README

README.md

@@ -0,0 +1,61 @@
+
+
+----------------------------------------------------------------
+ Clona Whisper small or medium modellproject
+----------------------------------------------------------------
+
+git clone https://huggingface.co/bjelkenhed/whisper-small-sv
+git clone https://huggingface.co/bjelkenhed/whisper-medium-sv
+
+
+----------------------------------------------------------------
+ Install requirements
+----------------------------------------------------------------
+
+sudo add-apt-repository -y ppa:jonathonf/ffmpeg-4
+sudo apt update
+sudo apt install -y ffmpeg
+sudo apt-get install git-lfs
+
+----------------------------------------------------------------
+ Install python libraries
+----------------------------------------------------------------
+
+env_name=babelbox
+
+python3 -m venv $env_name
+
+source ~/$env_name/bin/activate
+
+git clone https://github.com/huggingface/community-events.git
+
+pip install -r community-events/whisper-fine-tuning-event/requirements.txt
+
+python -c "import torch; print(torch.cuda.is_available())"
+
+
+----------------------------------------------------------------
+  Log in to Huggingface
+----------------------------------------------------------------
+
+git config --global credential.helper store
+huggingface-cli login
+
+Type your Huggingface access token
+
+
+----------------------------------------------------------------
+  Create HF modell or clone existing
+----------------------------------------------------------------
+
+#huggingface-cli repo create whisper-small-sv
+
+git clone https://huggingface.co/bjelkenhed/whisper-small-sv
+
+git config --global user.email "[email protected]"
+git config --global user.name "Magnus Bjelkenhed"
+
+tmux new -s mysession
+
+
+tmux a -t mysession

run.sh

@@ -10,8 +10,9 @@ python run_speech_recognition_seq2seq_streaming.py \
         --model_index_name="Whisper Small Swedish" \
         --max_steps="5000" \
         --output_dir="./" \
-        --per_device_train_batch_size="64" \
+        --per_device_train_batch_size="32" \
         --per_device_eval_batch_size="32" \
+	--gradient_accumulation_steps="2" \
         --logging_steps="25" \
         --learning_rate="1e-5" \
         --warmup_steps="500" \