Hi there, I used this to fine-tune the 0.6B base variant on a custom dataset so that I can use it in Home Assistant. When converting to OpenVINO INT8 using the official OV notebook, I'm getting vastly different filenames than what you offer on HF:
-rw-r--r-- 1 root root 904 Apr 30 11:18 added_tokens.json
-rw-r--r-- 1 root root 4.9K Apr 30 11:18 config.json
-rw-r--r-- 1 root root 1.6M Apr 30 11:18 merges.txt
-rw-r--r-- 1 root root 61M Apr 30 11:18 openvino_talker_code_predictor_embedding_model.bin
-rw-r--r-- 1 root root 41K Apr 30 11:18 openvino_talker_code_predictor_embedding_model.xml
-rw-r--r-- 1 root root 211M Apr 30 11:18 openvino_talker_code_predictor_model.bin
-rw-r--r-- 1 root root 557K Apr 30 11:18 openvino_talker_code_predictor_model.xml
-rw-r--r-- 1 root root 6.1M Apr 30 11:18 openvino_talker_embedding_model.bin
-rw-r--r-- 1 root root 2.9K Apr 30 11:18 openvino_talker_embedding_model.xml
-rw-r--r-- 1 root root 424M Apr 30 11:18 openvino_talker_language_model.bin
-rw-r--r-- 1 root root 2.7M Apr 30 11:18 openvino_talker_language_model.xml
-rw-r--r-- 1 root root 594M Apr 30 11:18 openvino_talker_text_embedding_model.bin
-rw-r--r-- 1 root root 2.9K Apr 30 11:18 openvino_talker_text_embedding_model.xml
-rw-r--r-- 1 root root 13M Apr 30 11:18 openvino_talker_text_projection_model.bin
-rw-r--r-- 1 root root 6.8K Apr 30 11:18 openvino_talker_text_projection_model.xml
drwxr-sr-x 1 root root 0 Apr 30 11:30 .ov_cache
-rw-r--r-- 1 root root 1008 Apr 30 11:18 special_tokens_map.json
drwxr-sr-x 1 root root 414 Apr 30 11:19 speech_tokenizer
-rw-r--r-- 1 root root 7.3K Apr 30 11:18 tokenizer_config.json
-rw-r--r-- 1 root root 11M Apr 30 11:18 tokenizer.json
-rw-r--r-- 1 root root 2.7M Apr 30 11:18 vocab.json
I tried renaming them to match expected filenames and was able to load the model into openarc, but when i try to use it I just get Port for tensor name token_ids was not found.. Could you share the conversion procedure like you did with ASR?
Hi there, I used this to fine-tune the 0.6B base variant on a custom dataset so that I can use it in Home Assistant. When converting to OpenVINO INT8 using the official OV notebook, I'm getting vastly different filenames than what you offer on HF:
I tried renaming them to match expected filenames and was able to load the model into openarc, but when i try to use it I just get
Port for tensor name token_ids was not found.. Could you share the conversion procedure like you did with ASR?