File size: 1,790 Bytes
179528d 0bbe73f 4815a23 179528d 1eb4f7b 40ce004 105e839 40ce004 3445ad5 40ce004 3445ad5 40ce004 3445ad5 40ce004 3445ad5 105e839 3445ad5 105e839 4815a23 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 |
---
license: mit
language:
- en
tags:
- sparse sparsity quantized onnx embeddings int8
---
This is the sparsified ONNX variant of the [bge-large-en-v1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) embeddings model created with [DeepSparse Optimum](https://github.com/neuralmagic/optimum-deepsparse) for ONNX export/inference pipeline and Neural Magic's [Sparsify](https://account.neuralmagic.com/signin?client_id=d04a5f0c-983d-11ed-88a6-971073f187d3&return_to=https%3A//accounts.neuralmagic.com/v1/connect/authorize%3Fscope%3Dsparsify%3Aread%2Bsparsify%3Awrite%2Buser%3Aapi-key%3Aread%2Buser%3Aprofile%3Awrite%2Buser%3Aprofile%3Aread%26response_type%3Dcode%26code_challenge_method%3DS256%26redirect_uri%3Dhttps%3A//apps.neuralmagic.com/sparsify/oidc/callback.html%26state%3Da9b466a6193c4a7b92cba469408d2495%26client_id%3Dd04a5f0c-983d-11ed-88a6-971073f187d3%26code_challenge%3DP0EkmKBpplTb7crJOGS8YLSwT8UH-BeuD0wuE4JTORQ%26response_mode%3Dquery) for One-Shot INT8 quantization and unstructured pruning (50%).
Model achieves 98% accuracy recovery on the STSB validation dataset vs. [dense ONNX variant](https://huggingface.co/zeroshot/bge-base-en-v1.5-dense).
Current up-to-date list of sparse and quantized bge ONNX models:
[zeroshot/bge-large-en-v1.5-sparse](https://huggingface.co/zeroshot/bge-large-en-v1.5-sparse)
[zeroshot/bge-large-en-v1.5-quant](https://huggingface.co/zeroshot/bge-large-en-v1.5-quant)
[zeroshot/bge-base-en-v1.5-sparse](https://huggingface.co/zeroshot/bge-base-en-v1.5-sparse)
[zeroshot/bge-base-en-v1.5-quant](https://huggingface.co/zeroshot/bge-base-en-v1.5-quant)
[zeroshot/bge-small-en-v1.5-sparse](https://huggingface.co/zeroshot/bge-small-en-v1.5-sparse)
[zeroshot/bge-small-en-v1.5-quant](https://huggingface.co/zeroshot/bge-small-en-v1.5-quant) |