Minor-fixes (#415)

* switch to new fairseq

* clean up requirements

* update changelog

* fix decimals in avg time extension

* add voices and piano model recommendation

* fix fairseq version

* switch fairseq to new mac wheel

* fix tortoise voices for React UI

* note additional torch version

Author: rsxdalv

README.md

@@ -74,6 +74,24 @@
 
 ## Changelog
 
+Nov 11:
+* Switch to a fixed fairseq version for windows reducing installation conflicts and speeding up updates.
+
+## October 2024
+
+Oct 28:
+* Added installer tests, model downloader, and pip CPU-only option for Torch.
+
+Oct 24:
+* Downgraded Gradio to 5.1.0 due to a bug.
+* Added test workflows and fixed minor bugs.
+
+Oct 22:
+* Fixed Dockerfile issues for smoother deployment.
+
+Oct 21:
+* Redesigned README: improved Whisper extension, added changelogs for August, September, and October, updated screenshots, and reorganized content.
+
 Oct 19:
 * Fixed extension logs and added new extensions.
 

extensions/builtin/extension_decorator_average_execution_time/main.py

@@ -27,7 +27,7 @@ def wrapper(*args, **kwargs):
         result = fn(*args, **kwargs)
         end = time.time()
         times.append(end - start)
-        print("Average execution time: ", sum(times) / len(times))
+        print(f"Average execution time: {sum(times) / len(times):.3f}")
         return result
 
     return wrapper

installer_scripts/js/initializeApp.js

@@ -8,11 +8,9 @@ const { applyDatabaseConfig } = require("./applyDatabaseConfig.js");
 
 const DEBUG_DRY_RUN = false;
 
-const torchVersion = "2.3.1";
+const torchVersion = "2.3.1"; // 2.4.1+cu118
 const cudaVersion = "11.8";
 
-// xformers==0.0.19 # For torch==2.0.0 project plane
-// xformers==xformers-0.0.22.post7 # For torch==2.1.0 project plane
 const pythonVersion = `3.10.11`;
 const pythonPackage = `python=${pythonVersion}`;
 const ffmpegPackage = `conda-forge::ffmpeg=4.4.2[build=lgpl*]`;

requirements.txt

@@ -1,7 +1,5 @@
 safetensors
-# safetensors==0.3.1 # Until they fix it
 ffmpeg-python # Apache 2.0
-# gradio==3.48.0
 gradio==5.1.0 # MIT License
 python-dotenv==1.0.0
 soundfile==0.12.1; sys_platform == 'win32' # torchaudio

requirements_bark_hubert_quantizer.txt

@@ -1,4 +1,7 @@
 # pre-install fairseq for bark_hubert_quantizer
-fairseq @ https://github.com/Sharrnah/fairseq/releases/download/v0.12.4/fairseq-0.12.4-cp310-cp310-win_amd64.whl ; sys_platform == 'win32' # MIT License
-fairseq==0.12.2 ; sys_platform == 'linux' or sys_platform == 'darwin' # MIT License
+fairseq @ https://github.com/rsxdalv/fairseq/releases/download/v0.12.3/fairseq-0.12.13-cp310-cp310-win_amd64.whl ; sys_platform == 'win32' # MIT License
+# fairseq==0.12.2 ; sys_platform == 'linux' or sys_platform == 'darwin' # MIT License
+fairseq==0.12.2 ; sys_platform == 'linux' # MIT License
+# https://github.com/rsxdalv/fairseq/releases/download/v0.12.3/fairseq-0.12.13-cp310-cp310-macosx_11_0_universal2.whl
+fairseq @ https://github.com/rsxdalv/fairseq/releases/download/v0.12.3/fairseq-0.12.13-cp310-cp310-macosx_11_0_universal2.whl ; sys_platform == 'darwin' # MIT License
 bark_hubert_quantizer @ git+https://github.com/rsxdalv/bark-voice-cloning-HuBERT-quantizer@bark_hubert_quantizer#egg=bark_hubert_quantizer # MIT License

tts_webui/stable_audio/stable_audio_tab.py

@@ -257,9 +257,15 @@ def model_select_ui():
 
 
 def model_download_ui():
-    gr.Markdown(
-        "Models can be found on the [HuggingFace model hub](https://huggingface.co/models?search=stable-audio-open-1.0)."
-    )
+    gr.Markdown("""
+Models can be found on the [HuggingFace model hub](https://huggingface.co/models?search=stable-audio-open-1.0).
+
+Recommended models:
+
+voices: RoyalCities/Vocal_Textures_Main
+piano: RoyalCities/RC_Infinite_Pianos
+original: stabilityai/stable-audio-open-1.0
+        """)
     pretrained_name_text = gr.Textbox(
         label="HuggingFace repo name, e.g. stabilityai/stable-audio-open-1.0",
         value="",

tts_webui/tortoise/tortoise_tab.py

@@ -36,6 +36,7 @@ def tortoise_ui():
                         value="Press refresh to load the list",
                         show_label=False,
                         container=False,
+                        allow_custom_value=True,
                     )
                     gr_open_button_simple(
                         TORTOISE_VOICE_DIR_ABS, api_name="tortoise_open_voices"