Hugging Face's logo

MLL-Lab
/
LAGEN-models

Robotics
PyTorch
Safetensors
vision-language-action
latency-sensitive-bench
multiple-checkpoints
Model card Files
xet
 Community
1
talha15032 commited on
Commit
d0b9610
·
verified ·
1 Parent(s): 837086a

Publish checkpoint-only zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z

Browse files
Files changed (7) hide show
  1. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/README.md +5 -0
  2. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/checkpoint_only.json +58 -0
  3. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/checkpoints/model.pt +3 -0
  4. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/config.yaml +40 -0
  5. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/dataset_statistics.json +127 -0
  6. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/provenance.json +58 -0
  7. zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/task_contract.json +131 -0
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/README.md ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ # qwenoft — mikasa-intercept-grab-fast
2
+
3
+ Checkpoint-only upload. The final inference checkpoint exists locally and is uploaded here; the source Slurm job hit the 4h walltime before the canonical eval_l0.json/publication block completed.
4
+
5
+ Use `checkpoints/model.pt` with the included config/statistics/task_contract. See `checkpoint_only.json` / `provenance.json` for details.
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/checkpoint_only.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backbone": {
3
+ "allow_patterns": [
4
+ "*.json",
5
+ "*.safetensors",
6
+ "*.txt",
7
+ "*.jinja",
8
+ "*.model"
9
+ ],
10
+ "repo_id": "Qwen/Qwen3-VL-4B-Instruct",
11
+ "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17"
12
+ },
13
+ "checkpoint_bytes": 9785101729,
14
+ "code_revisions": {
15
+ "benchmark": {
16
+ "dirty": true,
17
+ "revision": "2f18d4620afd1861059d8cecfc2b1410a4a4777d"
18
+ },
19
+ "gr00t": {
20
+ "dirty": false,
21
+ "revision": "aa0c592853102010080e4f61609db8e3dc0779d4"
22
+ },
23
+ "haic": {
24
+ "dirty": true,
25
+ "revision": "b96533e17dfb6d5a1bd3a67032a66acc76e2fd80"
26
+ },
27
+ "mikasa": {
28
+ "dirty": true,
29
+ "revision": "a5f72ac4faf1f83ea1ccbf7f33ca04173554d3bc"
30
+ },
31
+ "parkour": {
32
+ "dirty": false,
33
+ "revision": "26274b2ad83cd6f45d061973a4d102fd492010de"
34
+ },
35
+ "starvla": {
36
+ "dirty": false,
37
+ "revision": "73cd1e6175d8c6e5896e742b6a8c8ca3aef52dca"
38
+ }
39
+ },
40
+ "datasets": [
41
+ {
42
+ "path": "zero-latency/mikasa-intercept-grab-fast/teacher-rollouts-h1-state-v1",
43
+ "repo_id": "latency-sensitive-bench/benchmark-datasets",
44
+ "revision": "5421c9d272d7a939bb04045b047af84519c1fba8"
45
+ }
46
+ ],
47
+ "model": "qwenoft",
48
+ "reason": "Training produced inference/checkpoints/model.pt, but the original 4h Slurm job ended before train_task.py wrote eval_l0.json/provenance/README and called publish_run.",
49
+ "run_id": "mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z",
50
+ "status": "checkpoint_only_uploaded_after_walltime",
51
+ "task": "mikasa",
52
+ "teacher": {
53
+ "path": "zero-latency/mikasa-intercept-grab-fast/small-policy/ppo-v1/checkpoint.pt",
54
+ "repo_id": "latency-sensitive-bench/benchmark-models",
55
+ "revision": "46abd9ece5745d808d16aa8013766a26b4f6c371"
56
+ },
57
+ "uploaded_at_utc": "2026-09-07T16:29:56Z"
58
+ }
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/checkpoints/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:03bd214f4c34a1d32ac046d4baaf301abcfe92e2470858ca9ff18338056475c7
3
+ size 9785101729
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/config.yaml ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ framework:
2
+ name: QwenOFT
3
+ qwenvl:
4
+ base_vlm: Qwen/Qwen3-VL-4B-Instruct
5
+ attn_implementation: flash_attention_2
6
+ action_model:
7
+ action_model_type: MLP
8
+ action_dim: 7
9
+ state_dim: 7
10
+ action_hidden_dim: 2560
11
+ future_action_window_size: 0
12
+ past_action_window_size: 0
13
+ loss_type: l1
14
+ state_encoding: continuous_projector
15
+ task_objective: null
16
+ action_env_dim: 7
17
+ action_horizon: 1
18
+ version_id: '0.21'
19
+ datasets:
20
+ vla_data:
21
+ include_state: true
22
+ obs_image_size:
23
+ - 224
24
+ - 224
25
+ image_mode: single
26
+ stitch_grid:
27
+ - 1
28
+ - 2
29
+ num_obs_frames: 1
30
+ trainer:
31
+ pretrained_checkpoint: null
32
+ backbone:
33
+ repo_id: Qwen/Qwen3-VL-4B-Instruct
34
+ revision: ebb281ec70b05090aa6165b016eac8ec08e71b17
35
+ allow_patterns:
36
+ - '*.json'
37
+ - '*.safetensors'
38
+ - '*.txt'
39
+ - '*.jinja'
40
+ - '*.model'
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/dataset_statistics.json ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "action": {
4
+ "mean": [
5
+ -0.11131570488214493,
6
+ 0.040192198008298874,
7
+ -0.24009454250335693,
8
+ -0.23081380128860474,
9
+ -0.41221386194229126,
10
+ 0.18129923939704895,
11
+ 0.013331156224012375
12
+ ],
13
+ "std": [
14
+ 0.22855518758296967,
15
+ 0.5893623232841492,
16
+ 0.26777687668800354,
17
+ 0.27581456303596497,
18
+ 0.3006018102169037,
19
+ 0.3146352171897888,
20
+ 0.40770089626312256
21
+ ],
22
+ "max": [
23
+ 0.3204230070114136,
24
+ 0.9921634197235107,
25
+ 0.6247130036354065,
26
+ 0.8002768158912659,
27
+ 0.6306125521659851,
28
+ 1.0,
29
+ 0.774597704410553
30
+ ],
31
+ "min": [
32
+ -1.0,
33
+ -1.0,
34
+ -1.0,
35
+ -0.9346221089363098,
36
+ -1.0,
37
+ -0.5651661157608032,
38
+ -1.0
39
+ ],
40
+ "q01": [
41
+ -0.8506493151187897,
42
+ -1.0,
43
+ -0.9802482903003692,
44
+ -0.6838930904865265,
45
+ -1.0,
46
+ -0.4035292208194733,
47
+ -0.831850928068161
48
+ ],
49
+ "q99": [
50
+ 0.2628934979438781,
51
+ 0.8585556507110594,
52
+ 0.18524706959724418,
53
+ 0.5309111773967736,
54
+ 0.22950317859649652,
55
+ 0.8685460448265068,
56
+ 0.6467343747615811
57
+ ],
58
+ "mask": [
59
+ true,
60
+ true,
61
+ true,
62
+ true,
63
+ true,
64
+ true,
65
+ true
66
+ ]
67
+ },
68
+ "state": {
69
+ "mean": [
70
+ -0.18996892869472504,
71
+ -0.09119448065757751,
72
+ 0.08502458781003952,
73
+ -2.616269588470459,
74
+ 0.33758142590522766,
75
+ -0.17528389394283295,
76
+ 0.04933461919426918
77
+ ],
78
+ "std": [
79
+ 0.08925069123506546,
80
+ 0.1439206600189209,
81
+ 0.08522012829780579,
82
+ 1.4712677001953125,
83
+ 0.18586333096027374,
84
+ 0.10226143896579742,
85
+ 0.01495052408427
86
+ ],
87
+ "max": [
88
+ -0.004026889801025391,
89
+ 0.23487472534179688,
90
+ 0.3442644476890564,
91
+ 3.1415677070617676,
92
+ 0.9648659229278564,
93
+ 0.0931963101029396,
94
+ 0.07999999821186066
95
+ ],
96
+ "min": [
97
+ -0.41803836822509766,
98
+ -0.37668657302856445,
99
+ 0.015313848853111267,
100
+ -3.1415457725524902,
101
+ -0.07448819279670715,
102
+ -0.40120765566825867,
103
+ 0.02547522261738777
104
+ ],
105
+ "q01": [
106
+ -0.3719064712524414,
107
+ -0.3454733848571777,
108
+ 0.018063589930534363,
109
+ -3.1388832092285157,
110
+ -0.029389445297420023,
111
+ -0.3605163902044296,
112
+ 0.0291226364672184
113
+ ],
114
+ "q99": [
115
+ -0.03631005287170415,
116
+ 0.17424454689025873,
117
+ 0.3186076879501343,
118
+ 3.1387048721313477,
119
+ 0.7408432960510251,
120
+ 0.053878990560769925,
121
+ 0.07999999821186066
122
+ ]
123
+ },
124
+ "num_transitions": 8911,
125
+ "num_trajectories": 250
126
+ }
127
+ }
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/provenance.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backbone": {
3
+ "allow_patterns": [
4
+ "*.json",
5
+ "*.safetensors",
6
+ "*.txt",
7
+ "*.jinja",
8
+ "*.model"
9
+ ],
10
+ "repo_id": "Qwen/Qwen3-VL-4B-Instruct",
11
+ "revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17"
12
+ },
13
+ "checkpoint_bytes": 9785101729,
14
+ "code_revisions": {
15
+ "benchmark": {
16
+ "dirty": true,
17
+ "revision": "2f18d4620afd1861059d8cecfc2b1410a4a4777d"
18
+ },
19
+ "gr00t": {
20
+ "dirty": false,
21
+ "revision": "aa0c592853102010080e4f61609db8e3dc0779d4"
22
+ },
23
+ "haic": {
24
+ "dirty": true,
25
+ "revision": "b96533e17dfb6d5a1bd3a67032a66acc76e2fd80"
26
+ },
27
+ "mikasa": {
28
+ "dirty": true,
29
+ "revision": "a5f72ac4faf1f83ea1ccbf7f33ca04173554d3bc"
30
+ },
31
+ "parkour": {
32
+ "dirty": false,
33
+ "revision": "26274b2ad83cd6f45d061973a4d102fd492010de"
34
+ },
35
+ "starvla": {
36
+ "dirty": false,
37
+ "revision": "73cd1e6175d8c6e5896e742b6a8c8ca3aef52dca"
38
+ }
39
+ },
40
+ "datasets": [
41
+ {
42
+ "path": "zero-latency/mikasa-intercept-grab-fast/teacher-rollouts-h1-state-v1",
43
+ "repo_id": "latency-sensitive-bench/benchmark-datasets",
44
+ "revision": "5421c9d272d7a939bb04045b047af84519c1fba8"
45
+ }
46
+ ],
47
+ "model": "qwenoft",
48
+ "reason": "Training produced inference/checkpoints/model.pt, but the original 4h Slurm job ended before train_task.py wrote eval_l0.json/provenance/README and called publish_run.",
49
+ "run_id": "mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z",
50
+ "status": "checkpoint_only_uploaded_after_walltime",
51
+ "task": "mikasa",
52
+ "teacher": {
53
+ "path": "zero-latency/mikasa-intercept-grab-fast/small-policy/ppo-v1/checkpoint.pt",
54
+ "repo_id": "latency-sensitive-bench/benchmark-models",
55
+ "revision": "46abd9ece5745d808d16aa8013766a26b4f6c371"
56
+ },
57
+ "uploaded_at_utc": "2026-09-07T16:29:56Z"
58
+ }
zero-latency/mikasa-intercept-grab-fast/vla/starvla-qwenoft-h1/mikasa_intercept_grab_fast_qwenoft_formal_latest_r8gpu8_20260907T082154Z/task_contract.json ADDED
@@ -0,0 +1,131 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "latency_bench_starvla_h1_v1",
3
+ "task": "mikasa_intercept_grab_fast",
4
+ "robot_type": "latency_mikasa_h1",
5
+ "state_dim": 7,
6
+ "action_dim": 7,
7
+ "action_horizon": 1,
8
+ "fps": 20.0,
9
+ "state_key": "observation.state",
10
+ "action_key": "action",
11
+ "video_keys": [
12
+ "video.top",
13
+ "video.wrist"
14
+ ],
15
+ "auxiliary": {},
16
+ "normalization": {
17
+ "state": {
18
+ "mean": [
19
+ -0.18996892869472504,
20
+ -0.09119448065757751,
21
+ 0.08502458781003952,
22
+ -2.616269588470459,
23
+ 0.33758142590522766,
24
+ -0.17528389394283295,
25
+ 0.04933461919426918
26
+ ],
27
+ "std": [
28
+ 0.08925069123506546,
29
+ 0.1439206600189209,
30
+ 0.08522012829780579,
31
+ 1.4712677001953125,
32
+ 0.18586333096027374,
33
+ 0.10226143896579742,
34
+ 0.01495052408427
35
+ ],
36
+ "min": [
37
+ -0.41803836822509766,
38
+ -0.37668657302856445,
39
+ 0.015313848853111267,
40
+ -3.1415457725524902,
41
+ -0.07448819279670715,
42
+ -0.40120765566825867,
43
+ 0.02547522261738777
44
+ ],
45
+ "max": [
46
+ -0.004026889801025391,
47
+ 0.23487472534179688,
48
+ 0.3442644476890564,
49
+ 3.1415677070617676,
50
+ 0.9648659229278564,
51
+ 0.0931963101029396,
52
+ 0.07999999821186066
53
+ ],
54
+ "q01": [
55
+ -0.3719064712524414,
56
+ -0.3454733848571777,
57
+ 0.018063589930534363,
58
+ -3.1388832092285157,
59
+ -0.029389445297420023,
60
+ -0.3605163902044296,
61
+ 0.0291226364672184
62
+ ],
63
+ "q99": [
64
+ -0.03631005287170415,
65
+ 0.17424454689025873,
66
+ 0.3186076879501343,
67
+ 3.1387048721313477,
68
+ 0.7408432960510251,
69
+ 0.053878990560769925,
70
+ 0.07999999821186066
71
+ ]
72
+ },
73
+ "action": {
74
+ "mean": [
75
+ -0.11131570488214493,
76
+ 0.040192198008298874,
77
+ -0.24009454250335693,
78
+ -0.23081380128860474,
79
+ -0.41221386194229126,
80
+ 0.18129923939704895,
81
+ 0.013331156224012375
82
+ ],
83
+ "std": [
84
+ 0.22855518758296967,
85
+ 0.5893623232841492,
86
+ 0.26777687668800354,
87
+ 0.27581456303596497,
88
+ 0.3006018102169037,
89
+ 0.3146352171897888,
90
+ 0.40770089626312256
91
+ ],
92
+ "min": [
93
+ -1.0,
94
+ -1.0,
95
+ -1.0,
96
+ -0.9346221089363098,
97
+ -1.0,
98
+ -0.5651661157608032,
99
+ -1.0
100
+ ],
101
+ "max": [
102
+ 0.3204230070114136,
103
+ 0.9921634197235107,
104
+ 0.6247130036354065,
105
+ 0.8002768158912659,
106
+ 0.6306125521659851,
107
+ 1.0,
108
+ 0.774597704410553
109
+ ],
110
+ "q01": [
111
+ -0.8506493151187897,
112
+ -1.0,
113
+ -0.9802482903003692,
114
+ -0.6838930904865265,
115
+ -1.0,
116
+ -0.4035292208194733,
117
+ -0.831850928068161
118
+ ],
119
+ "q99": [
120
+ 0.2628934979438781,
121
+ 0.8585556507110594,
122
+ 0.18524706959724418,
123
+ 0.5309111773967736,
124
+ 0.22950317859649652,
125
+ 0.8685460448265068,
126
+ 0.6467343747615811
127
+ ]
128
+ }
129
+ },
130
+ "prompt": "Intercept the rolling ball and grasp it to stop it."
131
+ }