Create install.sh

[oxidian24]

I've written installation script that maintains control over the files, logs every action (download, extraction, move). The script uses sudo to get your permission to create the folder and move the files to /usr/local/lib/openvino-models.

There might be some things to improve or to check like the paths.

mkdir openvino-models is just temporary directory for the logs and git folders, don't copy that to your libraries.

/usr/local/lib/openvino-models# ls * -hC1 --size
142M ggml-base.bin
40M ggml-base-encoder-openvino.bin
284K ggml-base-encoder-openvino.xml
7,4M ggml-medium.bin
8,2M ggml-medium-encoder-openvino.bin
1000K ggml-medium-encoder-openvino.xml
69M ggml-small.bin
169M ggml-small-encoder-openvino.bin
804K ggml-small-encoder-openvino.xml
18M ggml-small.en-tdrz.bin
33M ggml-small.en-tdrz-encoder-openvino.bin
512K ggml-small.en-tdrz-encoder-openvino.xml
97M htdemucs_v4.bin
1,8M htdemucs_v4.xml
5,9M noise-suppression-denseunet-ll-0001.bin
676K noise-suppression-denseunet-ll-0001.xml

deepfilternet2:
total 9,3M
3,2M df_dec.bin
112K df_dec.xml
2,5M enc.bin
176K enc.xml
3,2M erb_dec.bin
184K erb_dec.xml

deepfilternet3:
celkem 8,7M
3,2M df_dec.bin
124K df_dec.xml
1,9M enc.bin
188K enc.xml
3,2M erb_dec.bin
188K erb_dec.xml

/usr/local/lib/openvino-models/musicgen# ls * -hC1 --size 2,0M attention_mask_from_prepare_4d_causal_10s.raw 500K attention_mask_from_prepare_4d_causal_5s.raw
57M encodec_combined_weights.bin
57M encodec_encoder_combined_weights.bin
444K encodec_encoder_10s.xml
444K encodec_encoder_5s.xml
260K encodec_20s.xml
264K encodec_5s.xml
780K musicgen-small-tokenizer.bin
8,0K musicgen-small-tokenizer.xml
53M t5.bin
556K t5.xml

/usr/local/lib/openvino-models/musicgen/mono:
total 1,1G
17M embed_tokens.bin
16K embed_tokens.xml
3,1M enc_to_dec_proj.bin
4,0K enc_to_dec_proj.xml
43M initial_cross_attn_kv_producer.bin
176K initial_cross_attn_kv_producer.xml
17M lm_heads.bin
12K lm_heads.xml
673M musicgen_decoder_combined_weights.bin
338M musicgen_decoder_combined_weights_int8.bin
3,1M musicgen_decoder_static_batch1_int8.xml
2,5M musicgen_decoder_static_batch1.xml
3,1M musicgen_decoder_static_int8.xml
2,5M musicgen_decoder_static.xml
2,5M musicgen_decoder_static0_10s.xml
2,5M musicgen_decoder_static0_5s.xml
8,0M sinusoidal_positional_embedding_weights_2048_1024.raw

/usr/local/lib/openvino-models/musicgen/stereo:
total 1,3G
13M embed_tokens.bin
32K embed_tokens.xml
3,1M enc_to_dec_proj.bin
4,0K enc_to_dec_proj.xml
193M initial_cross_attn_kv_producer.bin
148K initial_cross_attn_kv_producer.xml
33M lm_heads.bin
24K lm_heads.xml
673M musicgen_decoder_combined_weights.bin
338M musicgen_decoder_combined_weights_int8.bin
3,1M musicgen_decoder_static_batch1_int8.xml
2,5M musicgen_decoder_static_batch1.xml
3,1M musicgen_decoder_static_int8.xml
2,5M musicgen_decoder_static.xml
2,5M musicgen_decoder_static0_10s.xml
2,5M musicgen_decoder_static0_5s.xml
8,0M sinusoidal_positional_embedding_weights_2048_1024.raw

The Whisper section might be very long, I wanted to keep the selection of the models which one to download. You can use any function or complete script if you want, but would be good to test it once more. And to edit the large models names, I did not really test the large models name, because I did not see the content of the archives.