From ca182b650e1061f0630ad275baf047e260c864df Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?P=C3=A1l=20Zs=C3=A1mboki?=
 <79205753+zsamboki@users.noreply.github.com>
Date: Thu, 13 Jul 2023 18:37:20 +0200
Subject: [PATCH] Fix code typo in int8-asr.mdx

Having `bias="None"` in `LoraConfig` raised a `NotImplementedError`. Replaced it with `bias="none"` as per the [`LoraConfig` reference](https://huggingface.co/docs/peft/main/en/package_reference/tuners#peft.LoraConfig) and now the code works, I can run training.
---
 docs/source/task_guides/int8-asr.mdx | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/source/task_guides/int8-asr.mdx b/docs/source/task_guides/int8-asr.mdx
index f1ace5ba00..37d63b6d6d 100644
--- a/docs/source/task_guides/int8-asr.mdx
+++ b/docs/source/task_guides/int8-asr.mdx
@@ -205,7 +205,7 @@ Let's also apply LoRA to the training to make it even more efficient. Load a [`~
 ```py
 from peft import LoraConfig, PeftModel, LoraModel, LoraConfig, get_peft_model
 
-config = LoraConfig(r=32, lora_alpha=64, target_modules=["q_proj", "v_proj"], lora_dropout=0.05, bias="None")
+config = LoraConfig(r=32, lora_alpha=64, target_modules=["q_proj", "v_proj"], lora_dropout=0.05, bias="none")
 ```
 
 After you set up the [`~peft.LoraConfig`], wrap it and the base model with the [`get_peft_model`] function to create a [`PeftModel`]. Print out the number of trainable parameters to see how much more efficient LoRA is compared to fully training the model!
@@ -375,4 +375,4 @@ with torch.cuda.amp.autocast():
     text = pipe(audio, generate_kwargs={"forced_decoder_ids": forced_decoder_ids}, max_new_tokens=255)["text"]
 text
 "मी तुमच्यासाठी काही करू शकतो का?"
-```
\ No newline at end of file
+```