config.json
| 1 | { |
| 2 | "action_in_proj_cfg": { |
| 3 | "_target_": "alpamayo_r1.models.action_in_proj.PerWaypointActionInProjV2", |
| 4 | "hidden_size": 512, |
| 5 | "max_freq": 100.0, |
| 6 | "num_enc_layers": 2, |
| 7 | "num_fourier_feats": 20 |
| 8 | }, |
| 9 | "action_out_proj_cfg": { |
| 10 | "_target_": "torch.nn.Linear" |
| 11 | }, |
| 12 | "action_space_cfg": { |
| 13 | "_target_": "alpamayo_r1.action_space.UnicycleAccelCurvatureActionSpace", |
| 14 | "a_lambda": 0.0001, |
| 15 | "a_ridge": 0.0001, |
| 16 | "accel_bounds": [ |
| 17 | -9.8, |
| 18 | 9.8 |
| 19 | ], |
| 20 | "accel_mean": 0.02902694707164455, |
| 21 | "accel_std": 0.6810426736454882, |
| 22 | "curvature_bounds": [ |
| 23 | -0.33, |
| 24 | 0.33 |
| 25 | ], |
| 26 | "curvature_mean": 0.0002692167976330542, |
| 27 | "curvature_std": 0.026148280660833106, |
| 28 | "dt": 0.1, |
| 29 | "kappa_lambda": 0.0001, |
| 30 | "kappa_ridge": 0.0001, |
| 31 | "n_waypoints": 64, |
| 32 | "theta_lambda": 1e-06, |
| 33 | "theta_ridge": 1e-08, |
| 34 | "v_lambda": 1e-06, |
| 35 | "v_ridge": 0.0001 |
| 36 | }, |
| 37 | "add_special_tokens": true, |
| 38 | "architectures": [ |
| 39 | "AlpamayoR1" |
| 40 | ], |
| 41 | "attn_implementation": "flash_attention_2", |
| 42 | "diffusion_cfg": { |
| 43 | "_target_": "alpamayo_r1.diffusion.flow_matching.FlowMatching", |
| 44 | "int_method": "euler", |
| 45 | "x_dims": "???" |
| 46 | }, |
| 47 | "dtype": "bfloat16", |
| 48 | "expert_cfg": { |
| 49 | "dtype": "bfloat16", |
| 50 | "head_dim": 128, |
| 51 | "hidden_size": 2048, |
| 52 | "intermediate_size": 8256, |
| 53 | "num_attention_heads": 16 |
| 54 | }, |
| 55 | "expert_non_causal_attention": true, |
| 56 | "hist_traj_tokenizer_cfg": { |
| 57 | "_target_": "alpamayo_r1.models.delta_tokenizer.DeltaTrajectoryTokenizer" |
| 58 | }, |
| 59 | "keep_same_dtype": true, |
| 60 | "max_pixels": 196608, |
| 61 | "min_pixels": 163840, |
| 62 | "model_dtype": "bfloat16", |
| 63 | "model_type": "alpamayo_r1", |
| 64 | "tokens_per_future_traj": 128, |
| 65 | "tokens_per_history_traj": 48, |
| 66 | "traj_token_ids": { |
| 67 | "future": 155685, |
| 68 | "future_end": 155683, |
| 69 | "future_start": 155681, |
| 70 | "history": 155684, |
| 71 | "history_end": 155676, |
| 72 | "history_start": 155674 |
| 73 | }, |
| 74 | "traj_token_start_idx": 151669, |
| 75 | "traj_tokenizer_cfg": { |
| 76 | "_recursive_": false, |
| 77 | "_target_": "alpamayo_r1.action_space.discrete_action_space.DiscreteTrajectoryTokenizer", |
| 78 | "action_space_cfg": { |
| 79 | "_target_": "alpamayo_r1.action_space.UnicycleAccelCurvatureActionSpace", |
| 80 | "a_lambda": 0.0001, |
| 81 | "a_ridge": 0.0001, |
| 82 | "accel_bounds": [ |
| 83 | -9.8, |
| 84 | 9.8 |
| 85 | ], |
| 86 | "accel_mean": 0.02902694707164455, |
| 87 | "accel_std": 0.6810426736454882, |
| 88 | "curvature_bounds": [ |
| 89 | -0.33, |
| 90 | 0.33 |
| 91 | ], |
| 92 | "curvature_mean": 0.0002692167976330542, |
| 93 | "curvature_std": 0.026148280660833106, |
| 94 | "dt": 0.1, |
| 95 | "kappa_lambda": 0.0001, |
| 96 | "kappa_ridge": 0.0001, |
| 97 | "n_waypoints": 64, |
| 98 | "theta_lambda": 1e-06, |
| 99 | "theta_ridge": 1e-08, |
| 100 | "v_lambda": 1e-06, |
| 101 | "v_ridge": 0.0001 |
| 102 | }, |
| 103 | "dims_max": [ |
| 104 | 10, |
| 105 | 10 |
| 106 | ], |
| 107 | "dims_min": [ |
| 108 | -10, |
| 109 | -10 |
| 110 | ], |
| 111 | "num_bins": 3000 |
| 112 | }, |
| 113 | "traj_vocab_size": 4000, |
| 114 | "transformers_version": "4.57.1", |
| 115 | "vlm_backend": "qwenvl3", |
| 116 | "vocab_size": 155697 |
| 117 | } |
| 118 | |