Correct hyper-link to Architecture Blog
Browse files
README.md
CHANGED
|
@@ -79,7 +79,7 @@ DualFFN acts as an audio-specific expert, boosting the LLM's performance with mi
|
|
| 79 |
Our implementation preserves 91% of the original LLM’s training speed with the inclusion of DualFFN, which has 2.2B parameters.
|
| 80 |
Thus, the total number of parameter for Higgs Audio v2 is 3.6B (LLM) + 2.2B (Audio Dual FFN), and it has the same training / inference FLOPs as Llama-3.2-3B.
|
| 81 |
Ablation study shows that the model equipped with DualFFN consistently outperforms its counterpart in terms of word error rate (WER) and speaker similarity.
|
| 82 |
-
See [our architecture blog](https://github.com/boson-ai/higgs-audio/tech_blogs/ARCHITECTURE_BLOG.md) for more information.
|
| 83 |
|
| 84 |
|
| 85 |
## Evaluation
|
|
|
|
| 79 |
Our implementation preserves 91% of the original LLM’s training speed with the inclusion of DualFFN, which has 2.2B parameters.
|
| 80 |
Thus, the total number of parameter for Higgs Audio v2 is 3.6B (LLM) + 2.2B (Audio Dual FFN), and it has the same training / inference FLOPs as Llama-3.2-3B.
|
| 81 |
Ablation study shows that the model equipped with DualFFN consistently outperforms its counterpart in terms of word error rate (WER) and speaker similarity.
|
| 82 |
+
See [our architecture blog](https://github.com/boson-ai/higgs-audio/blob/main/tech_blogs/ARCHITECTURE_BLOG.md) for more information.
|
| 83 |
|
| 84 |
|
| 85 |
## Evaluation
|