Downloads · 30 days
211
26% of all-time downloads
atjsh/llmlingua-2-js-bert-base-multilingual-cased-meetingbank-onnx-v4
llmlingua-2-js-bert-base-multilingual-cased-meetingbank-onnx-v4 is a token classification model from atjsh. Use it when you need labels on individual words, such as names. It is set up for transformers.js. The card lists the license as apache-2.0.
This is a neutral format-conversion archive of microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank at source revision 5f0c82792b7ea14c6484e015b6a072009496b7f2.
Downloads · 30 days
211
26% of all-time downloads
All-time downloads
804
Public
Repo size
2.1 GB
Likes
0
Public
Click a slice to open those files.
.onnx2.7 GB · 100%
From the Hugging Face model README
This is a neutral format-conversion archive of
microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank
at source revision 5f0c82792b7ea14c6484e015b6a072009496b7f2.
The source repository declares Apache-2.0. The retained license and conversion
notice are in LICENSE and NOTICE. These files were not retrained.
onnx-community/convert-to-onnx@af0ef3070cef5e863b371628ad3dcdc10de4d091conversion/CONVERTER-LICENSEoptimum-cli export onnx --model <pinned-local-source> --task token-classification <output-directory>python conversion/quantize.py onnx/model.onnxconversion/requirements.txtconversion/export.log and conversion/quantize.logconversion/command-status.jsonconversion/manifest.jsonconversion/source-verification.jsonThe commands above are parameterized reproduction recipes, not reconstructed
historical shell invocations. conversion/quantize.py is an adapted fail-fast
extraction of the pinned converter's ModelConverter._apply_quantizations
flow, not a verbatim upstream file. It remains under the pinned converter's
Apache-2.0 terms; its exact upstream app.py and license identities are in
conversion/manifest.json. Exact historical argv and shell exit-code
records were not retained; conversion/command-status.json lists those gaps.
The published converter logs are sanitized copies: converter messages and
warnings are preserved while machine-local paths are replaced with stable
placeholders. The source inventory was independently checked against immutable
Hub revision metadata and downloaded bytes.
The q8 dtype maps to onnx/model_quantized.onnx, an exact copy of the signed
int8 output in this archive; the adapted reproduction script performs that
copy explicitly. The emitted bnb4 file is byte-for-byte identical to the
emitted q4 file. Both converter filenames are retained and measured as
emitted; this archive does not claim that they contain distinct encodings.
| Dtype | File | Bytes | SHA-256 |
|---|---|---|---|
| fp32 | onnx/model.onnx | 709,574,565 | b908335cb2af64c7f0aabf8dd5694b2c4e5594ee38f106e6f3201c6f13ae0c32 |
| fp16 | onnx/model_fp16.onnx | 354,974,103 | e6e7b852d6520515c77d880f5f8c9bff845851807b01f2c5273ce35c5e680294 |
| int8 | onnx/model_int8.onnx | 178,094,385 | a1268c6b428fe7e3efe217a0cd62fd2a0051fbf8c70aaa20806e2024164809d3 |
| uint8 | onnx/model_uint8.onnx | 178,094,427 | d5ce58e66eb569a219d7c8f7d3e4680a8e0cdd2a0d7cf047b5049aac8a37e370 |
| q8 | onnx/model_quantized.onnx | 178,094,385 | a1268c6b428fe7e3efe217a0cd62fd2a0051fbf8c70aaa20806e2024164809d3 |
| q4 | onnx/model_q4.onnx | 422,925,811 | a2a538a305f299c69251985431cf1b4a42b3a4996d23dfd8746fbeabec14c746 |
| q4f16 | onnx/model_q4f16.onnx | 232,876,968 | b53244859bf91e7bff99319ed4d84032bbc7842ebfc8068a431ccb30cc9e3177 |
| bnb4 (emitted q4 alias) | onnx/model_bnb4.onnx | 422,925,811 | a2a538a305f299c69251985431cf1b4a42b3a4996d23dfd8746fbeabec14c746 |
conversion/runtime-cpu.json records the results of a three-run ONNX Runtime
1.28.0 CPU smoke on an Apple M4; its exact historical argv was not retained.
fp32, int8, uint8, q8, q4, and bnb4 loaded, returned finite
[1, 16, 2] logits, and repeated byte-identically. fp16 and q4f16 failed to
load with an ONNX type mismatch. These are observations from that run, not a
runtime-support, numerical-fidelity, or product-suitability claim.
measurement/ records the demo's reproducible quality, warm-latency,
download-size, and renderer-RSS inputs. measurement/selected-defaults.json
is the deterministic result of the demo's balanced Pareto formula. It is not a
repository recommendation, artifact qualification, backend-support claim, or
change to the converter-trust publication policy.