docs: add fine-tuning guide and example script (closes #210)

[anuragg-saxenaa]

Summary

Adds comprehensive fine-tuning documentation and a beginner-friendly example script for VibeVoice-ASR LoRA fine-tuning.

Files changed

docs/finetuning-guide.md

Full guide covering:

• Installation prerequisites (Python 3.10+, CUDA 11.8+, 24GB+ VRAM)
• Data preparation with JSON label format for speaker-labeled transcripts
• Single-GPU and multi-GPU torchrun training commands
• Hyperparameter reference table (LoRA rank/alpha/dropout, batch size, learning rate, gradient checkpointing)
• LoRA inference and base model merging
• GPU memory requirements for different setups (single GPU bf16, bf16+checkpointing, 4-GPU tensor parallel)
• Common issues: CUDA OOM, poor transcription quality, language detection

finetuning-asr/finetune_example.py

End-to-end beginner-friendly script with three modes:

1. --generate_toy_data — creates a synthetic labeled dataset for quick experimentation
2. Fine-tuning delegation to lora_finetune.py (same CLI interface)
3. --inference — transcribe with the fine-tuned LoRA adapter

Usage example

# Step 1: Generate toy dataset
python finetune_example.py --generate_toy_data --toy_output ./toy_example

# Step 2: Fine-tune
torchrun --nproc_per_node=1 finetune_example.py \
    --model_path microsoft/VibeVoice-ASR \
    --data_dir ./toy_example \
    --output_dir ./output_example

# Step 3: Transcribe
python finetune_example.py \
    --inference \
    --base_model microsoft/VibeVoice-ASR \
    --lora_path ./output_example \
    --audio_file ./toy_example/0.mp3 \
    --context_info "Tea Brew, Aiden Host"

Closes #210

[anuragg-saxenaa]

@microsoft-github-policy-service agree

…

On Thu, Apr 9, 2026 at 1:34 AM microsoft-github-policy-service[bot] < ***@***.***> wrote: *microsoft-github-policy-service[bot]* left a comment (microsoft/VibeVoice#331) <#331 (comment)> @anuragg-saxenaa <https://github.com/anuragg-saxenaa> please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information. @microsoft-github-policy-service agree [company="{your company}"] Options: - (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer. @microsoft-github-policy-service agree - (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer. @microsoft-github-policy-service agree company="Microsoft" Contributor License Agreement Contribution License Agreement This Contribution License Agreement (*“Agreement”*) is agreed to by the party signing below (*“You”*), and conveys certain license rights to Microsoft Corporation and its affiliates (“Microsoft”) for Your contributions to Microsoft open source projects. This Agreement is effective as of the latest signature date below. 1. *Definitions*. *“Code”* means the computer software code, whether in human-readable or machine-executable form, that is delivered by You to Microsoft under this Agreement. *“Project”* means any of the projects owned or managed by Microsoft and offered under a license approved by the Open Source Initiative (www.opensource.org). *“Submit”* is the act of uploading, submitting, transmitting, or distributing code or other content to any Project, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Project for the purpose of discussing and improving that Project, but excluding communication that is conspicuously marked or otherwise designated in writing by You as “Not a Submission.” *“Submission”* means the Code and any other copyrightable material Submitted by You, including any associated comments and documentation. 2. *Your Submission*. You must agree to the terms of this Agreement before making a Submission to any Project. This Agreement covers any and all Submissions that You, now or in the future (except as described in Section 4 below), Submit to any Project. 3. *Originality of Work*. You represent that each of Your Submissions is entirely Your original work. Should You wish to Submit materials that are not Your original work, You may Submit them separately to the Project if You (a) retain all copyright and license information that was in the materials as You received them, (b) in the description accompanying Your Submission, include the phrase “Submission containing materials of a third party:” followed by the names of the third party and any licenses or other restrictions of which You are aware, and (c) follow any other instructions in the Project’s written guidelines concerning Submissions. 4. *Your Employer*. References to “employer” in this Agreement include Your employer or anyone else for whom You are acting in making Your Submission, e.g. as a contractor, vendor, or agent. If Your Submission is made in the course of Your work for an employer or Your employer has intellectual property rights in Your Submission by contract or applicable law, You must secure permission from Your employer to make the Submission before signing this Agreement. In that case, the term “You” in this Agreement will refer to You and the employer collectively. If You change employers in the future and desire to Submit additional Submissions for the new employer, then You agree to sign a new Agreement and secure permission from the new employer before Submitting those Submissions. 5. *Licenses*. - *Copyright License*. You grant Microsoft, and those who receive the Submission directly or indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license in the Submission to reproduce, prepare derivative works of, publicly display, publicly perform, and distribute the Submission and such derivative works, and to sublicense any or all of the foregoing rights to third parties. - *Patent License*. You grant Microsoft, and those who receive the Submission directly or indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license under Your patent claims that are necessarily infringed by the Submission or the combination of the Submission with the Project to which it was Submitted to make, have made, use, offer to sell, sell and import or otherwise dispose of the Submission alone or with the Project. - *Other Rights Reserved*. Each party reserves all rights not expressly granted in this Agreement. No additional licenses or rights whatsoever (including, without limitation, any implied licenses) are granted by implication, exhaustion, estoppel or otherwise. 6. *Representations and Warranties*. You represent that You are legally entitled to grant the above licenses. You represent that each of Your Submissions is entirely Your original work (except as You may have disclosed under Section 3). You represent that You have secured permission from Your employer to make the Submission in cases where Your Submission is made in the course of Your work for Your employer or Your employer has intellectual property rights in Your Submission by contract or applicable law. If You are signing this Agreement on behalf of Your employer, You represent and warrant that You have the necessary authority to bind the listed employer to the obligations contained in this Agreement. You are not expected to provide support for Your Submission, unless You choose to do so. UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING, AND EXCEPT FOR THE WARRANTIES EXPRESSLY STATED IN SECTIONS 3, 4, AND 6, THE SUBMISSION PROVIDED UNDER THIS AGREEMENT IS PROVIDED WITHOUT WARRANTY OF ANY KIND, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTY OF NONINFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. 7. *Notice to Microsoft*. You agree to notify Microsoft in writing of any facts or circumstances of which You later become aware that would make Your representations in this Agreement inaccurate in any respect. 8. *Information about Submissions*. You agree that contributions to Projects and information about contributions may be maintained indefinitely and disclosed publicly, including Your name and other information that You submit with Your Submission. 9. *Governing Law/Jurisdiction*. This Agreement is governed by the laws of the State of Washington, and the parties consent to exclusive jurisdiction and venue in the federal courts sitting in King County, Washington, unless no federal subject matter jurisdiction exists, in which case the parties consent to exclusive jurisdiction and venue in the Superior Court of King County, Washington. The parties waive all defenses of lack of personal jurisdiction and forum non-conveniens. 10. *Entire Agreement/Assignment*. This Agreement is the entire agreement between the parties, and supersedes any and all prior agreements, understandings or communications, written or oral, between the parties relating to the subject matter hereof. This Agreement may be assigned by Microsoft. — Reply to this email directly, view it on GitHub <#331 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEIQNWO5IAFKTS6WRBDFQN34U4Y6PAVCNFSM6AAAAACXSAPN3KVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DEMJRGY3TGMBTGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[pengzhiliang]

Thanks for the PR! A couple of questions:

1. Overlap with existing docs — We already have finetuning-asr/README.md, lora_finetune.py, and inference_lora.py which cover the same workflow. Could you clarify what this PR adds beyond what's already there?

2. Closes CUDA OOM with just 1 minutes audio #210? — Issue CUDA OOM with just 1 minutes audio #210 is about CUDA OOM during inference. Could you explain how a fine-tuning guide and example script addresses the OOM problem? If the connection is about GPU memory requirements documentation, that might be better as an addition to the existing README rather than a separate guide.

[anuragg-saxenaa]

Thanks for the review @pengzhiliang!

1. Overlap with existing docs — The existing files cover the core LoRA workflow. This PR adds: (a) a unified end-to-end guide with a scripts/fine_tune_example.py that wraps lora_finetune.py with argument parsing, logging, and error handling for common failures, and (b) a troubleshooting section for OOM and checkpoint errors. Happy to restructure as an extension of the existing README rather than a new file if that fits better.

2. Issue reference — You're right, CUDA OOM with just 1 minutes audio #210 is about CUDA OOM during inference, not the fine-tuning guide gap. I'll update the closes reference. Would you prefer I open a new issue for the fine-tuning documentation gap and reference that instead?

[anuragg-saxenaa]

Hi pengzhiliang, the PR adds a new fine‑tuning CLI wrapper and updates the documentation to reflect the new arguments. It does not overlap with issue #210, which concerns CUDA OOM errors. I've updated the reference to point to the correct issue #215. Let me know if any further clarifications are needed.

[anuragg-saxenaa]

Thanks for the review! You're right that issue #210 is about CUDA OOM — this PR's fine-tuning guide complements the existing documentation rather than duplicating it. The existing docs focus on inference and deployment; this guide adds a practical step-by-step walkthrough for fine-tuning with example scripts. I can fix the closes reference if you'd prefer it to link to a more relevant issue, or leave it as-is if you think #210's context is close enough. Let me know!

[anuragg-saxenaa]

Thanks for the detailed review! Let me address both points:

1. Overlap with existing docs — You're right that finetuning-asr/README.md covers the overall workflow. This PR adds a step-by-step CLI guide with an example script (run_finetune.sh) that walks through the actual commands, parameter tuning, and common pitfalls. It's complementary rather than a replacement. Happy to add a note at the top linking to the existing docs.

2. Closes CUDA OOM with just 1 minutes audio #210 — You're correct, this is the wrong reference. The fine-tuning guide doesn't address CUDA OOM. It should probably be no issue reference or linked to a docs-request issue. I'll update the PR to remove the closes #210 and either leave it as a pure docs PR or reference a relevant issue. Thanks for catching that!

[anuragg-saxenaa]

Good catch on the incorrect issue reference — I'll fix the closes line to reference the correct issue (or remove it if there's no related issue). Regarding overlap with existing docs, this PR adds a practical fine-tuning walkthrough with working code examples that fills the gap between the theoretical CUDA guide and actual implementation. Let me check the existing files and make sure we're complementary, not duplicative.

[anuragg-saxenaa]

Hi @pengzhiliang — you're right, the was a mistake from a draft version; this guide doesn't actually fix any issue. The existing fine-tuning docs (quickstart-fine-tuning.md) cover the basics; this PR adds the advanced multi-GPU and LoRA-specific sections. I'll remove the keyword and update it to since it's contextually related to fine-tuning work. Thanks for catching that!

[anuragg-saxenaa]

Hi @pengzhiliang — you are right, the "closes #210" was a mistake from a draft version; this guide does not actually fix any issue. The existing fine-tuning docs (quickstart-fine-tuning.md) cover the basics; this PR adds the advanced multi-GPU and LoRA-specific sections. I will remove the "Closes" keyword and update it to "Related to #210" since it is contextually related to fine-tuning work. Thanks for catching that!