SI

sir-convert-a-lot-qwen-finetuning

Specific fine-tuning workflow for Qwen3-TTS, focusing on Swedish speech data curation and model evaluation.

Install

mkdir -p .claude/skills/sir-convert-a-lot-qwen-finetuning && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/14778" && unzip -o skill.zip -d .claude/skills/sir-convert-a-lot-qwen-finetuning && rm skill.zip

Installs to .claude/skills/sir-convert-a-lot-qwen-finetuning

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Model-specific operator skill for Qwen3-TTS fine-tuning on Hemma and Colab. Use when the task is specifically about Qwen TTS training, Swedish language expansion with Qwen, Qwen preprocessing or runtime policy, or deciding whether a fine-tuned Qwen model should enter the Sir Convert-a-Lot sidecar candidate lane.
313 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Advanced

Key capabilities

  • Fine-tune Qwen3-TTS-12Hz-1.7B-Base model
  • Expand Swedish language support for Qwen
  • Choose between Hemma and Colab H100 for training
  • Manage ROCm, Triton flash attention, or GPU container policy
  • Curate and preprocess Swedish speech data
  • Evaluate trained Qwen models for sidecar candidacy

How it works

This skill manages the fine-tuning process for Qwen3-TTS models, specifically for Swedish language expansion, by guiding infrastructure selection, data preprocessing, and model evaluation.

Inputs & outputs

You give it
Qwen3-TTS model, Swedish speech data
You get back
fine-tuned Qwen model, evaluation results, infrastructure configuration

When to use sir-convert-a-lot-qwen-finetuning

  • Fine-tune Qwen3-TTS for Swedish
  • Preprocess speech corpus data
  • Evaluate speech model performance
  • Configure ROCm or Triton training environments

About this skill

Sir Convert-a-Lot Qwen Finetuning

Use When

  • Fine-tuning Qwen/Qwen3-TTS-12Hz-1.7B-Base for Swedish.
  • Working on Qwen preprocessing, training, evaluation, or recovery.
  • Choosing between Hemma and Colab for a governed Qwen run.
  • Deciding whether Qwen evidence permits promotion into a sidecar candidate lane.

Use the broader .codex/skills/speech-model-finetuning-on-hemma/SKILL.md for model-agnostic training guidance and .codex/skills/sir-convert-a-lot-colab-hemma/SKILL.md for Colab/Hemma transfer.

Read Order

  1. references/architecture-and-experiment-contract.md
  2. docs/backlog/epics/epic-sircon-05-qwen3-tts-swedish-language-expansion-fine-tuning-on-hemma-and-colab.md
  3. docs/reference/ref-sircon-research-qwen-training-eval-pilot-progress-ledger-2026-03-15-qwen-training-eval-pilot-progress-ledger-2026-03-15.md
  4. docs/runbooks/run-sircon-qwen3-tts-swedish-finetuning-runbook-for-hemma-and-colab-qwen3-tts-swedish-finetuning-runbook-for-hemma-and-colab.md
  5. docs/decisions/adr-sircon-0005-hemma-sidecar-tts-architecture-and-non-pdf-gpu-governance.md
  6. docs/decisions/adr-sircon-0006-reusable-multi-backend-tts-sidecar-capability-contract.md

The reference owns agent-facing architecture and evidence rules. The runbook owns operator procedure. The progress ledger owns current experiment classes, state vectors, surface status, results, and next-step truth.

Classify First

  • benchmark: serving and runtime evidence only.
  • single-speaker adaptation: voice-transfer experiments, not general Swedish.
  • language expansion: multi-speaker Swedish support.

Choose language expansion when the requested outcome is general Swedish support.

Workflow

  1. Confirm the governing backlog slice and the current ledger state.
  2. Classify the experiment and state its single primary question.
  3. Check the architecture and evidence contract before changing code or run shape.
  4. Use the Qwen runbook for commands, host procedure, stop conditions, and recovery.
  5. Record results and state-vector changes in the progress ledger.
  6. Promote only through the documented evidence ladder.

Do not copy live operator status, command transcripts, or experiment results into this skill.

When not to use it

  • The task is normal sidecar benchmarking not involving model training
  • The task involves Chatterbox, F5, OpenVoice, or MMS implementation work
  • The task is a generic speech-model training question without Qwen-specific decisions

Limitations

  • The skill is model-specific to Qwen3-TTS
  • The skill is focused on Swedish language expansion
  • The skill is limited to Hemma and Colab H100 for training

How it compares

This skill provides a specialized workflow for Qwen TTS fine-tuning with a focus on Swedish language expansion and specific hardware considerations, unlike general speech model training.

Compared to similar skills

sir-convert-a-lot-qwen-finetuning side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
sir-convert-a-lot-qwen-finetuning (this skill)02moNo flagsAdvanced
robotics-code-generator148moNo flagsAdvanced
modal58moReviewIntermediate
hugging-face-cli36moReviewIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

robotics-code-generator

HumaizaNaz

Generates clean, runnable ROS 2, Gazebo, Isaac Sim, and VLA code for humanoid robotics

1490

modal

davila7

Run Python code in the cloud with serverless containers, GPUs, and autoscaling. Use when deploying ML models, running batch processing jobs, scheduling compute-intensive tasks, or serving APIs that require GPU acceleration or dynamic scaling.

587

hugging-face-cli

patchy631

Execute Hugging Face Hub operations using the `hf` CLI. Use when the user needs to download models/datasets/spaces, upload files to Hub repositories, create repos, manage local cache, or run compute jobs on HF infrastructure. Covers authentication, file transfers, repository creation, cache operations, and cloud compute.

350

computer-use-agents

davila7

Build AI agents that interact with computers like humans do - viewing screens, moving cursors, clicking buttons, and typing text. Covers Anthropic's Computer Use, OpenAI's Operator/CUA, and open-source alternatives. Critical focus on sandboxing, security, and handling the unique challenges of vision-based control. Use when: computer use, desktop automation agent, screen control AI, vision-based agent, GUI automation.

1040

machine-learning-ops-ml-pipeline

sickn33

Design and implement a complete ML pipeline for: $ARGUMENTS

436

ray-train

davila7

Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.

336

Search skills

Search the agent skills registry