An ONNX copy of hotchpotch/bekko-embedding-v1-a25m at revision
44f0b8af0f487acd0ccf1a7cb7ae7a29a6dfc09c, made from its official ONNX export onnx/model.onnx. The only change is storage:
the weights are moved into the external-data sidecar onnx/model.onnx_data so they can be
memory-mapped instead of copied. The numerical values are unchanged.
Same tokenizer, same MEAN pooling, same embedding dimension; L2-normalize the pooled vector.
The initializers of the official export were saved as ONNX external data (one file,
onnx/model.onnx_data, tensors of at least 1024 bytes). Nothing was quantized or otherwise altered.
uv run scripts/prepare-embedding-artifact.py --repo hotchpotch/bekko-embedding-v1-a25m --revision 44f0b8af0f487acd0ccf1a7cb7ae7a29a6dfc09c \
--onnx onnx/model.onnx --pooling mean --mode external --out <dir>
The script is scripts/prepare-embedding-artifact.py in the Slipwiki repository. Dependencies:
onnx 1.23.2, onnxruntime 1.24.4, numpy 2.3.5,
tokenizers 0.22.2, onnx-ir 1.1.0, CPython 3.12.14. Re-running it reproduces the
model files byte for byte.
On 13 built-in texts (English, Chinese, Japanese, French, German, very short and truncated), the model output is bitwise identical to the official file's at batch size 1 and at batch size 8 (maximum absolute difference 0).
| File | Bytes | SHA-256 |
|---|---|---|
1_Pooling/config.json |
312 | 029891eaab443d10165e66e6f3d5d9008a16b712481f71ae7c938234e8fdeca1 |
LICENSE |
1395 | eddadd645883bb447653fd61247716b8a3aafa382042bc0d27123949481d66e6 |
NOTICE |
545 | f2b36819a0ca5fc5e8b16a0e134c7ae15687dc46cf83e1cdc101a87d0ce6c500 |
config.json |
1749 | 1719a7afae7032fd3b1320af2673740a17ac5fa7e6a12bb5b09e2611bfe8305a |
modules.json |
229 | 8f4b264b80206c830bebbdcae377e137925650a433b689343a63bdc9b3145460 |
onnx/model.onnx |
255684 | 2f7c6ce7746b3b615891d5042bc5bd125203e50a24f314f4f0337c4c35b9cbfa |
onnx/model.onnx_data |
199049728 | b09ee7dd01db2c677cd0446fe854ab3c04b46c656e450b83cb749f1403cffcb9 |
sentence_bert_config.json |
58 | 9ee3935137f87c9ce6662ae17fe7afb7c9c64744a41247359fd228dc37e627b3 |
special_tokens_map.json |
1051 | 6e3204c5a4004719c185007a745d1940471a6fb02521cfa0a00306dbb6bc5903 |
tokenizer.json |
34363442 | 8bd47075711f75a143d1b78e01a41cc65c1c591b00d3cfeffc23db07adce1392 |
tokenizer_config.json |
46634 | 2ee40d1066ac855e1ea38ac422dbe1f3ddf84c9d415e79272adce569668f3de4 |
mit, as the base model; see LICENSE and NOTICE. This is a re-stored (external-data) copy of the base model; the numerical values are unchanged.
The tokenizer is carried over unchanged from the base model, which reuses the Gemma 2 tokenizer (via mmBERT); its terms are as stated by the upstream model.
hotchpotch/bekko-embedding-v1-a25m