Hugging Face's logo

AbstractPhil
/
captionbert-8192-rotary

Feature Extraction
Transformers
TensorBoard
Safetensors
English
captionbert_structured_rotary
sentence-similarity
consensus-distillation
rotary-embeddings
custom_code
Model card Files
xet
 Metrics Community
captionbert-8192-rotary / campaigns /structured-20261106 /validation /deployment-colab-parity.json
AbstractPhil's picture
AbstractPhil
Publish full structured rotary report, final model card, all results and tested serving release
e349b70 verified
Raw History Blame Contribute Delete
6.44 kB
{
"target": "AbstractPhil/captionbert-8192-rotary",
"revision": "ad14b77f4cc9017d513173f3b340e735af02ef14",
"remote": true,
"native_reference": "independent training state.pt; exact v0.3.0 source",
"source_arm": "N16",
"source_seed": 1,
"source_config_id": "aafe9c59b9d696107cb4393f3a26815e84714daddb7d6eb028b5d6137b88418e",
"state_tensors_equal": true,
"safetensors_metadata": {
"format": "pt"
},
"loading_info": {
"missing_keys": [],
"unexpected_keys": [],
"mismatched_keys": [],
"error_msgs": []
},
"native_training_source_sha256": "094bf61ef954c61a2b200e5e8760348160b76a5fbfd186602dd992e4b62ccfa6",
"tf32": false,
"cases": [
{
"case": "natural",
"precision": "fp32",
"input_shape": [
2,
11
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
11,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.16567078699881677
},
{
"case": "all_padding",
"precision": "fp32",
"input_shape": [
2,
17
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
17,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.019933960000344086
},
{
"case": "synthetic_1x1",
"precision": "fp32",
"input_shape": [
1,
1
],
"embedding_shape": [
1,
768
],
"hidden_shape": [
1,
1,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.056073969000863144
},
{
"case": "synthetic_2x17",
"precision": "fp32",
"input_shape": [
2,
17
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
17,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.019162890002917266
},
{
"case": "synthetic_3x128",
"precision": "fp32",
"input_shape": [
3,
128
],
"embedding_shape": [
3,
768
],
"hidden_shape": [
3,
128,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.019288038998638513
},
{
"case": "synthetic_1x512",
"precision": "fp32",
"input_shape": [
1,
512
],
"embedding_shape": [
1,
768
],
"hidden_shape": [
1,
512,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.019747980000829557
},
{
"case": "boundary_shape_1025",
"precision": "fp32",
"input_shape": [
1,
1025
],
"embedding_shape": [
1,
768
],
"hidden_shape": [
1,
1025,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.02409016999808955
},
{
"case": "boundary_shape_8192",
"precision": "fp32",
"input_shape": [
1,
8192
],
"embedding_shape": [
1,
768
],
"hidden_shape": [
1,
8192,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.19234563700229046
},
{
"case": "natural",
"precision": "fp16_autocast",
"input_shape": [
2,
11
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
11,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.08385088800059748
},
{
"case": "all_padding",
"precision": "fp16_autocast",
"input_shape": [
2,
17
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
17,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.019784469997830456
},
{
"case": "natural",
"precision": "bf16_autocast",
"input_shape": [
2,
11
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
11,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.07453705900115892
},
{
"case": "all_padding",
"precision": "bf16_autocast",
"input_shape": [
2,
17
],
"embedding_shape": [
2,
768
],
"hidden_shape": [
2,
17,
512
],
"max_embedding_delta": 0.0,
"max_hidden_delta": 0.0,
"seconds": 0.020809778998227557
},
{
"case": "left_padding",
"invariance_required": false,
"max_unpadded_embedding_delta": 3.725290298461914e-08,
"max_native_auto_embedding_delta": 0.0
},
{
"case": "right_padding",
"invariance_required": true,
"max_unpadded_embedding_delta": 5.029141902923584e-08,
"max_native_auto_embedding_delta": 0.0
}
],
"source_checkpoint_sha256": "f3f5442c5a0ff08a02c0a3d9cc46b94f57e5a69204bb98ba99b0dc6d407fd6d9",
"serving_safetensors_sha256": "982569dba04fa0e40d8f53b7d40ccfbef5355d502bb3cd23f4951fab79c10af9",
"environment": {
"python": "3.13.16",
"torch": "2.15.0+cu132",
"transformers": "5.22.1",
"safetensors": "0.8.0",
"cuda": "13.2",
"gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
"huggingface_hub": "2.4.0"
},
"hidden_states_count": 13,
"attention_layers": 12,
"attention_head_count": 16,
"memory": {
"peak_allocated_gb": 0.761166336,
"peak_reserved_gb": 0.88080384,
"max_observed_device_gb": 1.632632832,
"total_device_limit_gb": 92.0
},
"training_performed": false,
"padding_contract": "Right padding matches training. Left padding is measured against the native model, not assumed invariant.",
"long_shapes_are_engineering_only": true,
"verifier_source_commit": "192e29ec49cf70d2ef8e9eb3430b70eb79c6cc20",
"source_experiment": "experiments/aafe9c59b9d696107cb4393f3a26815e84714daddb7d6eb028b5d6137b88418e/N16-seed1"
}