config.json
3.0 KB · 121 lines · json Raw
1 {
2 "action_in_proj_cfg": {
3 "_target_": "alpamayo1_5.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_": "alpamayo1_5.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 "architectures": [
38 "Alpamayo1_5"
39 ],
40 "attn_implementation": "flash_attention_2",
41 "diffusion_cfg": {
42 "_target_": "alpamayo1_5.diffusion.flow_matching.FlowMatching",
43 "int_method": "euler",
44 "x_dims": null
45 },
46 "dtype": "bfloat16",
47 "expert_cfg": {
48 "head_dim": 128,
49 "hidden_size": 2048,
50 "intermediate_size": 8256,
51 "num_attention_heads": 16
52 },
53 "expert_non_causal_attention": true,
54 "hist_traj_tokenizer_cfg": {
55 "_target_": "alpamayo1_5.models.delta_tokenizer.DeltaTrajectoryTokenizer"
56 },
57 "include_camera_ids": true,
58 "include_frame_nums": true,
59 "keep_same_dtype": true,
60 "max_pixels": 196608,
61 "min_pixels": 163840,
62 "model_dtype": "bfloat16",
63 "model_type": "alpamayo1_5",
64 "padding_side": "left",
65 "tokens_per_future_traj": 128,
66 "tokens_per_history_traj": 48,
67 "traj_token_ids": {
68 "future": 155685,
69 "future_end": 155683,
70 "future_start": 155681,
71 "history": 155684,
72 "history_end": 155676,
73 "history_start": 155674
74 },
75 "traj_token_start_idx": 151669,
76 "traj_tokenizer_cfg": {
77 "_recursive_": false,
78 "_target_": "alpamayo1_5.action_space.discrete_action_space.DiscreteTrajectoryTokenizer",
79 "action_space_cfg": {
80 "_target_": "alpamayo1_5.action_space.UnicycleAccelCurvatureActionSpace",
81 "a_lambda": 0.0001,
82 "a_ridge": 0.0001,
83 "accel_bounds": [
84 -9.8,
85 9.8
86 ],
87 "accel_mean": 0.02902694707164455,
88 "accel_std": 0.6810426736454882,
89 "curvature_bounds": [
90 -0.33,
91 0.33
92 ],
93 "curvature_mean": 0.0002692167976330542,
94 "curvature_std": 0.026148280660833106,
95 "dt": 0.1,
96 "kappa_lambda": 0.0001,
97 "kappa_ridge": 0.0001,
98 "n_waypoints": 64,
99 "theta_lambda": 1e-06,
100 "theta_ridge": 1e-08,
101 "v_lambda": 1e-06,
102 "v_ridge": 0.0001
103 },
104 "dims_max": [
105 10,
106 10
107 ],
108 "dims_min": [
109 -10,
110 -10
111 ],
112 "num_bins": 3000
113 },
114 "traj_vocab_size": 4000,
115 "transformers_version": "4.57.1",
116 "vlm_backend": "qwenvl3",
117 "vlm_name_or_path": "nvidia/Cosmos-Reason2-8B",
118 "vocab_size": 155697,
119 "add_special_tokens": true
120 }
121