Promote v9 preview checkpoint
Browse files- ESS-AIST-81M.safetensors +2 -2
- README.md +49 -30
- ess_ait_86m_spec.yaml +18 -12
- event_eval.json +135 -135
- manifest.json +2 -6
- parameter_breakdown.json +2 -2
- prefix_eval.json +29 -29
- retrieval_512_gt1030.json +27 -27
- subject_eval.json +72 -72
ESS-AIST-81M.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:415d6f5ac8299fd17265a6d1ae5ccafeed26729b212231778188dbadceaf6fba
|
| 3 |
+
size 323643096
|
README.md
CHANGED
|
@@ -9,7 +9,6 @@ tags:
|
|
| 9 |
- retrieval
|
| 10 |
- image-text-audio
|
| 11 |
- feature-extraction
|
| 12 |
-
- gguf
|
| 13 |
library_name: pytorch
|
| 14 |
pipeline_tag: feature-extraction
|
| 15 |
datasets:
|
|
@@ -20,12 +19,27 @@ datasets:
|
|
| 20 |
|
| 21 |
`ESS-AIST-81M Preview` is the current Cortext trial checkpoint from the ESS line.
|
| 22 |
|
| 23 |
-
- release checkpoint: `
|
|
|
|
| 24 |
- text encoder: `MongoDB/mdbr-leaf-ir`
|
| 25 |
- image encoder: `mobilenetv4_conv_medium.e180_r384_in12k`
|
| 26 |
- audio encoder: native `mn20_as` EfficientAT LoRA audio backbone
|
| 27 |
|
| 28 |
-
This
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
## Embedding Layout
|
| 31 |
|
|
@@ -44,7 +58,8 @@ Recommended normalized runtime views:
|
|
| 44 |
|
| 45 |
## Exact Release Metrics
|
| 46 |
|
| 47 |
-
All numbers below are from the exact published checkpoint
|
|
|
|
| 48 |
|
| 49 |
Evaluation scope note:
|
| 50 |
|
|
@@ -60,19 +75,19 @@ Source:
|
|
| 60 |
|
| 61 |
Speech holdout:
|
| 62 |
|
| 63 |
-
- `A->T_r1 = 0.
|
| 64 |
-
- `T->A_r1 = 0.
|
| 65 |
-
- `A->T_r5 = 0.
|
| 66 |
-
- `T->A_r5 = 0.
|
| 67 |
|
| 68 |
SALT:
|
| 69 |
|
| 70 |
-
- `I->T_r1 = 0.
|
| 71 |
-
- `T->I_r1 = 0.
|
| 72 |
-
- `A->T_r1 = 0.
|
| 73 |
-
- `T->A_r1 = 0.
|
| 74 |
-
- `I->A_r1 = 0.
|
| 75 |
-
- `A->I_r1 = 0.
|
| 76 |
|
| 77 |
### Held-Out ESS Eval
|
| 78 |
|
|
@@ -82,22 +97,26 @@ Sources:
|
|
| 82 |
- `event_eval.json`
|
| 83 |
- `prefix_eval.json`
|
| 84 |
|
| 85 |
-
Subject:
|
| 86 |
|
| 87 |
-
- `subject_key` same/different AUC: `0.
|
| 88 |
-
- `subject_key` same-topic-different-subject rejection AUC: `0.
|
| 89 |
|
| 90 |
-
Event:
|
| 91 |
|
| 92 |
-
- `
|
| 93 |
-
- `event_key` same
|
| 94 |
-
- `
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
Interpretation:
|
| 97 |
|
| 98 |
-
-
|
| 99 |
-
-
|
| 100 |
-
-
|
|
|
|
| 101 |
|
| 102 |
## Architecture
|
| 103 |
|
|
@@ -111,15 +130,14 @@ This preview is a frozen-encoder / trainable-projector stack:
|
|
| 111 |
- text projection params: `8,926,720`
|
| 112 |
- total exact loaded params: `80,812,854`
|
| 113 |
|
| 114 |
-
The audio path is not the old dual-audio teacher path. It uses the native
|
|
|
|
| 115 |
|
| 116 |
## Files
|
| 117 |
|
| 118 |
| File | Purpose |
|
| 119 |
|---|---|
|
| 120 |
| `ESS-AIST-81M.safetensors` | Full preview release artifact |
|
| 121 |
-
| `ESS-AIST-81M_q8_0.gguf` | Conservative GGUF quantization |
|
| 122 |
-
| `ESS-AIST-81M_q5_1.gguf` | Smaller GGUF quantization |
|
| 123 |
| `export_metadata.json` | ESS export contract |
|
| 124 |
| `manifest.json` | Release manifest |
|
| 125 |
| `parameter_breakdown.json` | Exact parameter accounting |
|
|
@@ -131,8 +149,9 @@ The audio path is not the old dual-audio teacher path. It uses the native audioh
|
|
| 131 |
|
| 132 |
## Caveats
|
| 133 |
|
| 134 |
-
- This is the current preview checkpoint, not the
|
| 135 |
-
-
|
| 136 |
-
-
|
|
|
|
| 137 |
- `SALT` and `speech holdout` are useful release gates for this line, but they are no longer fully external benchmarks in the same way they were for the earlier pre-ESS artifacts.
|
| 138 |
- Use this for internal Cortext trials, not as the final memory-model release.
|
|
|
|
| 9 |
- retrieval
|
| 10 |
- image-text-audio
|
| 11 |
- feature-extraction
|
|
|
|
| 12 |
library_name: pytorch
|
| 13 |
pipeline_tag: feature-extraction
|
| 14 |
datasets:
|
|
|
|
| 19 |
|
| 20 |
`ESS-AIST-81M Preview` is the current Cortext trial checkpoint from the ESS line.
|
| 21 |
|
| 22 |
+
- release checkpoint: `ess_aist_full_v9_subjectfix_l4k/best_model.pt`
|
| 23 |
+
- exported checkpoint epoch: `3`
|
| 24 |
- text encoder: `MongoDB/mdbr-leaf-ir`
|
| 25 |
- image encoder: `mobilenetv4_conv_medium.e180_r384_in12k`
|
| 26 |
- audio encoder: native `mn20_as` EfficientAT LoRA audio backbone
|
| 27 |
|
| 28 |
+
This preview is the current bridge artifact for Cortext. It keeps the ESS
|
| 29 |
+
`semantic / subject / event` slice layout, but the `v9` dataset repair moved
|
| 30 |
+
the `subject` slice much closer to the entity signal Cortext actually needs.
|
| 31 |
+
|
| 32 |
+
GGUF quantizations for this exact release live in:
|
| 33 |
+
|
| 34 |
+
- `augmem/ESS-AIST-81M-preview-GGUF`
|
| 35 |
+
|
| 36 |
+
Tradeoff:
|
| 37 |
+
|
| 38 |
+
- held-out subject/entity separation is much stronger than the earlier `v7` preview
|
| 39 |
+
- speech and SALT retrieval are weaker than the earlier `v7` retrieval-max point
|
| 40 |
+
|
| 41 |
+
For Cortext, this is still the better preview because the entity-side signal is
|
| 42 |
+
materially stronger.
|
| 43 |
|
| 44 |
## Embedding Layout
|
| 45 |
|
|
|
|
| 58 |
|
| 59 |
## Exact Release Metrics
|
| 60 |
|
| 61 |
+
All numbers below are from the exact published checkpoint state exported from
|
| 62 |
+
`ess_aist_full_v9_subjectfix_l4k/best_model.pt` at checkpoint epoch `3`.
|
| 63 |
|
| 64 |
Evaluation scope note:
|
| 65 |
|
|
|
|
| 75 |
|
| 76 |
Speech holdout:
|
| 77 |
|
| 78 |
+
- `A->T_r1 = 0.3276`
|
| 79 |
+
- `T->A_r1 = 0.3202`
|
| 80 |
+
- `A->T_r5 = 0.6120`
|
| 81 |
+
- `T->A_r5 = 0.6046`
|
| 82 |
|
| 83 |
SALT:
|
| 84 |
|
| 85 |
+
- `I->T_r1 = 0.3179`
|
| 86 |
+
- `T->I_r1 = 0.3425`
|
| 87 |
+
- `A->T_r1 = 0.1226`
|
| 88 |
+
- `T->A_r1 = 0.1272`
|
| 89 |
+
- `I->A_r1 = 0.1970`
|
| 90 |
+
- `A->I_r1 = 0.2148`
|
| 91 |
|
| 92 |
### Held-Out ESS Eval
|
| 93 |
|
|
|
|
| 97 |
- `event_eval.json`
|
| 98 |
- `prefix_eval.json`
|
| 99 |
|
| 100 |
+
Subject / entity surface:
|
| 101 |
|
| 102 |
+
- `subject_key` same/different AUC: `0.9881`
|
| 103 |
+
- `subject_key` same-topic-different-subject rejection AUC: `0.9881`
|
| 104 |
|
| 105 |
+
Event / disambiguation surface:
|
| 106 |
|
| 107 |
+
- `subject_key` event same/different AUC: `0.8855`
|
| 108 |
+
- `event_key` event same/different AUC: `0.8193`
|
| 109 |
+
- `subject_key` same-subject-different-event rejection AUC: `0.7381`
|
| 110 |
+
- `event_key` same-subject-different-event rejection AUC: `0.6807`
|
| 111 |
+
- `subject_key` topic-shift rejection AUC: `0.9513`
|
| 112 |
+
- `event_key` topic-shift rejection AUC: `0.8969`
|
| 113 |
|
| 114 |
Interpretation:
|
| 115 |
|
| 116 |
+
- the repaired `v9` held-out surface is no longer near-random on subject/entity
|
| 117 |
+
- the current `subject` slice is the strongest entity carrier in the model
|
| 118 |
+
- event structure is usable, but still entangled with subject
|
| 119 |
+
- this is the right bridge checkpoint for Cortext, not the final `semantic/entity` architecture
|
| 120 |
|
| 121 |
## Architecture
|
| 122 |
|
|
|
|
| 130 |
- text projection params: `8,926,720`
|
| 131 |
- total exact loaded params: `80,812,854`
|
| 132 |
|
| 133 |
+
The audio path is not the old dual-audio teacher path. It uses the native
|
| 134 |
+
audioheavy LoRA EfficientAT backbone.
|
| 135 |
|
| 136 |
## Files
|
| 137 |
|
| 138 |
| File | Purpose |
|
| 139 |
|---|---|
|
| 140 |
| `ESS-AIST-81M.safetensors` | Full preview release artifact |
|
|
|
|
|
|
|
| 141 |
| `export_metadata.json` | ESS export contract |
|
| 142 |
| `manifest.json` | Release manifest |
|
| 143 |
| `parameter_breakdown.json` | Exact parameter accounting |
|
|
|
|
| 149 |
|
| 150 |
## Caveats
|
| 151 |
|
| 152 |
+
- This is the current preview checkpoint, not the final Cortext model family.
|
| 153 |
+
- The current runtime slices are still named `semantic / subject / event`; the next family will move toward `semantic / entity`.
|
| 154 |
+
- Subject/entity is now strong on the repaired `v9` held-out surface, but event remains entangled and the engine still needs attention over active anchors for weak-reference resolution.
|
| 155 |
+
- Retrieval on `speech holdout` and `SALT` is lower than the earlier `v7` preview.
|
| 156 |
- `SALT` and `speech holdout` are useful release gates for this line, but they are no longer fully external benchmarks in the same way they were for the earlier pre-ESS artifacts.
|
| 157 |
- Use this for internal Cortext trials, not as the final memory-model release.
|
ess_ait_86m_spec.yaml
CHANGED
|
@@ -61,13 +61,13 @@ early_stopping_patience: 8
|
|
| 61 |
log_dir: runs
|
| 62 |
benchmark_eval_every_epochs: 1
|
| 63 |
|
| 64 |
-
# Next-run ESS corpus,
|
| 65 |
-
#
|
| 66 |
-
ess_corpus_dir: checkpoints/ess_ait_86m_20260430T035907Z/
|
| 67 |
-
ess_train_jsonl: checkpoints/ess_ait_86m_20260430T035907Z/
|
| 68 |
-
ess_val_jsonl: checkpoints/ess_ait_86m_20260430T035907Z/
|
| 69 |
-
ess_train_text_cache: checkpoints/ess_ait_86m_20260430T035907Z/
|
| 70 |
-
ess_val_text_cache: checkpoints/ess_ait_86m_20260430T035907Z/
|
| 71 |
|
| 72 |
# Multimodal subject-media attachment from the finalized v19 generated bundle.
|
| 73 |
ess_subject_media_dataset_dir: checkpoints/ess_ait_86m_20260430T035907Z/ess_subject_media_pilot52_full_v19
|
|
@@ -100,8 +100,8 @@ ess_subject_slice: [512, 1024]
|
|
| 100 |
ess_event_slice: [1024, 1536]
|
| 101 |
|
| 102 |
# Corpus composition at build time:
|
| 103 |
-
# - train: 219957 semantic /
|
| 104 |
-
# - val: 19954 semantic /
|
| 105 |
#
|
| 106 |
# Do not sample raw row frequency. Subject supervision is too small and must be
|
| 107 |
# explicitly oversampled to shape the subject block.
|
|
@@ -115,7 +115,9 @@ ess_sampling:
|
|
| 115 |
train_dataset_weights:
|
| 116 |
speech_chatterbox_150k: 5.0
|
| 117 |
wit_entity_subject: 0.25
|
| 118 |
-
librispeech_subject: 0.
|
|
|
|
|
|
|
| 119 |
val_family_weights:
|
| 120 |
semantic: 0.50
|
| 121 |
subject: 0.15
|
|
@@ -123,7 +125,9 @@ ess_sampling:
|
|
| 123 |
val_dataset_weights:
|
| 124 |
speech_chatterbox_150k: 5.0
|
| 125 |
wit_entity_subject: 0.25
|
| 126 |
-
librispeech_subject: 0.
|
|
|
|
|
|
|
| 127 |
family_from_active_supervision:
|
| 128 |
semantic: semantic
|
| 129 |
subject: subject
|
|
@@ -136,7 +140,9 @@ ess_sampling:
|
|
| 136 |
- subject rows are intentionally oversampled relative to raw corpus count
|
| 137 |
- semantic remains dominant to protect 512d retrieval
|
| 138 |
- speech_chatterbox semantic rows are oversampled within semantic because only ~20k rows survive dedupe into v6
|
| 139 |
-
- librispeech_subject
|
|
|
|
|
|
|
| 140 |
- event stays high enough to shape prefix_1536 without overwhelming semantic
|
| 141 |
|
| 142 |
ess_loss_weights:
|
|
|
|
| 61 |
log_dir: runs
|
| 62 |
benchmark_eval_every_epochs: 1
|
| 63 |
|
| 64 |
+
# Next-run ESS corpus, replacing the weak LibriSpeech subject text with
|
| 65 |
+
# identity-prefixed subject rows and book-level hard negatives.
|
| 66 |
+
ess_corpus_dir: checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix
|
| 67 |
+
ess_train_jsonl: checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/train.jsonl
|
| 68 |
+
ess_val_jsonl: checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/val.jsonl
|
| 69 |
+
ess_train_text_cache: checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/cache/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix_train_leaf_ir_text_features.npy
|
| 70 |
+
ess_val_text_cache: checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/cache/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix_val_leaf_ir_text_features.npy
|
| 71 |
|
| 72 |
# Multimodal subject-media attachment from the finalized v19 generated bundle.
|
| 73 |
ess_subject_media_dataset_dir: checkpoints/ess_ait_86m_20260430T035907Z/ess_subject_media_pilot52_full_v19
|
|
|
|
| 100 |
ess_event_slice: [1024, 1536]
|
| 101 |
|
| 102 |
# Corpus composition at build time:
|
| 103 |
+
# - train: 219957 semantic / 121223 event / 188415 subject
|
| 104 |
+
# - val: 19954 semantic / 13891 event / 21012 subject
|
| 105 |
#
|
| 106 |
# Do not sample raw row frequency. Subject supervision is too small and must be
|
| 107 |
# explicitly oversampled to shape the subject block.
|
|
|
|
| 115 |
train_dataset_weights:
|
| 116 |
speech_chatterbox_150k: 5.0
|
| 117 |
wit_entity_subject: 0.25
|
| 118 |
+
librispeech_subject: 0.02
|
| 119 |
+
librispeech_subject_lexical: 0.02
|
| 120 |
+
librispeech_event_lexical: 0.05
|
| 121 |
val_family_weights:
|
| 122 |
semantic: 0.50
|
| 123 |
subject: 0.15
|
|
|
|
| 125 |
val_dataset_weights:
|
| 126 |
speech_chatterbox_150k: 5.0
|
| 127 |
wit_entity_subject: 0.25
|
| 128 |
+
librispeech_subject: 0.02
|
| 129 |
+
librispeech_subject_lexical: 0.02
|
| 130 |
+
librispeech_event_lexical: 0.05
|
| 131 |
family_from_active_supervision:
|
| 132 |
semantic: semantic
|
| 133 |
subject: subject
|
|
|
|
| 140 |
- subject rows are intentionally oversampled relative to raw corpus count
|
| 141 |
- semantic remains dominant to protect 512d retrieval
|
| 142 |
- speech_chatterbox semantic rows are oversampled within semantic because only ~20k rows survive dedupe into v6
|
| 143 |
+
- librispeech_subject now carries identity-prefixed text plus book-level hard negatives, so its dataset weight is raised modestly
|
| 144 |
+
- librispeech_subject_lexical now mirrors that stronger identity text and is also raised modestly within subject
|
| 145 |
+
- librispeech_event_lexical adds chapter/book lexical event structure without dominating the SALT event anchor
|
| 146 |
- event stays high enough to shape prefix_1536 without overwhelming semantic
|
| 147 |
|
| 148 |
ess_loss_weights:
|
event_eval.json
CHANGED
|
@@ -1,265 +1,265 @@
|
|
| 1 |
{
|
| 2 |
-
"checkpoint": "/shared/augmem/triembed/checkpoints/
|
| 3 |
"split": "val",
|
| 4 |
-
"records_path": "/shared/augmem/triembed/checkpoints/ess_ait_86m_20260430T035907Z/
|
| 5 |
"views": {
|
| 6 |
"semantic_key": {
|
| 7 |
"event_same_different_auc": {
|
| 8 |
-
"auc": 0.
|
| 9 |
-
"positive_pairs":
|
| 10 |
-
"negative_pairs":
|
| 11 |
-
"positive_mean": 0.
|
| 12 |
-
"negative_mean": 0.
|
| 13 |
},
|
| 14 |
"same_subject_different_event_rejection_auc": {
|
| 15 |
-
"auc": 0.
|
| 16 |
-
"positive_pairs":
|
| 17 |
-
"negative_pairs":
|
| 18 |
-
"positive_mean": 0.
|
| 19 |
-
"negative_mean": 0.
|
| 20 |
},
|
| 21 |
"stale_same_source_rejection_auc": {
|
| 22 |
"auc": null,
|
| 23 |
-
"positive_pairs":
|
| 24 |
"negative_pairs": 0,
|
| 25 |
-
"positive_mean": 0.
|
| 26 |
"negative_mean": null
|
| 27 |
},
|
| 28 |
"wrong_active_rejection_auc": {
|
| 29 |
"auc": null,
|
| 30 |
-
"positive_pairs":
|
| 31 |
"negative_pairs": 0,
|
| 32 |
-
"positive_mean": 0.
|
| 33 |
"negative_mean": null
|
| 34 |
},
|
| 35 |
"topic_shift_rejection_auc": {
|
| 36 |
-
"auc": 0.
|
| 37 |
-
"positive_pairs":
|
| 38 |
-
"negative_pairs":
|
| 39 |
-
"positive_mean": 0.
|
| 40 |
-
"negative_mean": 0.
|
| 41 |
}
|
| 42 |
},
|
| 43 |
"subject_key": {
|
| 44 |
"event_same_different_auc": {
|
| 45 |
-
"auc": 0.
|
| 46 |
-
"positive_pairs":
|
| 47 |
-
"negative_pairs":
|
| 48 |
-
"positive_mean": 0.
|
| 49 |
-
"negative_mean": 0.
|
| 50 |
},
|
| 51 |
"same_subject_different_event_rejection_auc": {
|
| 52 |
-
"auc": 0.
|
| 53 |
-
"positive_pairs":
|
| 54 |
-
"negative_pairs":
|
| 55 |
-
"positive_mean": 0.
|
| 56 |
-
"negative_mean": 0.
|
| 57 |
},
|
| 58 |
"stale_same_source_rejection_auc": {
|
| 59 |
"auc": null,
|
| 60 |
-
"positive_pairs":
|
| 61 |
"negative_pairs": 0,
|
| 62 |
-
"positive_mean": 0.
|
| 63 |
"negative_mean": null
|
| 64 |
},
|
| 65 |
"wrong_active_rejection_auc": {
|
| 66 |
"auc": null,
|
| 67 |
-
"positive_pairs":
|
| 68 |
"negative_pairs": 0,
|
| 69 |
-
"positive_mean": 0.
|
| 70 |
"negative_mean": null
|
| 71 |
},
|
| 72 |
"topic_shift_rejection_auc": {
|
| 73 |
-
"auc": 0.
|
| 74 |
-
"positive_pairs":
|
| 75 |
-
"negative_pairs":
|
| 76 |
-
"positive_mean": 0.
|
| 77 |
-
"negative_mean": 0.
|
| 78 |
}
|
| 79 |
},
|
| 80 |
"event_key": {
|
| 81 |
"event_same_different_auc": {
|
| 82 |
-
"auc": 0.
|
| 83 |
-
"positive_pairs":
|
| 84 |
-
"negative_pairs":
|
| 85 |
-
"positive_mean": 0.
|
| 86 |
-
"negative_mean": 0.
|
| 87 |
},
|
| 88 |
"same_subject_different_event_rejection_auc": {
|
| 89 |
-
"auc": 0.
|
| 90 |
-
"positive_pairs":
|
| 91 |
-
"negative_pairs":
|
| 92 |
-
"positive_mean": 0.
|
| 93 |
-
"negative_mean": 0.
|
| 94 |
},
|
| 95 |
"stale_same_source_rejection_auc": {
|
| 96 |
"auc": null,
|
| 97 |
-
"positive_pairs":
|
| 98 |
"negative_pairs": 0,
|
| 99 |
-
"positive_mean": 0.
|
| 100 |
"negative_mean": null
|
| 101 |
},
|
| 102 |
"wrong_active_rejection_auc": {
|
| 103 |
"auc": null,
|
| 104 |
-
"positive_pairs":
|
| 105 |
"negative_pairs": 0,
|
| 106 |
-
"positive_mean": 0.
|
| 107 |
"negative_mean": null
|
| 108 |
},
|
| 109 |
"topic_shift_rejection_auc": {
|
| 110 |
-
"auc": 0.
|
| 111 |
-
"positive_pairs":
|
| 112 |
-
"negative_pairs":
|
| 113 |
-
"positive_mean": 0.
|
| 114 |
-
"negative_mean": 0.
|
| 115 |
}
|
| 116 |
},
|
| 117 |
"full_key": {
|
| 118 |
"event_same_different_auc": {
|
| 119 |
-
"auc": 0.
|
| 120 |
-
"positive_pairs":
|
| 121 |
-
"negative_pairs":
|
| 122 |
-
"positive_mean": 0.
|
| 123 |
-
"negative_mean": 0.
|
| 124 |
},
|
| 125 |
"same_subject_different_event_rejection_auc": {
|
| 126 |
-
"auc": 0.
|
| 127 |
-
"positive_pairs":
|
| 128 |
-
"negative_pairs":
|
| 129 |
-
"positive_mean": 0.
|
| 130 |
-
"negative_mean": 0.
|
| 131 |
},
|
| 132 |
"stale_same_source_rejection_auc": {
|
| 133 |
"auc": null,
|
| 134 |
-
"positive_pairs":
|
| 135 |
"negative_pairs": 0,
|
| 136 |
-
"positive_mean": 0.
|
| 137 |
"negative_mean": null
|
| 138 |
},
|
| 139 |
"wrong_active_rejection_auc": {
|
| 140 |
"auc": null,
|
| 141 |
-
"positive_pairs":
|
| 142 |
"negative_pairs": 0,
|
| 143 |
-
"positive_mean": 0.
|
| 144 |
"negative_mean": null
|
| 145 |
},
|
| 146 |
"topic_shift_rejection_auc": {
|
| 147 |
-
"auc": 0.
|
| 148 |
-
"positive_pairs":
|
| 149 |
-
"negative_pairs":
|
| 150 |
-
"positive_mean": 0.
|
| 151 |
-
"negative_mean": 0.
|
| 152 |
}
|
| 153 |
},
|
| 154 |
"prefix_512": {
|
| 155 |
"event_same_different_auc": {
|
| 156 |
-
"auc": 0.
|
| 157 |
-
"positive_pairs":
|
| 158 |
-
"negative_pairs":
|
| 159 |
-
"positive_mean": 0.
|
| 160 |
-
"negative_mean": 0.
|
| 161 |
},
|
| 162 |
"same_subject_different_event_rejection_auc": {
|
| 163 |
-
"auc": 0.
|
| 164 |
-
"positive_pairs":
|
| 165 |
-
"negative_pairs":
|
| 166 |
-
"positive_mean": 0.
|
| 167 |
-
"negative_mean": 0.
|
| 168 |
},
|
| 169 |
"stale_same_source_rejection_auc": {
|
| 170 |
"auc": null,
|
| 171 |
-
"positive_pairs":
|
| 172 |
"negative_pairs": 0,
|
| 173 |
-
"positive_mean": 0.
|
| 174 |
"negative_mean": null
|
| 175 |
},
|
| 176 |
"wrong_active_rejection_auc": {
|
| 177 |
"auc": null,
|
| 178 |
-
"positive_pairs":
|
| 179 |
"negative_pairs": 0,
|
| 180 |
-
"positive_mean": 0.
|
| 181 |
"negative_mean": null
|
| 182 |
},
|
| 183 |
"topic_shift_rejection_auc": {
|
| 184 |
-
"auc": 0.
|
| 185 |
-
"positive_pairs":
|
| 186 |
-
"negative_pairs":
|
| 187 |
-
"positive_mean": 0.
|
| 188 |
-
"negative_mean": 0.
|
| 189 |
}
|
| 190 |
},
|
| 191 |
"prefix_1024": {
|
| 192 |
"event_same_different_auc": {
|
| 193 |
-
"auc": 0.
|
| 194 |
-
"positive_pairs":
|
| 195 |
-
"negative_pairs":
|
| 196 |
-
"positive_mean": 0.
|
| 197 |
-
"negative_mean": 0.
|
| 198 |
},
|
| 199 |
"same_subject_different_event_rejection_auc": {
|
| 200 |
-
"auc": 0.
|
| 201 |
-
"positive_pairs":
|
| 202 |
-
"negative_pairs":
|
| 203 |
-
"positive_mean": 0.
|
| 204 |
-
"negative_mean": 0.
|
| 205 |
},
|
| 206 |
"stale_same_source_rejection_auc": {
|
| 207 |
"auc": null,
|
| 208 |
-
"positive_pairs":
|
| 209 |
"negative_pairs": 0,
|
| 210 |
-
"positive_mean": 0.
|
| 211 |
"negative_mean": null
|
| 212 |
},
|
| 213 |
"wrong_active_rejection_auc": {
|
| 214 |
"auc": null,
|
| 215 |
-
"positive_pairs":
|
| 216 |
"negative_pairs": 0,
|
| 217 |
-
"positive_mean": 0.
|
| 218 |
"negative_mean": null
|
| 219 |
},
|
| 220 |
"topic_shift_rejection_auc": {
|
| 221 |
-
"auc": 0.
|
| 222 |
-
"positive_pairs":
|
| 223 |
-
"negative_pairs":
|
| 224 |
-
"positive_mean": 0.
|
| 225 |
-
"negative_mean": 0.
|
| 226 |
}
|
| 227 |
},
|
| 228 |
"prefix_1536": {
|
| 229 |
"event_same_different_auc": {
|
| 230 |
-
"auc": 0.
|
| 231 |
-
"positive_pairs":
|
| 232 |
-
"negative_pairs":
|
| 233 |
-
"positive_mean": 0.
|
| 234 |
-
"negative_mean": 0.
|
| 235 |
},
|
| 236 |
"same_subject_different_event_rejection_auc": {
|
| 237 |
-
"auc": 0.
|
| 238 |
-
"positive_pairs":
|
| 239 |
-
"negative_pairs":
|
| 240 |
-
"positive_mean": 0.
|
| 241 |
-
"negative_mean": 0.
|
| 242 |
},
|
| 243 |
"stale_same_source_rejection_auc": {
|
| 244 |
"auc": null,
|
| 245 |
-
"positive_pairs":
|
| 246 |
"negative_pairs": 0,
|
| 247 |
-
"positive_mean": 0.
|
| 248 |
"negative_mean": null
|
| 249 |
},
|
| 250 |
"wrong_active_rejection_auc": {
|
| 251 |
"auc": null,
|
| 252 |
-
"positive_pairs":
|
| 253 |
"negative_pairs": 0,
|
| 254 |
-
"positive_mean": 0.
|
| 255 |
"negative_mean": null
|
| 256 |
},
|
| 257 |
"topic_shift_rejection_auc": {
|
| 258 |
-
"auc": 0.
|
| 259 |
-
"positive_pairs":
|
| 260 |
-
"negative_pairs":
|
| 261 |
-
"positive_mean": 0.
|
| 262 |
-
"negative_mean": 0.
|
| 263 |
}
|
| 264 |
}
|
| 265 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"checkpoint": "/shared/augmem/triembed/checkpoints/ess_aist_full_v9_subjectfix_l4k/best_model.pt",
|
| 3 |
"split": "val",
|
| 4 |
+
"records_path": "/shared/augmem/triembed/checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/val.jsonl",
|
| 5 |
"views": {
|
| 6 |
"semantic_key": {
|
| 7 |
"event_same_different_auc": {
|
| 8 |
+
"auc": 0.827451141075551,
|
| 9 |
+
"positive_pairs": 181625,
|
| 10 |
+
"negative_pairs": 453539,
|
| 11 |
+
"positive_mean": 0.8296324522244216,
|
| 12 |
+
"negative_mean": 0.6291148030260264
|
| 13 |
},
|
| 14 |
"same_subject_different_event_rejection_auc": {
|
| 15 |
+
"auc": 0.6695206182798321,
|
| 16 |
+
"positive_pairs": 181625,
|
| 17 |
+
"negative_pairs": 175759,
|
| 18 |
+
"positive_mean": 0.8296324522244216,
|
| 19 |
+
"negative_mean": 0.7356056283701664
|
| 20 |
},
|
| 21 |
"stale_same_source_rejection_auc": {
|
| 22 |
"auc": null,
|
| 23 |
+
"positive_pairs": 181625,
|
| 24 |
"negative_pairs": 0,
|
| 25 |
+
"positive_mean": 0.8296324522244216,
|
| 26 |
"negative_mean": null
|
| 27 |
},
|
| 28 |
"wrong_active_rejection_auc": {
|
| 29 |
"auc": null,
|
| 30 |
+
"positive_pairs": 181625,
|
| 31 |
"negative_pairs": 0,
|
| 32 |
+
"positive_mean": 0.8296324522244216,
|
| 33 |
"negative_mean": null
|
| 34 |
},
|
| 35 |
"topic_shift_rejection_auc": {
|
| 36 |
+
"auc": 0.8994650340280883,
|
| 37 |
+
"positive_pairs": 181625,
|
| 38 |
+
"negative_pairs": 340388,
|
| 39 |
+
"positive_mean": 0.8296324522244216,
|
| 40 |
+
"negative_mean": 0.5938547173777865
|
| 41 |
}
|
| 42 |
},
|
| 43 |
"subject_key": {
|
| 44 |
"event_same_different_auc": {
|
| 45 |
+
"auc": 0.8854762132187833,
|
| 46 |
+
"positive_pairs": 181625,
|
| 47 |
+
"negative_pairs": 453539,
|
| 48 |
+
"positive_mean": 0.8070377156889529,
|
| 49 |
+
"negative_mean": 0.5549017710295172
|
| 50 |
},
|
| 51 |
"same_subject_different_event_rejection_auc": {
|
| 52 |
+
"auc": 0.7381349173591332,
|
| 53 |
+
"positive_pairs": 181625,
|
| 54 |
+
"negative_pairs": 175759,
|
| 55 |
+
"positive_mean": 0.8070377156889529,
|
| 56 |
+
"negative_mean": 0.6661491519755037
|
| 57 |
},
|
| 58 |
"stale_same_source_rejection_auc": {
|
| 59 |
"auc": null,
|
| 60 |
+
"positive_pairs": 181625,
|
| 61 |
"negative_pairs": 0,
|
| 62 |
+
"positive_mean": 0.8070377156889529,
|
| 63 |
"negative_mean": null
|
| 64 |
},
|
| 65 |
"wrong_active_rejection_auc": {
|
| 66 |
"auc": null,
|
| 67 |
+
"positive_pairs": 181625,
|
| 68 |
"negative_pairs": 0,
|
| 69 |
+
"positive_mean": 0.8070377156889529,
|
| 70 |
"negative_mean": null
|
| 71 |
},
|
| 72 |
"topic_shift_rejection_auc": {
|
| 73 |
+
"auc": 0.9512869887738572,
|
| 74 |
+
"positive_pairs": 181625,
|
| 75 |
+
"negative_pairs": 340388,
|
| 76 |
+
"positive_mean": 0.8070377156889529,
|
| 77 |
+
"negative_mean": 0.5185315754949175
|
| 78 |
}
|
| 79 |
},
|
| 80 |
"event_key": {
|
| 81 |
"event_same_different_auc": {
|
| 82 |
+
"auc": 0.8193492434516296,
|
| 83 |
+
"positive_pairs": 181625,
|
| 84 |
+
"negative_pairs": 453539,
|
| 85 |
+
"positive_mean": 0.8111014214781179,
|
| 86 |
+
"negative_mean": 0.560698310072904
|
| 87 |
},
|
| 88 |
"same_subject_different_event_rejection_auc": {
|
| 89 |
+
"auc": 0.6806606788615208,
|
| 90 |
+
"positive_pairs": 181625,
|
| 91 |
+
"negative_pairs": 175759,
|
| 92 |
+
"positive_mean": 0.8111014214781179,
|
| 93 |
+
"negative_mean": 0.673238928121084
|
| 94 |
},
|
| 95 |
"stale_same_source_rejection_auc": {
|
| 96 |
"auc": null,
|
| 97 |
+
"positive_pairs": 181625,
|
| 98 |
"negative_pairs": 0,
|
| 99 |
+
"positive_mean": 0.8111014214781179,
|
| 100 |
"negative_mean": null
|
| 101 |
},
|
| 102 |
"wrong_active_rejection_auc": {
|
| 103 |
"auc": null,
|
| 104 |
+
"positive_pairs": 181625,
|
| 105 |
"negative_pairs": 0,
|
| 106 |
+
"positive_mean": 0.8111014214781179,
|
| 107 |
"negative_mean": null
|
| 108 |
},
|
| 109 |
"topic_shift_rejection_auc": {
|
| 110 |
+
"auc": 0.8968700248558907,
|
| 111 |
+
"positive_pairs": 181625,
|
| 112 |
+
"negative_pairs": 340388,
|
| 113 |
+
"positive_mean": 0.8111014214781179,
|
| 114 |
+
"negative_mean": 0.5184466918134688
|
| 115 |
}
|
| 116 |
},
|
| 117 |
"full_key": {
|
| 118 |
"event_same_different_auc": {
|
| 119 |
+
"auc": 0.8518429254835135,
|
| 120 |
+
"positive_pairs": 181625,
|
| 121 |
+
"negative_pairs": 453539,
|
| 122 |
+
"positive_mean": 0.8179552220574594,
|
| 123 |
+
"negative_mean": 0.5906047553254696
|
| 124 |
},
|
| 125 |
"same_subject_different_event_rejection_auc": {
|
| 126 |
+
"auc": 0.6938164439467958,
|
| 127 |
+
"positive_pairs": 181625,
|
| 128 |
+
"negative_pairs": 175759,
|
| 129 |
+
"positive_mean": 0.8179552220574594,
|
| 130 |
+
"negative_mean": 0.6977020526065576
|
| 131 |
},
|
| 132 |
"stale_same_source_rejection_auc": {
|
| 133 |
"auc": null,
|
| 134 |
+
"positive_pairs": 181625,
|
| 135 |
"negative_pairs": 0,
|
| 136 |
+
"positive_mean": 0.8179552220574594,
|
| 137 |
"negative_mean": null
|
| 138 |
},
|
| 139 |
"wrong_active_rejection_auc": {
|
| 140 |
"auc": null,
|
| 141 |
+
"positive_pairs": 181625,
|
| 142 |
"negative_pairs": 0,
|
| 143 |
+
"positive_mean": 0.8179552220574594,
|
| 144 |
"negative_mean": null
|
| 145 |
},
|
| 146 |
"topic_shift_rejection_auc": {
|
| 147 |
+
"auc": 0.9288029360785891,
|
| 148 |
+
"positive_pairs": 181625,
|
| 149 |
+
"negative_pairs": 340388,
|
| 150 |
+
"positive_mean": 0.8179552220574594,
|
| 151 |
+
"negative_mean": 0.553700150467076
|
| 152 |
}
|
| 153 |
},
|
| 154 |
"prefix_512": {
|
| 155 |
"event_same_different_auc": {
|
| 156 |
+
"auc": 0.827451141075551,
|
| 157 |
+
"positive_pairs": 181625,
|
| 158 |
+
"negative_pairs": 453539,
|
| 159 |
+
"positive_mean": 0.8296324522244216,
|
| 160 |
+
"negative_mean": 0.6291148030260264
|
| 161 |
},
|
| 162 |
"same_subject_different_event_rejection_auc": {
|
| 163 |
+
"auc": 0.6695206182798321,
|
| 164 |
+
"positive_pairs": 181625,
|
| 165 |
+
"negative_pairs": 175759,
|
| 166 |
+
"positive_mean": 0.8296324522244216,
|
| 167 |
+
"negative_mean": 0.7356056283701664
|
| 168 |
},
|
| 169 |
"stale_same_source_rejection_auc": {
|
| 170 |
"auc": null,
|
| 171 |
+
"positive_pairs": 181625,
|
| 172 |
"negative_pairs": 0,
|
| 173 |
+
"positive_mean": 0.8296324522244216,
|
| 174 |
"negative_mean": null
|
| 175 |
},
|
| 176 |
"wrong_active_rejection_auc": {
|
| 177 |
"auc": null,
|
| 178 |
+
"positive_pairs": 181625,
|
| 179 |
"negative_pairs": 0,
|
| 180 |
+
"positive_mean": 0.8296324522244216,
|
| 181 |
"negative_mean": null
|
| 182 |
},
|
| 183 |
"topic_shift_rejection_auc": {
|
| 184 |
+
"auc": 0.8994650340280883,
|
| 185 |
+
"positive_pairs": 181625,
|
| 186 |
+
"negative_pairs": 340388,
|
| 187 |
+
"positive_mean": 0.8296324522244216,
|
| 188 |
+
"negative_mean": 0.5938547173777865
|
| 189 |
}
|
| 190 |
},
|
| 191 |
"prefix_1024": {
|
| 192 |
"event_same_different_auc": {
|
| 193 |
+
"auc": 0.8613090604277244,
|
| 194 |
+
"positive_pairs": 181625,
|
| 195 |
+
"negative_pairs": 453539,
|
| 196 |
+
"positive_mean": 0.8195616576529665,
|
| 197 |
+
"negative_mean": 0.5971227166902979
|
| 198 |
},
|
| 199 |
"same_subject_different_event_rejection_auc": {
|
| 200 |
+
"auc": 0.7009197337402358,
|
| 201 |
+
"positive_pairs": 181625,
|
| 202 |
+
"negative_pairs": 175759,
|
| 203 |
+
"positive_mean": 0.8195616576529665,
|
| 204 |
+
"negative_mean": 0.7044079956623066
|
| 205 |
},
|
| 206 |
"stale_same_source_rejection_auc": {
|
| 207 |
"auc": null,
|
| 208 |
+
"positive_pairs": 181625,
|
| 209 |
"negative_pairs": 0,
|
| 210 |
+
"positive_mean": 0.8195616576529665,
|
| 211 |
"negative_mean": null
|
| 212 |
},
|
| 213 |
"wrong_active_rejection_auc": {
|
| 214 |
"auc": null,
|
| 215 |
+
"positive_pairs": 181625,
|
| 216 |
"negative_pairs": 0,
|
| 217 |
+
"positive_mean": 0.8195616576529665,
|
| 218 |
"negative_mean": null
|
| 219 |
},
|
| 220 |
"topic_shift_rejection_auc": {
|
| 221 |
+
"auc": 0.934646406192986,
|
| 222 |
+
"positive_pairs": 181625,
|
| 223 |
+
"negative_pairs": 340388,
|
| 224 |
+
"positive_mean": 0.8195616576529665,
|
| 225 |
+
"negative_mean": 0.5615511370717131
|
| 226 |
}
|
| 227 |
},
|
| 228 |
"prefix_1536": {
|
| 229 |
"event_same_different_auc": {
|
| 230 |
+
"auc": 0.8518429254835135,
|
| 231 |
+
"positive_pairs": 181625,
|
| 232 |
+
"negative_pairs": 453539,
|
| 233 |
+
"positive_mean": 0.8179552220574594,
|
| 234 |
+
"negative_mean": 0.5906047553254696
|
| 235 |
},
|
| 236 |
"same_subject_different_event_rejection_auc": {
|
| 237 |
+
"auc": 0.6938164439467958,
|
| 238 |
+
"positive_pairs": 181625,
|
| 239 |
+
"negative_pairs": 175759,
|
| 240 |
+
"positive_mean": 0.8179552220574594,
|
| 241 |
+
"negative_mean": 0.6977020526065576
|
| 242 |
},
|
| 243 |
"stale_same_source_rejection_auc": {
|
| 244 |
"auc": null,
|
| 245 |
+
"positive_pairs": 181625,
|
| 246 |
"negative_pairs": 0,
|
| 247 |
+
"positive_mean": 0.8179552220574594,
|
| 248 |
"negative_mean": null
|
| 249 |
},
|
| 250 |
"wrong_active_rejection_auc": {
|
| 251 |
"auc": null,
|
| 252 |
+
"positive_pairs": 181625,
|
| 253 |
"negative_pairs": 0,
|
| 254 |
+
"positive_mean": 0.8179552220574594,
|
| 255 |
"negative_mean": null
|
| 256 |
},
|
| 257 |
"topic_shift_rejection_auc": {
|
| 258 |
+
"auc": 0.9288029360785891,
|
| 259 |
+
"positive_pairs": 181625,
|
| 260 |
+
"negative_pairs": 340388,
|
| 261 |
+
"positive_mean": 0.8179552220574594,
|
| 262 |
+
"negative_mean": 0.553700150467076
|
| 263 |
}
|
| 264 |
}
|
| 265 |
}
|
manifest.json
CHANGED
|
@@ -1,10 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"model_id": "ESS-AIST-81M",
|
| 3 |
-
"trimodal_checkpoint": "/shared/augmem/triembed/checkpoints/
|
| 4 |
"audio_checkpoint": "/shared/augmem/triembed/checkpoints/mn20_native_lora_aistmix_audioheavy100k175k175k_continue_from_balanced_20260426T143137Z/latest_model.pt",
|
| 5 |
-
"safetensors": "/shared/augmem/triembed/dist/ESS-AIST-81M-preview/ESS-AIST-81M.safetensors"
|
| 6 |
-
"gguf": [
|
| 7 |
-
"/shared/augmem/triembed/dist/ESS-AIST-81M-preview/ESS-AIST-81M_q8_0.gguf",
|
| 8 |
-
"/shared/augmem/triembed/dist/ESS-AIST-81M-preview/ESS-AIST-81M_q5_1.gguf"
|
| 9 |
-
]
|
| 10 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"model_id": "ESS-AIST-81M",
|
| 3 |
+
"trimodal_checkpoint": "/shared/augmem/triembed/checkpoints/ess_aist_full_v9_subjectfix_l4k/best_model.pt",
|
| 4 |
"audio_checkpoint": "/shared/augmem/triembed/checkpoints/mn20_native_lora_aistmix_audioheavy100k175k175k_continue_from_balanced_20260426T143137Z/latest_model.pt",
|
| 5 |
+
"safetensors": "/shared/augmem/triembed/dist/ESS-AIST-81M-preview-hf/ESS-AIST-81M.safetensors"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
}
|
parameter_breakdown.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
"text_encoder": 22861056,
|
| 3 |
-
"image_encoder":
|
| 4 |
"audio_encoder": 20639974,
|
| 5 |
"image_projection": 9975296,
|
| 6 |
"audio_projection": 9975296,
|
| 7 |
"text_projection": 8926720,
|
| 8 |
-
"total_exact_loaded_params":
|
| 9 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"text_encoder": 22861056,
|
| 3 |
+
"image_encoder": 8502493,
|
| 4 |
"audio_encoder": 20639974,
|
| 5 |
"image_projection": 9975296,
|
| 6 |
"audio_projection": 9975296,
|
| 7 |
"text_projection": 8926720,
|
| 8 |
+
"total_exact_loaded_params": 80880835
|
| 9 |
}
|
prefix_eval.json
CHANGED
|
@@ -1,48 +1,48 @@
|
|
| 1 |
{
|
| 2 |
-
"checkpoint": "/shared/augmem/triembed/checkpoints/
|
| 3 |
"split": "val",
|
| 4 |
"views": {
|
| 5 |
"semantic_key": {
|
| 6 |
-
"subject_same_different_auc": 0.
|
| 7 |
-
"event_same_different_auc": 0.
|
| 8 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 9 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 10 |
},
|
| 11 |
"subject_key": {
|
| 12 |
-
"subject_same_different_auc": 0.
|
| 13 |
-
"event_same_different_auc": 0.
|
| 14 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 15 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 16 |
},
|
| 17 |
"event_key": {
|
| 18 |
-
"subject_same_different_auc": 0.
|
| 19 |
-
"event_same_different_auc": 0.
|
| 20 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 21 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 22 |
},
|
| 23 |
"full_key": {
|
| 24 |
-
"subject_same_different_auc": 0.
|
| 25 |
-
"event_same_different_auc": 0.
|
| 26 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 27 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 28 |
},
|
| 29 |
"prefix_512": {
|
| 30 |
-
"subject_same_different_auc": 0.
|
| 31 |
-
"event_same_different_auc": 0.
|
| 32 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 33 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 34 |
},
|
| 35 |
"prefix_1024": {
|
| 36 |
-
"subject_same_different_auc": 0.
|
| 37 |
-
"event_same_different_auc": 0.
|
| 38 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 39 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 40 |
},
|
| 41 |
"prefix_1536": {
|
| 42 |
-
"subject_same_different_auc": 0.
|
| 43 |
-
"event_same_different_auc": 0.
|
| 44 |
-
"same_topic_different_subject_rejection_auc": 0.
|
| 45 |
-
"same_subject_different_event_rejection_auc": 0.
|
| 46 |
}
|
| 47 |
}
|
| 48 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"checkpoint": "/shared/augmem/triembed/checkpoints/ess_aist_full_v9_subjectfix_l4k/best_model.pt",
|
| 3 |
"split": "val",
|
| 4 |
"views": {
|
| 5 |
"semantic_key": {
|
| 6 |
+
"subject_same_different_auc": 0.9562558404233674,
|
| 7 |
+
"event_same_different_auc": 0.827451141075551,
|
| 8 |
+
"same_topic_different_subject_rejection_auc": 0.9562558404233674,
|
| 9 |
+
"same_subject_different_event_rejection_auc": 0.6695206182798321
|
| 10 |
},
|
| 11 |
"subject_key": {
|
| 12 |
+
"subject_same_different_auc": 0.9881162919768391,
|
| 13 |
+
"event_same_different_auc": 0.8854762132187833,
|
| 14 |
+
"same_topic_different_subject_rejection_auc": 0.9881162919768391,
|
| 15 |
+
"same_subject_different_event_rejection_auc": 0.7381349173591332
|
| 16 |
},
|
| 17 |
"event_key": {
|
| 18 |
+
"subject_same_different_auc": 0.9551271013544805,
|
| 19 |
+
"event_same_different_auc": 0.8193492434516296,
|
| 20 |
+
"same_topic_different_subject_rejection_auc": 0.9551271013544805,
|
| 21 |
+
"same_subject_different_event_rejection_auc": 0.6806606788615208
|
| 22 |
},
|
| 23 |
"full_key": {
|
| 24 |
+
"subject_same_different_auc": 0.9778614751548688,
|
| 25 |
+
"event_same_different_auc": 0.8518429254835135,
|
| 26 |
+
"same_topic_different_subject_rejection_auc": 0.9778614751548688,
|
| 27 |
+
"same_subject_different_event_rejection_auc": 0.6938164439467958
|
| 28 |
},
|
| 29 |
"prefix_512": {
|
| 30 |
+
"subject_same_different_auc": 0.9562558404233674,
|
| 31 |
+
"event_same_different_auc": 0.827451141075551,
|
| 32 |
+
"same_topic_different_subject_rejection_auc": 0.9562558404233674,
|
| 33 |
+
"same_subject_different_event_rejection_auc": 0.6695206182798321
|
| 34 |
},
|
| 35 |
"prefix_1024": {
|
| 36 |
+
"subject_same_different_auc": 0.9814636892484202,
|
| 37 |
+
"event_same_different_auc": 0.8613090604277244,
|
| 38 |
+
"same_topic_different_subject_rejection_auc": 0.9814636892484202,
|
| 39 |
+
"same_subject_different_event_rejection_auc": 0.7009197337402358
|
| 40 |
},
|
| 41 |
"prefix_1536": {
|
| 42 |
+
"subject_same_different_auc": 0.9778614751548688,
|
| 43 |
+
"event_same_different_auc": 0.8518429254835135,
|
| 44 |
+
"same_topic_different_subject_rejection_auc": 0.9778614751548688,
|
| 45 |
+
"same_subject_different_event_rejection_auc": 0.6938164439467958
|
| 46 |
}
|
| 47 |
}
|
| 48 |
}
|
retrieval_512_gt1030.json
CHANGED
|
@@ -1,40 +1,40 @@
|
|
| 1 |
{
|
| 2 |
"SALT-512": {
|
| 3 |
-
"A->I_r1": 0.
|
| 4 |
-
"A->I_r10": 0.
|
| 5 |
-
"A->I_r5": 0.
|
| 6 |
-
"A->T_r1": 0.
|
| 7 |
-
"A->T_r10": 0.
|
| 8 |
-
"A->T_r5": 0.
|
| 9 |
-
"I->A_r1": 0.
|
| 10 |
-
"I->A_r10": 0.
|
| 11 |
-
"I->A_r5": 0.
|
| 12 |
-
"I->T_r1": 0.
|
| 13 |
-
"I->T_r10": 0.
|
| 14 |
-
"I->T_r5": 0.
|
| 15 |
-
"T->A_r1": 0.
|
| 16 |
-
"T->A_r10": 0.
|
| 17 |
-
"T->A_r5": 0.
|
| 18 |
-
"T->I_r1": 0.
|
| 19 |
-
"T->I_r10": 0.
|
| 20 |
-
"T->I_r5": 0.
|
| 21 |
},
|
| 22 |
"_meta": {
|
| 23 |
"audio_suffix": "mn20_audioheavy_lora1280_audio_features",
|
| 24 |
-
"checkpoint": "/shared/augmem/triembed/checkpoints/
|
| 25 |
"device": "NVIDIA GeForce GT 1030",
|
| 26 |
"dims": [
|
| 27 |
512
|
| 28 |
],
|
| 29 |
-
"encoder_name": "mobilenetv4_conv_medium",
|
| 30 |
-
"image_suffix": "
|
| 31 |
},
|
| 32 |
"speech_chatterbox-512": {
|
| 33 |
-
"A->T_r1": 0.
|
| 34 |
-
"A->T_r10": 0.
|
| 35 |
-
"A->T_r5": 0.
|
| 36 |
-
"T->A_r1": 0.
|
| 37 |
-
"T->A_r10": 0.
|
| 38 |
-
"T->A_r5": 0.
|
| 39 |
}
|
| 40 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"SALT-512": {
|
| 3 |
+
"A->I_r1": 0.21484297513961792,
|
| 4 |
+
"A->I_r10": 0.6691338419914246,
|
| 5 |
+
"A->I_r5": 0.5125024914741516,
|
| 6 |
+
"A->T_r1": 0.12262453138828278,
|
| 7 |
+
"A->T_r10": 0.41028207540512085,
|
| 8 |
+
"A->T_r5": 0.30946189165115356,
|
| 9 |
+
"I->A_r1": 0.1970394104719162,
|
| 10 |
+
"I->A_r10": 0.6443288922309875,
|
| 11 |
+
"I->A_r5": 0.48849770426750183,
|
| 12 |
+
"I->T_r1": 0.3178635835647583,
|
| 13 |
+
"I->T_r10": 0.5503100752830505,
|
| 14 |
+
"I->T_r5": 0.4920984208583832,
|
| 15 |
+
"T->A_r1": 0.12722544372081757,
|
| 16 |
+
"T->A_r10": 0.41768354177474976,
|
| 17 |
+
"T->A_r5": 0.31926384568214417,
|
| 18 |
+
"T->I_r1": 0.3424685001373291,
|
| 19 |
+
"T->I_r10": 0.5625125169754028,
|
| 20 |
+
"T->I_r5": 0.5149030089378357
|
| 21 |
},
|
| 22 |
"_meta": {
|
| 23 |
"audio_suffix": "mn20_audioheavy_lora1280_audio_features",
|
| 24 |
+
"checkpoint": "/shared/augmem/triembed/checkpoints/ess_aist_full_v9_subjectfix_l4k/best_model.pt",
|
| 25 |
"device": "NVIDIA GeForce GT 1030",
|
| 26 |
"dims": [
|
| 27 |
512
|
| 28 |
],
|
| 29 |
+
"encoder_name": "mobilenetv4_conv_medium.e180_r384_in12k",
|
| 30 |
+
"image_suffix": "mobilenetv4_conv_medium"
|
| 31 |
},
|
| 32 |
"speech_chatterbox-512": {
|
| 33 |
+
"A->T_r1": 0.32760000228881836,
|
| 34 |
+
"A->T_r10": 0.717199981212616,
|
| 35 |
+
"A->T_r5": 0.6119999885559082,
|
| 36 |
+
"T->A_r1": 0.32019999623298645,
|
| 37 |
+
"T->A_r10": 0.7089999914169312,
|
| 38 |
+
"T->A_r5": 0.6046000123023987
|
| 39 |
}
|
| 40 |
}
|
subject_eval.json
CHANGED
|
@@ -1,118 +1,118 @@
|
|
| 1 |
{
|
| 2 |
-
"checkpoint": "/shared/augmem/triembed/checkpoints/
|
| 3 |
"split": "val",
|
| 4 |
-
"records_path": "/shared/augmem/triembed/checkpoints/ess_ait_86m_20260430T035907Z/
|
| 5 |
"views": {
|
| 6 |
"semantic_key": {
|
| 7 |
"subject_same_different_auc": {
|
| 8 |
-
"auc": 0.
|
| 9 |
-
"positive_pairs":
|
| 10 |
-
"negative_pairs":
|
| 11 |
-
"positive_mean": 0.
|
| 12 |
-
"negative_mean": 0.
|
| 13 |
},
|
| 14 |
"same_topic_different_subject_rejection_auc": {
|
| 15 |
-
"auc": 0.
|
| 16 |
-
"positive_pairs":
|
| 17 |
-
"negative_pairs":
|
| 18 |
-
"positive_mean": 0.
|
| 19 |
-
"negative_mean": 0.
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"subject_key": {
|
| 23 |
"subject_same_different_auc": {
|
| 24 |
-
"auc": 0.
|
| 25 |
-
"positive_pairs":
|
| 26 |
-
"negative_pairs":
|
| 27 |
-
"positive_mean": 0.
|
| 28 |
-
"negative_mean": 0.
|
| 29 |
},
|
| 30 |
"same_topic_different_subject_rejection_auc": {
|
| 31 |
-
"auc": 0.
|
| 32 |
-
"positive_pairs":
|
| 33 |
-
"negative_pairs":
|
| 34 |
-
"positive_mean": 0.
|
| 35 |
-
"negative_mean": 0.
|
| 36 |
}
|
| 37 |
},
|
| 38 |
"event_key": {
|
| 39 |
"subject_same_different_auc": {
|
| 40 |
-
"auc": 0.
|
| 41 |
-
"positive_pairs":
|
| 42 |
-
"negative_pairs":
|
| 43 |
-
"positive_mean": 0.
|
| 44 |
-
"negative_mean": 0.
|
| 45 |
},
|
| 46 |
"same_topic_different_subject_rejection_auc": {
|
| 47 |
-
"auc": 0.
|
| 48 |
-
"positive_pairs":
|
| 49 |
-
"negative_pairs":
|
| 50 |
-
"positive_mean": 0.
|
| 51 |
-
"negative_mean": 0.
|
| 52 |
}
|
| 53 |
},
|
| 54 |
"full_key": {
|
| 55 |
"subject_same_different_auc": {
|
| 56 |
-
"auc": 0.
|
| 57 |
-
"positive_pairs":
|
| 58 |
-
"negative_pairs":
|
| 59 |
-
"positive_mean": 0.
|
| 60 |
-
"negative_mean": 0.
|
| 61 |
},
|
| 62 |
"same_topic_different_subject_rejection_auc": {
|
| 63 |
-
"auc": 0.
|
| 64 |
-
"positive_pairs":
|
| 65 |
-
"negative_pairs":
|
| 66 |
-
"positive_mean": 0.
|
| 67 |
-
"negative_mean": 0.
|
| 68 |
}
|
| 69 |
},
|
| 70 |
"prefix_512": {
|
| 71 |
"subject_same_different_auc": {
|
| 72 |
-
"auc": 0.
|
| 73 |
-
"positive_pairs":
|
| 74 |
-
"negative_pairs":
|
| 75 |
-
"positive_mean": 0.
|
| 76 |
-
"negative_mean": 0.
|
| 77 |
},
|
| 78 |
"same_topic_different_subject_rejection_auc": {
|
| 79 |
-
"auc": 0.
|
| 80 |
-
"positive_pairs":
|
| 81 |
-
"negative_pairs":
|
| 82 |
-
"positive_mean": 0.
|
| 83 |
-
"negative_mean": 0.
|
| 84 |
}
|
| 85 |
},
|
| 86 |
"prefix_1024": {
|
| 87 |
"subject_same_different_auc": {
|
| 88 |
-
"auc": 0.
|
| 89 |
-
"positive_pairs":
|
| 90 |
-
"negative_pairs":
|
| 91 |
-
"positive_mean": 0.
|
| 92 |
-
"negative_mean": 0.
|
| 93 |
},
|
| 94 |
"same_topic_different_subject_rejection_auc": {
|
| 95 |
-
"auc": 0.
|
| 96 |
-
"positive_pairs":
|
| 97 |
-
"negative_pairs":
|
| 98 |
-
"positive_mean": 0.
|
| 99 |
-
"negative_mean": 0.
|
| 100 |
}
|
| 101 |
},
|
| 102 |
"prefix_1536": {
|
| 103 |
"subject_same_different_auc": {
|
| 104 |
-
"auc": 0.
|
| 105 |
-
"positive_pairs":
|
| 106 |
-
"negative_pairs":
|
| 107 |
-
"positive_mean": 0.
|
| 108 |
-
"negative_mean": 0.
|
| 109 |
},
|
| 110 |
"same_topic_different_subject_rejection_auc": {
|
| 111 |
-
"auc": 0.
|
| 112 |
-
"positive_pairs":
|
| 113 |
-
"negative_pairs":
|
| 114 |
-
"positive_mean": 0.
|
| 115 |
-
"negative_mean": 0.
|
| 116 |
}
|
| 117 |
}
|
| 118 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"checkpoint": "/shared/augmem/triembed/checkpoints/ess_aist_full_v9_subjectfix_l4k/best_model.pt",
|
| 3 |
"split": "val",
|
| 4 |
+
"records_path": "/shared/augmem/triembed/checkpoints/ess_ait_86m_20260430T035907Z/ess_corpus_v9_subject_media_wit4096_speech100k_wavcaps100k_librispeech360_subjectfix/val.jsonl",
|
| 5 |
"views": {
|
| 6 |
"semantic_key": {
|
| 7 |
"subject_same_different_auc": {
|
| 8 |
+
"auc": 0.9562558404233674,
|
| 9 |
+
"positive_pairs": 119068,
|
| 10 |
+
"negative_pairs": 104655,
|
| 11 |
+
"positive_mean": 0.9163722881572546,
|
| 12 |
+
"negative_mean": 0.7671980599613585
|
| 13 |
},
|
| 14 |
"same_topic_different_subject_rejection_auc": {
|
| 15 |
+
"auc": 0.9562558404233674,
|
| 16 |
+
"positive_pairs": 119068,
|
| 17 |
+
"negative_pairs": 104655,
|
| 18 |
+
"positive_mean": 0.9163722881572546,
|
| 19 |
+
"negative_mean": 0.7671980599613585
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"subject_key": {
|
| 23 |
"subject_same_different_auc": {
|
| 24 |
+
"auc": 0.9881162919768391,
|
| 25 |
+
"positive_pairs": 119068,
|
| 26 |
+
"negative_pairs": 104655,
|
| 27 |
+
"positive_mean": 0.8715863582124139,
|
| 28 |
+
"negative_mean": 0.507699292860264
|
| 29 |
},
|
| 30 |
"same_topic_different_subject_rejection_auc": {
|
| 31 |
+
"auc": 0.9881162919768391,
|
| 32 |
+
"positive_pairs": 119068,
|
| 33 |
+
"negative_pairs": 104655,
|
| 34 |
+
"positive_mean": 0.8715863582124139,
|
| 35 |
+
"negative_mean": 0.507699292860264
|
| 36 |
}
|
| 37 |
},
|
| 38 |
"event_key": {
|
| 39 |
"subject_same_different_auc": {
|
| 40 |
+
"auc": 0.9551271013544805,
|
| 41 |
+
"positive_pairs": 119068,
|
| 42 |
+
"negative_pairs": 104655,
|
| 43 |
+
"positive_mean": 0.9246441395894791,
|
| 44 |
+
"negative_mean": 0.7915105798808771
|
| 45 |
},
|
| 46 |
"same_topic_different_subject_rejection_auc": {
|
| 47 |
+
"auc": 0.9551271013544805,
|
| 48 |
+
"positive_pairs": 119068,
|
| 49 |
+
"negative_pairs": 104655,
|
| 50 |
+
"positive_mean": 0.9246441395894791,
|
| 51 |
+
"negative_mean": 0.7915105798808771
|
| 52 |
}
|
| 53 |
},
|
| 54 |
"full_key": {
|
| 55 |
"subject_same_different_auc": {
|
| 56 |
+
"auc": 0.9778614751548688,
|
| 57 |
+
"positive_pairs": 119068,
|
| 58 |
+
"negative_pairs": 104655,
|
| 59 |
+
"positive_mean": 0.907507678522579,
|
| 60 |
+
"negative_mean": 0.7068911181573889
|
| 61 |
},
|
| 62 |
"same_topic_different_subject_rejection_auc": {
|
| 63 |
+
"auc": 0.9778614751548688,
|
| 64 |
+
"positive_pairs": 119068,
|
| 65 |
+
"negative_pairs": 104655,
|
| 66 |
+
"positive_mean": 0.907507678522579,
|
| 67 |
+
"negative_mean": 0.7068911181573889
|
| 68 |
}
|
| 69 |
},
|
| 70 |
"prefix_512": {
|
| 71 |
"subject_same_different_auc": {
|
| 72 |
+
"auc": 0.9562558404233674,
|
| 73 |
+
"positive_pairs": 119068,
|
| 74 |
+
"negative_pairs": 104655,
|
| 75 |
+
"positive_mean": 0.9163722881572546,
|
| 76 |
+
"negative_mean": 0.7671980599613585
|
| 77 |
},
|
| 78 |
"same_topic_different_subject_rejection_auc": {
|
| 79 |
+
"auc": 0.9562558404233674,
|
| 80 |
+
"positive_pairs": 119068,
|
| 81 |
+
"negative_pairs": 104655,
|
| 82 |
+
"positive_mean": 0.9163722881572546,
|
| 83 |
+
"negative_mean": 0.7671980599613585
|
| 84 |
}
|
| 85 |
},
|
| 86 |
"prefix_1024": {
|
| 87 |
"subject_same_different_auc": {
|
| 88 |
+
"auc": 0.9814636892484202,
|
| 89 |
+
"positive_pairs": 119068,
|
| 90 |
+
"negative_pairs": 104655,
|
| 91 |
+
"positive_mean": 0.8989504970086771,
|
| 92 |
+
"negative_mean": 0.6642348380165352
|
| 93 |
},
|
| 94 |
"same_topic_different_subject_rejection_auc": {
|
| 95 |
+
"auc": 0.9814636892484202,
|
| 96 |
+
"positive_pairs": 119068,
|
| 97 |
+
"negative_pairs": 104655,
|
| 98 |
+
"positive_mean": 0.8989504970086771,
|
| 99 |
+
"negative_mean": 0.6642348380165352
|
| 100 |
}
|
| 101 |
},
|
| 102 |
"prefix_1536": {
|
| 103 |
"subject_same_different_auc": {
|
| 104 |
+
"auc": 0.9778614751548688,
|
| 105 |
+
"positive_pairs": 119068,
|
| 106 |
+
"negative_pairs": 104655,
|
| 107 |
+
"positive_mean": 0.907507678522579,
|
| 108 |
+
"negative_mean": 0.7068911181573889
|
| 109 |
},
|
| 110 |
"same_topic_different_subject_rejection_auc": {
|
| 111 |
+
"auc": 0.9778614751548688,
|
| 112 |
+
"positive_pairs": 119068,
|
| 113 |
+
"negative_pairs": 104655,
|
| 114 |
+
"positive_mean": 0.907507678522579,
|
| 115 |
+
"negative_mean": 0.7068911181573889
|
| 116 |
}
|
| 117 |
}
|
| 118 |
}
|