Add secret flag in command. More HF setup steps.

advanced/whisperx/README.md

@@ -11,38 +11,51 @@ Similar to the Whisper JAX example, if you are running locally, we recommend you
 Usually, when you run different AI models, they require specific dependencies that sometimes conflict with each other. This is particularly true in the whisper case - from `requirements.txt`, you may notice that there are quite a bit of specific version requirements.
 
 This is where having a separate service like Lepton becomes super useful: we can create a python environment (using e.g. conda or virtualenv), installed the required dependencies, run the photon as a web service, and then in the regular python environment, simply call the web service as if we were using a regular python function. Comparing to some apparent choices:
+
 - unlike a single python environment, we don't need to resolve version conflicts of different algorithms;
 - unlike packing everything in a separate opaque container image, we are much more lightweighted: only a python environment and dependencies are needed.
 
-## Running with a custom environment.
+## Prerequisite
+
+Note that one of the dependency relies on 3 Hugging Face Hub models that would require you to sign some terms of usage beforehand. Otherwise it will throw error. Simply proceed to the website for [Segmentation](https://huggingface.co/pyannote/segmentation) , [Voice Activity Detection (VAD)](https://huggingface.co/pyannote/voice-activity-detection) , and [Speaker Diarization](https://huggingface.co/pyannote/speaker-diarization) and sign the terms.
+
+![Pyannote Model Term Agreement](assets/pyannote.png)
+
+You would also need a Hugging Face Access Token at hand. Simply following the steps in the [official guide](https://huggingface.co/docs/hub/security-tokens).
+
+## Running with a custom environment
 
 We recommend you use conda or virtualenv to start a whisper-specific environment. For example, if you use conda, it's easy to do:
 
 ```shell
 # pick a python version of your favorite
-conda create -n whisperx python=3.10 
+conda create -n whisperx python=3.10
 conda activate whisperx
 ```
 
 After that, install lepton [per the installation instruction](https://www.lepton.ai/docs/overview/quickstart#1-installation), and install the required dependencies of this demo via:
+
 ```shell
 pip install -r requirements.txt
 ```
 
 After this, you can launch whisperx like:
+
 ```shell
 # Set your huggingface token. This is required to obtain the respective models.
 export HUGGING_FACE_HUB_TOKEN="replace-with-your-own-token"
 python main.py
 ```
 
 It will download the paramaters and start the server. After that, use the regular python client to access the model:
+
 ```python
 from leptonai.client import Client, local
 c = Client(local())
 ```
 
 and invoke transcription or translation as follows:
+
 ```python
 >> c.run(filename="assets/thequickbrownfox.wav")
 [{'start': 0.028,
@@ -122,10 +135,12 @@ lep login
 lep photon create -n whisperx -m main.py
 lep photon push -n whisperx
 # An A10 machine is usually big enough to run the large-v2 model.
-lep photon run -n whisperx --resource-shape gpu.a10
+# note you need to specify the secret that needs to be available in the run
+lep photon run -n whisperx --resource-shape gpu.a10 --secret HUGGING_FACE_ACCESS_TOKEN
 ```
 
 After that, you can use the `lep deployment status` to obtain the public address of the photon, and use the same slack app to connect to it:
+
 ```shell
 >> lep deployment status -n whisperx
 Created at: 2023-08-09 20:24:48
@@ -144,6 +159,7 @@ Replicas List:
 ```
 
 To access the model, we can create a client similar to the local case, simply replace `local()` with the workspace, deployment name, and token. Also, since we are running remote now, we will need to upload the audio files. This is done by calling the `run_updload` path:
+
 ```python
 >> from leptonai.client import Client
 >> from leptonai.photon import FileParam
@@ -193,6 +209,7 @@ To access the model, we can create a client similar to the local case, simply re
 ```
 
 Unlike local deployment, running on the Lepton cloud platform comes with a series of advantages, especially in the whisperx case:
+
 - You do not need to worry about reproducible software environment. The photon is guaranteed to run on the same environment as you created it.
 - Scaling is easier - you can simply increase the number of replicas if you need more capacity.
 - Automatic fault tolerance - if the photon crashes, it will be automatically restarted.