Vikhrmodels
/

Borealis-5b-it

Audio-Text-to-Text

feature-extraction

Model card Files Files and versions

AlexWortega commited on Dec 15, 2025

Commit

45d787f

·

verified ·

1 Parent(s): 37a410e

Update README.md

Files changed (1) hide show

README.md +8 -8

README.md CHANGED Viewed

@@ -281,14 +281,14 @@ vLLM provides ~2x speedup over HuggingFace with full audio processing support.
 ### ASR Benchmarks (WER / CER)
-| Dataset | Baseline | Borealis-5B-IT |
-|---------|----------|----------------|
-| Russian LibriSpeech | 6.63 / 3.49 | **5.64 / 2.59** |
-| Common Voice | **8.88 / 5.04** | 12.67 / 8.59 |
-| Tone Webinars | **56.87 / 52.47** | 60.55 / 54.20 |
-| Tone Books | 6.03 / 3.75 | **5.25 / 2.75** |
-| Tone Speak | **4.63 / 3.38** | 6.49 / 5.19 |
-| Sova RuDevices | **17.28 / 8.03** | 21.57 / 11.48 |
 *Baseline: Whisper Large V3. Lower is better.*

 ### ASR Benchmarks (WER / CER)
+ | Split               | Borealis baseline | Borealis step-2898 | Whisper-v3 |
+  |---------------------|-------------------|--------------------|------------|
+  | Russian_LibriSpeech | 6.63%             | 5.64%              | 11.68%     |
+  | Common_Voice        | 8.88%             | 12.67%             | 12.23%     |
+  | Tone_Webinars       | 56.87%            | 60.55%             | 7.77%      |
+  | Tone_Books          | 6.03%             | 5.25%              | 11.95%     |
+  | Tone_Speak          | 4.63%             | 6.49%              | 2.68%      |
+  | Sova_RuDevices      | 17.28%            | 21.57%             | 19.87%     |
 *Baseline: Whisper Large V3. Lower is better.*