
+MultiTaskAgent.tau = 0.008
+MultiTaskAgent.use_multigamma = True
+MultiTaskAgent.use_target_actor = True
+
+# Parameters for NCriticsTwoHot:
+# ==============================================================================
+NCriticsTwoHot.activation = 'leaky_relu'
+NCriticsTwoHot.d_hidden = 512
+NCriticsTwoHot.dropout_p = 0.0
+NCriticsTwoHot.max_return = 2100
+NCriticsTwoHot.min_return = -100
+NCriticsTwoHot.n_layers = 2
+NCriticsTwoHot.output_bins = 96
+NCriticsTwoHot.use_symlog = False
+
+# Parameters for PerceiverTurnEmbedding:
+# ==============================================================================
+# None.
+
+# Parameters for PopArtLayer:
+# ==============================================================================
+PopArtLayer.beta = 0.0005
+PopArtLayer.init_nu = 100.0
+
+# Parameters for ResidualActor:
+# ==============================================================================
+# None.
+
+# Parameters for SigmaReparam:
+# ==============================================================================
+SigmaReparam.bias = True
+SigmaReparam.fast_init = False
+
+# Parameters for TformerTrajEncoder:
+# ==============================================================================
+TformerTrajEncoder.activation = 'leaky_relu'
+TformerTrajEncoder.attention_type = @transformer.FlashAttention
+TformerTrajEncoder.d_ff = 3072
+TformerTrajEncoder.d_model = 768
+TformerTrajEncoder.dropout_attn = 0.0
+TformerTrajEncoder.dropout_emb = 0.05
+TformerTrajEncoder.dropout_ff = 0.05
+TformerTrajEncoder.dropout_qkv = 0.0
+TformerTrajEncoder.head_scaling = True
+TformerTrajEncoder.n_heads = 12
+TformerTrajEncoder.n_layers = 6
+TformerTrajEncoder.norm = 'layer'
+TformerTrajEncoder.normformer_norms = True
+TformerTrajEncoder.pos_emb = 'fixed'
+TformerTrajEncoder.sigma_reparam = True

alakazam2/ckpts/latest/policy.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5749e207e2844d32798abe1e50b4a8a9e0ceede488c78fb3d5eb7518af8f5bbd
+size 242350603

alakazam2/ckpts/policy_weights/policy_epoch_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4af96830f111687d973c2fa35f169735cfdb9eb3e24bdfc63ff3ad37e4e2720c
+size 242356043

alakazam2/ckpts/policy_weights/policy_epoch_10.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d88f19f754cbce4fed328a67f2ad1c714ff1bcc598ff8891ee97e0758cc41d2
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_12.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b24fe2be2ceea834e9744d1926f37d66653aaa00e9435735e16a4fb3ce625d8
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_14.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b7e0f6bc0793d4354e5a4602e94d4a2af091349029b935545148036cef32419
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_16.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79780c94dfad447bf7ece49ff9bbe5d39648dd79a50783d6681474391bc3b3e8
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_18.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7c8f76c362ba6c8c71259f9b7e0a0e8985dae57ced4613e30fb4f19061d4c65
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:907a87152482eb3301d29a1b47f0aac6a6bc10512e78831629f4e6f86600ae50
+size 242356043

alakazam2/ckpts/policy_weights/policy_epoch_20.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b56aeb44738de6ffd9a8eb700033ce2bdb3d750255653aba3d35f1849721d09
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_22.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:684c23b11feb1d3dd461686aa751801d477d326cdc02e64f9f407ca90fbb168b
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_24.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:666557150f2ccc31c9ae7cae90282720eb6fad874ea9c7fcbbbd913313b8112d
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_26.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91227a839e1b49c18cec8390eb2b1393d04c00198b04f471472adf1588615167
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_28.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0114dd16ae4a25de2c95d64a4e18dd457908cfb9d81e8da60cd4c57b4488f813
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_30.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:471900ef9eebfee75ccd6f8f4fec1aaa2e9362731781c63fb8ea6c2ed721bda9
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_32.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:081cf299f97c95bd2413943a1098741d565f25f449e2b030d47d40828be408c4
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_34.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cb4198d9774888e130de045f20b49dc00067d1ff80be099ee7510f4ddf2d534
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_36.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d7acbd7c048a3112c6d6460aea15e69f3c38a66bb52fba18b469188ea767900
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_38.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03724e7ee53ee469aeea2469233c89eef616361067dc9646b0a6fe4dd3b2bfb7
+size 242356427

alakazam2/ckpts/policy_weights/policy_epoch_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd907066db0bc80a556ccf983c129d5b2cfbb530ed4ba35bdfb17e71f3da8313
+size 242356043

alakazam2/ckpts/policy_weights/policy_epoch_40.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:deaf36d8d2f07b0f2d03539f4e248432a2f372682f38bc8ff9d37a1fe983a0d3
+size 242356427