From d8b46e5e4b7dcbc7c1744c278c04a57fe7441d09 Mon Sep 17 00:00:00 2001
From: MayNiklas <info@niklas-steffen.de>
Date: Mon, 25 Sep 2023 18:11:26 +0200
Subject: [PATCH] openai-whisper: 20230314 -> 20230918

---
 .../python-modules/openai-whisper/default.nix | 13 +++++-------
 .../openai-whisper/ffmpeg-path.patch          | 20 +++++++++----------
 2 files changed, 15 insertions(+), 18 deletions(-)

diff --git a/pkgs/development/python-modules/openai-whisper/default.nix b/pkgs/development/python-modules/openai-whisper/default.nix
index 28a3b8035ce6..68f692e4c37a 100644
--- a/pkgs/development/python-modules/openai-whisper/default.nix
+++ b/pkgs/development/python-modules/openai-whisper/default.nix
@@ -5,7 +5,7 @@
 , cudaSupport ? false
 
 # runtime
-, ffmpeg
+, ffmpeg-headless
 
 # propagates
 , numpy
@@ -14,7 +14,6 @@
 , tqdm
 , more-itertools
 , transformers
-, ffmpeg-python
 , numba
 , openai-triton
 , scipy
@@ -26,20 +25,20 @@
 
 buildPythonPackage rec {
   pname = "whisper";
-  version = "20230314";
+  version = "20230918";
   format = "setuptools";
 
   src = fetchFromGitHub {
     owner = "openai";
     repo = pname;
     rev = "refs/tags/v${version}";
-    hash = "sha256-qQCELjRFeRCT1k1CBc3netRtFvt+an/EbkrgnmiX/mc=";
+    hash = "sha256-wBAanFVEIIzTcoX40P9eI26UdEu0SC/xuife/zi2Xho=";
   };
 
   patches = [
     (substituteAll {
       src = ./ffmpeg-path.patch;
-      inherit ffmpeg;
+      ffmpeg = ffmpeg-headless;
     })
   ];
 
@@ -48,7 +47,6 @@ buildPythonPackage rec {
     tqdm
     more-itertools
     transformers
-    ffmpeg-python
     numba
     scipy
     tiktoken
@@ -61,7 +59,7 @@ buildPythonPackage rec {
 
   postPatch = ''
     substituteInPlace requirements.txt \
-      --replace "tiktoken==0.3.1" "tiktoken>=0.3.1"
+      --replace "tiktoken==0.3.3" "tiktoken>=0.3.3"
   ''
   # openai-triton is only needed for CUDA support.
   # triton needs CUDA to be build.
@@ -80,7 +78,6 @@ buildPythonPackage rec {
 
   disabledTests = [
     # requires network access to download models
-    "test_tokenizer"
     "test_transcribe"
     # requires NVIDIA drivers
     "test_dtw_cuda_equivalence"
diff --git a/pkgs/development/python-modules/openai-whisper/ffmpeg-path.patch b/pkgs/development/python-modules/openai-whisper/ffmpeg-path.patch
index 6ccde1dcabb8..784168d1f62d 100644
--- a/pkgs/development/python-modules/openai-whisper/ffmpeg-path.patch
+++ b/pkgs/development/python-modules/openai-whisper/ffmpeg-path.patch
@@ -1,13 +1,13 @@
 diff --git a/whisper/audio.py b/whisper/audio.py
-index a6074e8..da18350 100644
+index 4f5b6e0..bfe7924 100644
 --- a/whisper/audio.py
 +++ b/whisper/audio.py
-@@ -41,7 +41,7 @@ def load_audio(file: str, sr: int = SAMPLE_RATE):
-         out, _ = (
-             ffmpeg.input(file, threads=0)
-             .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=sr)
--            .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
-+            .run(cmd=["@ffmpeg@/bin/ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
-         )
-     except ffmpeg.Error as e:
-         raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e
+@@ -44,7 +44,7 @@ def load_audio(file: str, sr: int = SAMPLE_RATE):
+     # and resampling as necessary.  Requires the ffmpeg CLI in PATH.
+     # fmt: off
+     cmd = [
+-        "ffmpeg",
++        "@ffmpeg@/bin/ffmpeg",
+         "-nostdin",
+         "-threads", "0",
+         "-i", file,