Skip to content
cccskills
FreePublished on GitHub

distil-whisper

Routes Distil-Whisper inference, PyTorch distillation training, and Flax reproduction workflows.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill
See how to install

Files included (19)

  • SKILL.md2.3 KB
  • references/model-overview.md2.1 KB
  • references/repo-provenance.md1.4 KB
  • references/repo-routing-metadata.json435 B
  • references/troubleshooting.md2.9 KB
  • references/workflows.md2.4 KB
  • scripts/check-env.py3.5 KB
  • sub-skills/flax-reproduction/references/troubleshooting.md2.6 KB
  • sub-skills/flax-reproduction/references/workflows.md2.6 KB
  • sub-skills/flax-reproduction/scripts/convert_train_state_to_hf.py12.5 KB
  • sub-skills/flax-reproduction/scripts/create_student_model.py8.4 KB
  • sub-skills/flax-reproduction/SKILL.md1.9 KB
  • sub-skills/inference/references/troubleshooting.md1.8 KB
  • sub-skills/inference/references/workflows.md3.2 KB
  • sub-skills/inference/SKILL.md1.7 KB
  • sub-skills/pytorch-training/references/troubleshooting.md2.4 KB
  • sub-skills/pytorch-training/references/workflows.md3.6 KB
  • sub-skills/pytorch-training/scripts/create_student_model.py8.5 KB
  • sub-skills/pytorch-training/SKILL.md1.6 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Distil-Whisper

Purpose

Use this skill when the user wants to transcribe audio with Distil-Whisper, choose or compare checkpoints, run the PyTorch distillation stack under training/, or reproduce the original JAX/Flax stack under training/flax/.

First stop

  • Read references/model-overview.md to choose a checkpoint and decode path.
  • Run scripts/check-env.py to verify the CPU inspection stack before deeper work.
  • Read references/repo-provenance.md when deciding whether this skill still matches the current checkout.

Route map

  • sub-skills/inference/SKILL.md for checkpoint usage, short-form transcription, long-form transcription, speculative decoding, and inference-speed trade-offs.
  • sub-skills/pytorch-training/SKILL.md for pseudo-labelling, student initialization, distillation training, evaluation, and dataset mixing in the PyTorch stack.
  • sub-skills/flax-reproduction/SKILL.md for the JAX/Flax package, conversion helpers, long-form transcription, and original reproduction workflows.

Install notes

This repo has two practical dependency surfaces:

  • The Flax package under training/flax/.
  • The PyTorch training scripts under training/.

For a local checkout, install the Flax package in editable mode and then add the PyTorch training dependencies from the root training/ metadata. The verified inspection environment used a single CPU prefix with Python 3.11, a CPU-only torch wheel, JAX 0.4.18, Flax 0.7.2, SciPy 1.11.4, and the core Hugging Face audio stack. See references/repo-provenance.md for the exact verified versions.

Minimal smoke

  1. python scripts/check-env.py
  2. python -m pip check
  3. Read the relevant sub-skill before running a workflow command.

Runtime boundaries

  • Do not tell future agents to open or run files from the original repository checkout when a bundled reference or script exists.
  • Do not treat benchmark-only or TPU-host-specific helpers as default user workflows.
  • Keep training-scale runs, model downloads, and Hub pushes in the route-specific references unless the user explicitly wants to perform them.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

3d-resnets-pytorch

Free

Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

3ddfa

Free

Guide 3DDFA Python inference, geometry rendering, training/evaluation, and optional C++ ONNX workflows for 3D dense face alignment.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

3ddfa-v2

Free

Routes 3DDFA_V2 face-alignment setup, still-image demos, video tracking, and ONNX benchmarking workflows.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

ab3dmot

Free

Operate AB3DMOT 3D multi-object tracking workflows for KITTI and nuScenes data, tracking, evaluation, and visualization.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

accelerate

Free

Use Hugging Face Accelerate for PyTorch training-loop migration, distributed launch/configuration, DeepSpeed/FSDP/TPU backend setup, big-model inference/offload, checkpointing, tracking, and troubleshooting.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

acme

Free

Route Acme reinforcement-learning framework tasks across core loops, replay/data, JAX agents, and TensorFlow/Sonnet agents.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill340Updated Sep 3, 2026

See all skills by VectorSpaceLab

Report a problem with this skill