Skip to content

docs(quickstart): document server-side hardware requirements (#123) - #133

Merged
BrettKinny merged 1 commit into
mainfrom
docs/server-hardware-123
May 31, 2026
Merged

BrettKinny merged 1 commit into
mainfrom
docs/server-hardware-123

Conversation

@BrettKinny

Copy link
Copy Markdown
Owner

Closes #123.

New users on a GPU-less host hit could not select device driver "nvidia" from the compose runtime: nvidia block, with nothing in the docs explaining what server hardware is actually required.

Change

Adds a Server hardware subsection to docs/quickstart.md that makes the requirements explicit:

  • The default stack (FunASR SenseVoice ASR + LocalPiper TTS + a cloud LLM key) is CPU-only — no GPU required. That's the Quickstart happy path.
  • A GPU is only needed for two opt-in scenarios, captured in a table:
    • WhisperLocal ASR (faster-whisper float16 — the sole reason for the runtime: nvidia block), and
    • self-hosting the LLM locally (VRAM scales with model size; links to the local cookbooks).
  • Clarifies that make setup auto-detects the NVIDIA runtime and strips the runtime: nvidia / NVIDIA_* blocks from the rendered docker-compose.yml when absent — so the GPU config is opt-in, not something users must edit.
  • Documents the manual removal steps (the # --- BEGIN/END CUDA BLOCK --- markers) for anyone rendering the compose file by hand — i.e. the exact workaround the reporter performed.

Notes

  • Docs-only; no behaviour change. The make setup CUDA-stripping logic it describes already exists (Makefile:132-182).
  • Both relative links (cookbook/run-fully-local.md, cookbook/llama-swap-concurrent-models.md) resolve; frontmatter intact for the Frontmatter Check job.

🤖 Generated with Claude Code

New users hit `could not select device driver "nvidia"` when bringing up
the compose stack on a GPU-less host, with no docs explaining what
hardware the server side actually needs.

Add a "Server hardware" subsection to the Quickstart making explicit
that the default stack (FunASR SenseVoice ASR + LocalPiper TTS) is
CPU-only, and that the `runtime: nvidia` block is needed solely for the
optional WhisperLocal ASR path. Document that `make setup` auto-detects
the NVIDIA runtime and strips the CUDA block when absent, plus the
manual removal steps for hand-rendered compose files.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 31, 2026 11:36
@BrettKinny
BrettKinny merged commit acafeb3 into main May 31, 2026
9 of 10 checks passed
@BrettKinny
BrettKinny deleted the docs/server-hardware-123 branch May 31, 2026 11:37

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add server-side hardware requirement

2 participants