add small-rl-gen9beta

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4c0e2750fb1d118c645a49f1c3dd3b6266f1d6e04e22fcbd56895a45d96b17e
+size 56576177

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_10.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd9c1296b431039a8ccb5ad3ddd13bde8e55cccd3c84a93623f6920de53989fa
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_12.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c466284dd15acc688efae73e7e2f66ca9fc77070daaa28b58786091519ce82f7
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_14.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb33f5610cf213451343f04b73527591c7e2774e85402c88069168958e0859af
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_16.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9d97e58a357d2c3401f13ec5a05a22eb803ed26c716813cc866c7928bb1f204
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_18.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4e9b3a42f208c3e4538db8871b1a0d2c05f05266e2aaceda919f33181d4926a
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e8891b6c6a134a1bf6ea7d64c5e31a0c6ccc4e6fc34c1f6ecc378df833ddb6c
+size 56576177

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_20.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:beaa0fa4a4c400defc0801d3b31e383e045a497d792f4f6a9f6b18650385f8e6
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_22.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d93a4598039280403913ec35c41dd87a489f9f381c26854481a7a3afc973f991
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_24.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:921d0c1bfcbc9a4319bccad8003ee4fbb22bee93c9bc588e0118e6deb3c6b144
+size 56576351

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02904f0048c05b7008460aac2bfa07ed43e078669be6b1f3641aa4108eb13dec
+size 56576177

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4fad78a293aff52a9ec310e24a464c37b0969c783efa464c60e59df40a152f3
+size 56576177

small-rl-gen9beta/ckpts/policy_weights/policy_epoch_8.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9124f620013bdaa82afe502454c0e89503507f31c8720b77cab2bbfe20e6d1f5
+size 56576177

small-rl-gen9beta/config.txt

@@ -0,0 +1,130 @@
+import amago.agent
+import amago.experiment as amago2
+import amago.nets.actor_critic as amago3
+import amago.nets.traj_encoders as amago4
+
+# Parameters for Actor:
+# ==============================================================================
+# None.
+
+# 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 = (32, 0)
+
+# Parameters for MetamonAMAGOExperiment:
+# ==============================================================================
+MetamonAMAGOExperiment.agent_type = @agent.MultiTaskAgent
+MetamonAMAGOExperiment.critic_loss_weight = 10.0
+MetamonAMAGOExperiment.grad_clip = 1.5
+MetamonAMAGOExperiment.l2_coeff = 0.0001
+MetamonAMAGOExperiment.learning_rate = 0.00015
+MetamonAMAGOExperiment.lr_warmup_steps = 1000
+MetamonAMAGOExperiment.max_seq_len = 200
+MetamonAMAGOExperiment.traj_encoder_type = @traj_encoders.TformerTrajEncoder
+MetamonAMAGOExperiment.tstep_encoder_type = @MetamonTstepEncoder
+
+# Parameters for MetamonMaskedActor:
+# ==============================================================================
+MetamonMaskedActor.activation = 'leaky_relu'
+MetamonMaskedActor.continuous_dist_type = None
+MetamonMaskedActor.d_hidden = 300
+MetamonMaskedActor.dropout_p = 0.0
+MetamonMaskedActor.mask_illegal_actions = True
+MetamonMaskedActor.n_layers = 2
+
+# Parameters for MetamonTstepEncoder:
+# ==============================================================================
+MetamonTstepEncoder.d_model = 100
+MetamonTstepEncoder.dropout = 0.05
+MetamonTstepEncoder.extra_emb_dim = 18
+MetamonTstepEncoder.n_heads = 5
+MetamonTstepEncoder.n_layers = 3
+MetamonTstepEncoder.numerical_tokens = 6
+MetamonTstepEncoder.scratch_tokens = 4
+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 = @MetamonMaskedActor
+MultiTaskAgent.critic_type = @actor_critic.NCriticsTwoHot
+MultiTaskAgent.fake_filter = False
+MultiTaskAgent.fbc_filter_func = @agent.exp_filter
+MultiTaskAgent.gamma = 0.999
+MultiTaskAgent.num_actions_for_value_in_actor_loss = 3
+MultiTaskAgent.num_actions_for_value_in_critic_loss = 5
+MultiTaskAgent.num_critics = 4
+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 = 300
+NCriticsTwoHot.dropout_p = 0.0
+NCriticsTwoHot.max_return = 1100
+NCriticsTwoHot.min_return = -1100
+NCriticsTwoHot.n_layers = 2
+NCriticsTwoHot.output_bins = 64
+NCriticsTwoHot.use_symlog = False
+
+# Parameters for PopArtLayer:
+# ==============================================================================
+PopArtLayer.beta = 0.0005
+PopArtLayer.init_nu = 100.0
+
+# Parameters for SigmaReparam:
+# ==============================================================================
+SigmaReparam.bias = True
+SigmaReparam.fast_init = False
+
+# Parameters for TformerTrajEncoder:
+# ==============================================================================
+TformerTrajEncoder.activation = 'leaky_relu'
+TformerTrajEncoder.d_ff = 2048
+TformerTrajEncoder.d_model = 512
+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 = 8
+TformerTrajEncoder.n_layers = 3
+TformerTrajEncoder.norm = 'layer'
+TformerTrajEncoder.normformer_norms = True
+TformerTrajEncoder.pos_emb = 'fixed'
+TformerTrajEncoder.sigma_reparam = True
+
+# Parameters for TransformerTurnEmbedding:
+# ==============================================================================
+# None.