skills.rest
DeepLink-org
DeepLink-orgOFFICIAL
Explore all skills by this author →
WebsiteGitHub

hygon-audio-au

Evaluates ECAPA-TDNN language identification inference accuracy and latency on Hygon DCU.

7|1|Updated Apr 17, 2026
Search Tags:#docker#language identification#hygon dcu#inference benchmark#audio understanding#ecapa-tdnn#speechbrain
One-click install
npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill hygon-audio-au-deeplink-org
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hygon-audio-au
Source: https://github.com/DeepLink-org/DeepEval-Skills/tree/main/skills/Hygon/audio/hygon-audio-au
Command: npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill hygon-audio-au-deeplink-org

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running audio understanding model benchmarks on Hygon DCU hardware requires manual container setup, dataset mounting, inference execution, and metric collection, which is error-prone and hard to reproduce. ## Core Features & Use Cases - Automated DCU Container Setup: Launches a preconfigured Docker environment with Hygon DTK, PyTorch, and SpeechBrain, mounting model checkpoints, LMDB datasets, and config files. - Language Identification Inference: Runs the lang-id-voxlingua107-ecapa model on the foundation-lid (VoxLingua107) dataset covering 107 languages. - Metric Collection: Extracts accuracy, average inference time, success rate, and sample counts from acc_report.json into a structured result.json. - Use Case: Ask the agent to test the lang-id-voxlingua107-ecapa model on Hygon DCU, and it will start the container, run inference, and report accuracy and per-sample latency. ## Quick Start Ask the agent to test the lang-id-voxlingua107-ecapa language identification model inference performance on Hygon DCU with the foundation-lid dataset.

Frequently Asked Questions about hygon-audio-au

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark language identification models on Hygon DCU?▼

Launch the provided Docker container with model checkpoints, LMDB datasets, and config mounted, then run infer_runner.py inside the container. The script outputs accuracy and average inference time to acc_report.json.

What models and datasets does Hygon audio understanding evaluation support?▼

It currently supports the lang-id-voxlingua107-ecapa model, an ECAPA-TDNN language identification model covering 107 languages, evaluated on the foundation-lid dataset derived from VoxLingua107 in LMDB format.

What metrics are collected during audio inference evaluation?▼

The evaluation collects accuracy (language identification correctness), avg_inference_time per sample, success_rate, and total_samples. Results are written to acc_report.json and consolidated into result.json.

Why does SpeechBrain model loading fail inside the container?▼

SpeechBrain creates symbolic link caches in the model directory when loading checkpoints. Mount the model checkpoint directory with read-write (:rw) permissions instead of read-only to resolve this.

How do I fix DCU out-of-memory errors during inference?▼

Check VRAM usage with rocm-smi or hy-smi, then set HIP_VISIBLE_DEVICES to an idle DCU card before running the inference script. Ensure the variable is exported before starting Python.

Related Skills

Explore More Software Engineering Skills→
nishide-devnishide-dev

ml-data-pipeline

Create PyTorch Lightning DataModules for vision and graph ML pipelines.

Community
Intermediate
Open330Open330
1

benchmarking-ml-models

Benchmark ML models by measuring latency, throughput, memory, and accuracy.

Official
Intermediate
atrawogatrawog

gpu

Monitor GPU status, VRAM, and loaded models for Ollama inference.

Community
Intermediate
shogo-labsshogo-labs
12

benchmark-models

Run the same prompt across Claude, GPT, and Gemini to benchmark gstack skills.

Official
Advanced