Upstream evidence
Pin support to a commit, not a product name
Timestamp is UTC and records the upstream merge event.
An old binary can fail before memory allocation
A package built before the merge may report an unknown architecture or unsupported model type. That is a build-recognition problem. It is not evidence that the GGUF is too large, corrupted, or incompatible with your hardware.
Safe starting sequence
Verify provenance before assembling a command
- 01
Identify the exact build
Record the binary's source, build date, revision, and whether a wrapper ships its own embedded llama.cpp version.
- 02
Compare with the merged revision
Confirm that the build contains the Muse Glimmer merge commit or a later revision that retains the change.
- 03
Verify the official file manifest
Check the selected main GGUF and any intended Vision projector or DFlash drafter against the official tree.
- 04
Use current upstream documentation
Runtime arguments can change. Read the documentation shipped with the same build instead of transplanting flags from another model family.
Command status
Generic shape — placeholder only
llama-cli -m [MODEL_PATH] [RUNTIME_OPTIONS]
Why the copy action is disabled
This wiki has not verified an upstream Muse-specific, end-to-end recipe that covers the main file, image companion, and DFlash companion together. The placeholders above are intentionally incomplete and not an executable recipe. No vision or DFlash flags are guessed here.
Limitations
- A merge into master does not establish availability in a tagged release, package manager, desktop app, or wrapper.
- The upstream merge is architecture-support evidence, not a per-device compatibility or performance result.
- Arguments, companion integration, and cache behavior remain build-specific until documented and verified end to end.
- This page does not execute, upload, or copy commands.
Continue the runtime setup
- Estimate hardware fit
Screen memory capacity before attempting a runtime setup.
- Verify official GGUF artifacts
Match the main model with the intended companion files.
- Trace a loading failure
Distinguish architecture recognition from memory and file problems.
- Plan runtime memory
Understand file, companion, KV-cache, and reserve boundaries.