AbstractPhil commited on
Commit
30776f1
·
verified ·
1 Parent(s): 4f16a24

[CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d] epoch=1 step=98 is_best=True

Browse files
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/checkpoint_best.meta.json ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1,
3
+ "global_step": 98,
4
+ "avg_loss": 3.554797871551514,
5
+ "best_metric": 0.64,
6
+ "timestamp": "2025-08-29T16:48:01.544007",
7
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
8
+ "group": "checkpoint",
9
+ "metrics": {
10
+ "epoch": 1,
11
+ "train_loss": 3.554797871551514,
12
+ "val_loss": 1.9044383962631226,
13
+ "top1": 0.64,
14
+ "top3": 0.8092999996185303,
15
+ "rose_weight": 0.15,
16
+ "rose_margin": 0.25,
17
+ "rose_temperature": 1.0
18
+ },
19
+ "config_snapshot": {
20
+ "device": "cuda",
21
+ "seed": 42,
22
+ "precision": "fp32",
23
+ "hf_clip_id": "laion/CLIP-ViT-B-32-laion2B-s34B-b79K",
24
+ "freeze_encoder": true,
25
+ "normalize_output": true,
26
+ "dataset": "cifar100",
27
+ "data_root": "./data",
28
+ "image_size": 224,
29
+ "clip_mean": [
30
+ 0.48145466,
31
+ 0.4578275,
32
+ 0.40821073
33
+ ],
34
+ "clip_std": [
35
+ 0.26862954,
36
+ 0.26130258,
37
+ 0.27577711
38
+ ],
39
+ "train_aug": true,
40
+ "train_crop_scale": [
41
+ 0.9,
42
+ 1.0
43
+ ],
44
+ "hflip_p": 0.5,
45
+ "vocab_repo": "AbstractPhil/geometric-vocab-256d",
46
+ "vocab_split": "unicode",
47
+ "vocab_dim_override": 256,
48
+ "vocab_singleton": true,
49
+ "crystal_dims": 256,
50
+ "role_index": {
51
+ "anchor": 0,
52
+ "support": 1,
53
+ "contrast": 2,
54
+ "purpose": 3,
55
+ "observer": 4
56
+ },
57
+ "rose_loss_weight": 0.15,
58
+ "rose_margin": 0.25,
59
+ "rose_temperature": 1.0,
60
+ "rose_role_weights": {
61
+ "anchor": 1.0,
62
+ "need": -0.5,
63
+ "relation": 0.75,
64
+ "purpose": 0.75,
65
+ "observer": -0.5
66
+ },
67
+ "classes": 100,
68
+ "classes_per_penta": 2,
69
+ "penta_branches": 50,
70
+ "pentachora_blocks": 4,
71
+ "temperature": 0.07,
72
+ "batch_size": 512,
73
+ "epochs": 20,
74
+ "lr": 0.0001,
75
+ "weight_decay": 0.0001,
76
+ "label_smoothing": 0.0,
77
+ "use_cosine_sched": true,
78
+ "warmup_steps": 0,
79
+ "max_steps": null,
80
+ "amp": true,
81
+ "grad_clip_norm": null,
82
+ "log_every": 25,
83
+ "auto_train": false,
84
+ "num_workers": 2,
85
+ "pin_memory": true,
86
+ "persistent_workers": true,
87
+ "ckpt_dir": "./checkpoints",
88
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
89
+ "save_latest": true,
90
+ "save_best": true,
91
+ "resume": false,
92
+ "resume_which": "best",
93
+ "resume_source": "auto",
94
+ "features_enable": true,
95
+ "features_autogenerate_if_missing": true,
96
+ "features_force_rebuild": false,
97
+ "features_local_root": "./features_cache",
98
+ "features_save_raw": true,
99
+ "features_check_remote_meta": true,
100
+ "hf_enable": true,
101
+ "hf_branch": "main",
102
+ "hf_private": true,
103
+ "hf_token_env": [
104
+ "HF_TOKEN",
105
+ "HUGGINGFACE_HUB_TOKEN"
106
+ ],
107
+ "save_every": 1,
108
+ "upload_every": 5,
109
+ "max_uploads_per_run": 1,
110
+ "push_strategy": "on_interval",
111
+ "hf_repo_id": "AbstractPhil/rose-geoclip",
112
+ "hf_push_latest": false,
113
+ "hf_push_best": true,
114
+ "hf_create": true,
115
+ "features_batch_size": 512,
116
+ "features_push_to_hub": true,
117
+ "features_offload_after_prepare": false,
118
+ "features_repo_id": "AbstractPhil/CLIP-ViT-B-32-laion2B-s34B-b79K_cifar100_extracted_features",
119
+ "features_local_dir": "./features_cache/CLIP-ViT-B-32-laion2B-s34B-b79K-cifar100",
120
+ "features_embed_dim": 512,
121
+ "features_ready": true,
122
+ "use_precomputed_features": true
123
+ }
124
+ }
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/checkpoint_best.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f97fd8b694cbc34bded0fdb09fcb933f90d78d32b418025e07884d544aa8852b
3
+ size 790736
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/encoder_snapshot.meta.json ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1,
3
+ "global_step": 98,
4
+ "avg_loss": 3.554797871551514,
5
+ "best_metric": 0.64,
6
+ "timestamp": "2025-08-29T16:48:01.281346",
7
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
8
+ "group": "encoder",
9
+ "metrics": {
10
+ "epoch": 1,
11
+ "train_loss": 3.554797871551514,
12
+ "val_loss": 1.9044383962631226,
13
+ "top1": 0.64,
14
+ "top3": 0.8092999996185303,
15
+ "rose_weight": 0.15,
16
+ "rose_margin": 0.25,
17
+ "rose_temperature": 1.0
18
+ },
19
+ "config_snapshot": {
20
+ "device": "cuda",
21
+ "seed": 42,
22
+ "precision": "fp32",
23
+ "hf_clip_id": "laion/CLIP-ViT-B-32-laion2B-s34B-b79K",
24
+ "freeze_encoder": true,
25
+ "normalize_output": true,
26
+ "dataset": "cifar100",
27
+ "data_root": "./data",
28
+ "image_size": 224,
29
+ "clip_mean": [
30
+ 0.48145466,
31
+ 0.4578275,
32
+ 0.40821073
33
+ ],
34
+ "clip_std": [
35
+ 0.26862954,
36
+ 0.26130258,
37
+ 0.27577711
38
+ ],
39
+ "train_aug": true,
40
+ "train_crop_scale": [
41
+ 0.9,
42
+ 1.0
43
+ ],
44
+ "hflip_p": 0.5,
45
+ "vocab_repo": "AbstractPhil/geometric-vocab-256d",
46
+ "vocab_split": "unicode",
47
+ "vocab_dim_override": 256,
48
+ "vocab_singleton": true,
49
+ "crystal_dims": 256,
50
+ "role_index": {
51
+ "anchor": 0,
52
+ "support": 1,
53
+ "contrast": 2,
54
+ "purpose": 3,
55
+ "observer": 4
56
+ },
57
+ "rose_loss_weight": 0.15,
58
+ "rose_margin": 0.25,
59
+ "rose_temperature": 1.0,
60
+ "rose_role_weights": {
61
+ "anchor": 1.0,
62
+ "need": -0.5,
63
+ "relation": 0.75,
64
+ "purpose": 0.75,
65
+ "observer": -0.5
66
+ },
67
+ "classes": 100,
68
+ "classes_per_penta": 2,
69
+ "penta_branches": 50,
70
+ "pentachora_blocks": 4,
71
+ "temperature": 0.07,
72
+ "batch_size": 512,
73
+ "epochs": 20,
74
+ "lr": 0.0001,
75
+ "weight_decay": 0.0001,
76
+ "label_smoothing": 0.0,
77
+ "use_cosine_sched": true,
78
+ "warmup_steps": 0,
79
+ "max_steps": null,
80
+ "amp": true,
81
+ "grad_clip_norm": null,
82
+ "log_every": 25,
83
+ "auto_train": false,
84
+ "num_workers": 2,
85
+ "pin_memory": true,
86
+ "persistent_workers": true,
87
+ "ckpt_dir": "./checkpoints",
88
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
89
+ "save_latest": true,
90
+ "save_best": true,
91
+ "resume": false,
92
+ "resume_which": "best",
93
+ "resume_source": "auto",
94
+ "features_enable": true,
95
+ "features_autogenerate_if_missing": true,
96
+ "features_force_rebuild": false,
97
+ "features_local_root": "./features_cache",
98
+ "features_save_raw": true,
99
+ "features_check_remote_meta": true,
100
+ "hf_enable": true,
101
+ "hf_branch": "main",
102
+ "hf_private": true,
103
+ "hf_token_env": [
104
+ "HF_TOKEN",
105
+ "HUGGINGFACE_HUB_TOKEN"
106
+ ],
107
+ "save_every": 1,
108
+ "upload_every": 5,
109
+ "max_uploads_per_run": 1,
110
+ "push_strategy": "on_interval",
111
+ "hf_repo_id": "AbstractPhil/rose-geoclip",
112
+ "hf_push_latest": false,
113
+ "hf_push_best": true,
114
+ "hf_create": true,
115
+ "features_batch_size": 512,
116
+ "features_push_to_hub": true,
117
+ "features_offload_after_prepare": false,
118
+ "features_repo_id": "AbstractPhil/CLIP-ViT-B-32-laion2B-s34B-b79K_cifar100_extracted_features",
119
+ "features_local_dir": "./features_cache/CLIP-ViT-B-32-laion2B-s34B-b79K-cifar100",
120
+ "features_embed_dim": 512,
121
+ "features_ready": true,
122
+ "use_precomputed_features": true
123
+ }
124
+ }
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/encoder_snapshot.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38043de30a132f425752fe884744fe2b16894b1045b61725704846dcd7c1112a
3
+ size 3024
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/head_best.meta.json ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1,
3
+ "global_step": 98,
4
+ "avg_loss": 3.554797871551514,
5
+ "best_metric": 0.64,
6
+ "timestamp": "2025-08-29T16:48:01.272968",
7
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
8
+ "group": "head",
9
+ "metrics": {
10
+ "epoch": 1,
11
+ "train_loss": 3.554797871551514,
12
+ "val_loss": 1.9044383962631226,
13
+ "top1": 0.64,
14
+ "top3": 0.8092999996185303,
15
+ "rose_weight": 0.15,
16
+ "rose_margin": 0.25,
17
+ "rose_temperature": 1.0
18
+ },
19
+ "config_snapshot": {
20
+ "device": "cuda",
21
+ "seed": 42,
22
+ "precision": "fp32",
23
+ "hf_clip_id": "laion/CLIP-ViT-B-32-laion2B-s34B-b79K",
24
+ "freeze_encoder": true,
25
+ "normalize_output": true,
26
+ "dataset": "cifar100",
27
+ "data_root": "./data",
28
+ "image_size": 224,
29
+ "clip_mean": [
30
+ 0.48145466,
31
+ 0.4578275,
32
+ 0.40821073
33
+ ],
34
+ "clip_std": [
35
+ 0.26862954,
36
+ 0.26130258,
37
+ 0.27577711
38
+ ],
39
+ "train_aug": true,
40
+ "train_crop_scale": [
41
+ 0.9,
42
+ 1.0
43
+ ],
44
+ "hflip_p": 0.5,
45
+ "vocab_repo": "AbstractPhil/geometric-vocab-256d",
46
+ "vocab_split": "unicode",
47
+ "vocab_dim_override": 256,
48
+ "vocab_singleton": true,
49
+ "crystal_dims": 256,
50
+ "role_index": {
51
+ "anchor": 0,
52
+ "support": 1,
53
+ "contrast": 2,
54
+ "purpose": 3,
55
+ "observer": 4
56
+ },
57
+ "rose_loss_weight": 0.15,
58
+ "rose_margin": 0.25,
59
+ "rose_temperature": 1.0,
60
+ "rose_role_weights": {
61
+ "anchor": 1.0,
62
+ "need": -0.5,
63
+ "relation": 0.75,
64
+ "purpose": 0.75,
65
+ "observer": -0.5
66
+ },
67
+ "classes": 100,
68
+ "classes_per_penta": 2,
69
+ "penta_branches": 50,
70
+ "pentachora_blocks": 4,
71
+ "temperature": 0.07,
72
+ "batch_size": 512,
73
+ "epochs": 20,
74
+ "lr": 0.0001,
75
+ "weight_decay": 0.0001,
76
+ "label_smoothing": 0.0,
77
+ "use_cosine_sched": true,
78
+ "warmup_steps": 0,
79
+ "max_steps": null,
80
+ "amp": true,
81
+ "grad_clip_norm": null,
82
+ "log_every": 25,
83
+ "auto_train": false,
84
+ "num_workers": 2,
85
+ "pin_memory": true,
86
+ "persistent_workers": true,
87
+ "ckpt_dir": "./checkpoints",
88
+ "run_name": "CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d",
89
+ "save_latest": true,
90
+ "save_best": true,
91
+ "resume": false,
92
+ "resume_which": "best",
93
+ "resume_source": "auto",
94
+ "features_enable": true,
95
+ "features_autogenerate_if_missing": true,
96
+ "features_force_rebuild": false,
97
+ "features_local_root": "./features_cache",
98
+ "features_save_raw": true,
99
+ "features_check_remote_meta": true,
100
+ "hf_enable": true,
101
+ "hf_branch": "main",
102
+ "hf_private": true,
103
+ "hf_token_env": [
104
+ "HF_TOKEN",
105
+ "HUGGINGFACE_HUB_TOKEN"
106
+ ],
107
+ "save_every": 1,
108
+ "upload_every": 5,
109
+ "max_uploads_per_run": 1,
110
+ "push_strategy": "on_interval",
111
+ "hf_repo_id": "AbstractPhil/rose-geoclip",
112
+ "hf_push_latest": false,
113
+ "hf_push_best": true,
114
+ "hf_create": true,
115
+ "features_batch_size": 512,
116
+ "features_push_to_hub": true,
117
+ "features_offload_after_prepare": false,
118
+ "features_repo_id": "AbstractPhil/CLIP-ViT-B-32-laion2B-s34B-b79K_cifar100_extracted_features",
119
+ "features_local_dir": "./features_cache/CLIP-ViT-B-32-laion2B-s34B-b79K-cifar100",
120
+ "features_embed_dim": 512,
121
+ "features_ready": true,
122
+ "use_precomputed_features": true
123
+ }
124
+ }
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/head_best.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e6e6039dd245fb468824f27427d98787f33a69b025440b68b98661ce4c26333e
3
+ size 790728
CLIP-ViT-B-32-laion2B-s34B-b79K-unicode-v256d-c256d/training_latest.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1,
3
+ "step": 98,
4
+ "avg_loss": 3.554797871551514,
5
+ "best_metric": 0.64,
6
+ "ts": "2025-08-29 16:48:01"
7
+ }