Darayut commited on
Commit
81eab20
·
verified ·
1 Parent(s): 96fccd8

Card: inference time at 0% vs 80% sparsity

Browse files
Files changed (3) hide show
  1. README.md +2 -2
  2. assets/inference_time.png +0 -0
  3. assets/tradeoff.png +0 -3
README.md CHANGED
@@ -36,9 +36,9 @@ Test sets; greedy CTC decoding, no language model. CER for Khmer, WER for Englis
36
  - **English:** it keeps much less of its teacher's accuracy, with about 1.5–2× the teacher's WER.
37
  - **In-domain test sets:** FLEURS-km and FLEURS-en train are in the training mix, so their test sets are in-domain.
38
 
39
- <p align="center"><img src="https://huggingface.co/Darayut/sota-base/resolve/main/assets/tradeoff.png" width="100%" alt="Parameters, speed and test error of each 80%-sparse SOTA model next to its fine-tuned teacher"></p>
40
 
41
- *What 80% sparsity buys and costs.* Each SOTA model (orange) next to its fine-tuned teacher (blue); percentages are the change against the teacher, and speed is the measured real-time factor relative to the teacher.
42
 
43
  ## Architecture
44
 
 
36
  - **English:** it keeps much less of its teacher's accuracy, with about 1.5–2× the teacher's WER.
37
  - **In-domain test sets:** FLEURS-km and FLEURS-en train are in the training mix, so their test sets are in-domain.
38
 
39
+ <p align="center"><img src="https://huggingface.co/Darayut/sota-base/resolve/main/assets/inference_time.png" width="60%" alt="Inference time for 60 s of audio at 0% and 80% encoder sparsity, for the base and large models"></p>
40
 
41
+ *Inference time* for one 60 s utterance (batch 1, bf16, median of 20 runs), at 0% sparsity (the fine-tuned teacher) and 80% (SOTA).
42
 
43
  ## Architecture
44
 
assets/inference_time.png ADDED
assets/tradeoff.png DELETED

Git LFS Details

  • SHA256: 710ff0dcf1c8b6f512c70538cb7b2c11e112220ccb2a89b075f781076586899b
  • Pointer size: 131 Bytes
  • Size of remote file: 151 kB