fix: replace in-place tensor operations for torch.compile compatibility (#1179)

* docs: fix typo 'tecnical' to 'technical' in all documentation

Fixed spelling error in badge URLs and markdown links across main README and all translated documentation files. This improves consistency and professionalism.

Made-with: Cursor

* fix: replace in-place tensor operations for torch.compile compatibility

Replaced masked_fill_() with torch.where() in llama.py and fill_() with torch.zeros_like() in modded_dac.py. In-place operations can cause issues with torch.compile graph optimization and gradient tracking. This change makes the code more compile-friendly without changing behavior.

Made-with: Cursor

---------

Co-authored-by: AI Agency Bot <bot@aiagency.com>

README.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 
@@ -77,7 +77,7 @@ Fish Audio S2 is the latest model developed by [Fish Audio](https://fish.audio/)
 
 S2 supports fine-grained inline control of prosody and emotion using natural-language tags like `[laugh]`, `[whispers]`, and `[super happy]`, as well as native multi-speaker and multi-turn generation.
 
-Visit the [Fish Audio website](https://fish.audio/) for live playground. Read the [blog post](https://fish.audio/blog/fish-audio-open-sources-s2/) and [tecnical report](https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf) for more details.
+Visit the [Fish Audio website](https://fish.audio/) for live playground. Read the [blog post](https://fish.audio/blog/fish-audio-open-sources-s2/) and [technical report](https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf) for more details.
 
 ### Model Variants
 

docs/README.ar.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/README.ja.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/README.ko.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/README.pt-BR.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/README.zh.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/ar/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/en/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/ja/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/ko/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/pt/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

docs/zh/index.md

@@ -37,7 +37,7 @@
         <img alt="Fish Audio Blog" src="https://img.shields.io/badge/Blog-Fish_Audio_S2-1f7a8c?style=flat-square&logo=readme&logoColor=white"/>
     </a>
     <a target="_blank" href="https://github.com/fishaudio/fish-speech/blob/main/FishAudioS2TecReport.pdf">
-        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Tecnical_Report-b31b1b?style=flat-square"/>
+        <img alt="Paper | Technical Report" src="https://img.shields.io/badge/Paper-Technical_Report-b31b1b?style=flat-square"/>
     </a>
 </div>
 

fish_speech/models/dac/modded_dac.py

@@ -86,8 +86,12 @@ class KVCache(nn.Module):
         )
 
     def clear_cache(self, prompt_len):
-        self.k_cache[:, :, prompt_len:, :].fill_(0)
-        self.v_cache[:, :, prompt_len:, :].fill_(0)
+        self.k_cache[:, :, prompt_len:, :] = torch.zeros_like(
+            self.k_cache[:, :, prompt_len:, :]
+        )
+        self.v_cache[:, :, prompt_len:, :] = torch.zeros_like(
+            self.v_cache[:, :, prompt_len:, :]
+        )
 
 
 class Transformer(nn.Module):

fish_speech/models/text2semantic/llama.py

@@ -961,9 +961,11 @@ class Attention(nn.Module):
 
         if attn_mask is not None:
             if attn_mask.dtype == torch.bool:
-                attn_bias.masked_fill_(attn_mask.logical_not(), float("-inf"))
+                attn_bias = torch.where(
+                    attn_mask.logical_not(), float("-inf"), attn_bias
+                )
             else:
-                attn_bias += attn_mask
+                attn_bias = attn_bias + attn_mask
 
         attn_weight = query @ key.transpose(-2, -1) * scale_factor
         attn_weight += attn_bias