chatterbox_defaults(): hardware-adaptive setup, 6GB jit validation#13
Open
TroyHernandez wants to merge 4 commits into
Open
chatterbox_defaults(): hardware-adaptive setup, 6GB jit validation#13TroyHernandez wants to merge 4 commits into
TroyHernandez wants to merge 4 commits into
Conversation
…esults - New chatterbox_defaults(vram_gb = NULL): detects the GPU (CPU when absent or under 4GB) and returns GC options, backend, max_new_tokens, and chunking threshold as a classed object; print method emits a ready-to-paste setup snippet. Measured tiers labeled vs projected. - 6GB validation results (GTX 1660 Ti, June 2026) folded into gc_options 6GB note, the performance vignette, and CLAUDE.md: jit 35-38 ms/token (4.7GB) vs container 30 - jit is fastest on every measured card; traced additionally truncates long-form at its 350-position cap. The old 'traced wins on 6GB' guidance predated the jit backend. - 12GB projected row added to the tier table.
…card threshold - Vignette no longer claims jit is un-revalidated on 6GB (it was, this PR). - 'jit wins/fastest' tightened to 'fastest native path/backend' - it does not beat the container. - Cards under 5GB are treated as CPU (model floor ~4.6GB, measured 6GB peak 4.7GB); measured flag no longer sweeps 4-5.5GB or 13GB into measured tiers.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes the hardware-matrix → adaptive-defaults arc with the fresh 6GB numbers.
New export:
chatterbox_defaults(vram_gb = NULL)— detects the GPU via nvidia-smi (CPU when absent or < 4GB, which can't hold the ~4.6GB model) and returns a classed list: GCoptions(apply withdo.call(options, ...)before torch loads),backend,max_new_tokens,chunk_chars, and ameasuredflag. The print method emits a ready-to-paste setup snippet per tier; projected tiers (8/12GB) are labeled as such.6GB validation (GTX 1660 Ti, long-form ~22s audio, GC 0.75):
jit is fastest on every measured card; the old "traced wins on 6GB" guidance predated the jit backend and is corrected in
chatterbox_gc_options()'s 6GB note, the performance vignette, and CLAUDE.md. 12GB projected row added to the tier table.Tier-logic tests run CPU-only (
test_defaults.R); full suite passes. Version 0.1.0.9 + NEWS as separate bump commit.