whisper-1 and llm whisper-api (fixes)

[leifulstrup]

(A) llm-whisper-api does not work without adding whisper-1 as an extra model (it's missing from your standard list)

I tried this: https://simonwillison.net/2024/Oct/27/llm-whisper-api/

I listed llm models and whisper-1 does not show as an option. Your python code depends for extension assumes whisper-1 is included in the standard list.

(B) I then tried to add whisper-1 using your "adding more openai models" instructions and ran into another problem...

I then found used your instructions here: https://llm.datasette.io/en/stable/openai-models.html#adding-more-openai-models

I get this error:

File "/usr/local/Cellar/llm/0.20/libexec/lib/python3.13/site-packages/llm/default_plugins/openai_models.py", line 101, in register_models
model_name = extra_model["model_name"]
~~~~~~~~~~~^^^^^^^^^^^^^^
KeyError: 'model_name'

I reviewed the python code in openai_model.py code and see that it uses:

Load extra models

extra_path = llm.user_dir() / "extra-openai-models.yaml"
if not extra_path.exists():
    return
with open(extra_path) as f:
    extra_models = yaml.safe_load(f)
for extra_model in extra_models:
    model_id = extra_model["model_id"]
    aliases = extra_model.get("aliases", [])
    model_name = extra_model["model_name"]
    api_base = extra_model.get("api_base")
    api_type = extra_model.get("api_type")
    api_version = extra_model.get("api_version")
    api_engine = extra_model.get("api_engine")
    headers = extra_model.get("headers")
    kwargs = {}

It looks like model_name = extra_model["model_name"] should be model_name = extra_model.get("model_name", [])

and/OR the instructions for extra-openai-models.yaml should say:

• model_id: whisper-1
  aliases: ["whisper-1"]
  model_name: "whisper-1"

(C) You should include whisper-1 in the default list of openai models or explain here that it needs to be added with how to do it - https://simonwillison.net/2024/Oct/27/llm-whisper-api/

(D) I've fixed the yaml to add the line for model_name: "whisper-1" and it loads the extra model and llm whisper-api myfile.mp3 > transcript.txt works as advertised

(E) it would be great if that extension chunked a lonk mp3 or wave up into 25m or less chunks since whisper-1 balks if it is a long audio file

*** Thank you for all of your YT videos, blog, and bsky posts! You are amazingly talented.

The LLM cli tool is terrific and I'm using it frequently.