From 653b7ca5b83a8d017ac03dbee4dbe398ccd3da53 Mon Sep 17 00:00:00 2001
From: Andres Marafioti <andimarafioti@gmail.com>
Date: Wed, 28 Aug 2024 12:00:48 +0200
Subject: [PATCH] Assigning min new tokens to a compiled whisper graph on a
thread breaks it
---
STT/whisper_stt_handler.py | 2 +-
arguments_classes/whisper_stt_arguments.py | 6 ------
2 files changed, 1 insertion(+), 7 deletions(-)
diff --git a/STT/whisper_stt_handler.py b/STT/whisper_stt_handler.py
index ea37b8b..b1596a9 100644
--- a/STT/whisper_stt_handler.py
+++ b/STT/whisper_stt_handler.py
@@ -67,7 +67,7 @@ class WhisperSTTHandler(BaseHandler):
# generating more tokens than previously will trigger CUDA graphs capture
# one should warmup with a number of generated tokens above max tokens targeted for subsequent generation
warmup_gen_kwargs = {
- "min_new_tokens": self.gen_kwargs["min_new_tokens"],
+ "min_new_tokens": self.gen_kwargs["max_new_tokens"], # Yes, assign max_new_tokens to min_new_tokens
"max_new_tokens": self.gen_kwargs["max_new_tokens"],
**self.gen_kwargs,
}
diff --git a/arguments_classes/whisper_stt_arguments.py b/arguments_classes/whisper_stt_arguments.py
index 2edb4c2..bed382d 100644
--- a/arguments_classes/whisper_stt_arguments.py
+++ b/arguments_classes/whisper_stt_arguments.py
@@ -33,12 +33,6 @@ class WhisperSTTHandlerArguments:
"help": "The maximum number of new tokens to generate. Default is 128."
},
)
- stt_gen_min_new_tokens: int = field(
- default=0,
- metadata={
- "help": "The minimum number of new tokens to generate. Default is 0."
- },
- )
stt_gen_num_beams: int = field(
default=1,
metadata={
--
GitLab