Skip to content

docs(benchmarks,features): DFlash rows catch up with D14 (drift repair) - #43

Merged
mudler merged 1 commit into
mainfrom
docs/dflash-drift-repair
Aug 5, 2026
Merged

docs(benchmarks,features): DFlash rows catch up with D14 (drift repair)#43
mudler merged 1 commit into
mainfrom
docs/dflash-drift-repair

Conversation

@localai-bot

Copy link
Copy Markdown
Collaborator

STATUS.md has recorded SPEC-DFLASH DONE since D14 (2026-07-27): warp-scoped draft attention (242.9 → 77.9 ms), c1 our-on 29.32 vs vLLM-on 29.24 tok/s, non-overlapping 3-rep bands, 1.003x; spec-off 10.16 tok/s.

docs/BENCHMARKS.md (speculative table + open-gaps row) and docs/FEATURES.md still said ~2x over spec-off, below vLLM throughput / ◐. This updates the keyed rows in place per the public-document obligations. No engine change, no new measurement; the binding numbers are D14's (.agents/benchmark-record.md, CLAIM-DFLASH-D14, benchmark_binding=true).

Preflight: all gates green including doc-checkpoint --staged and now-current --staged.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UtefFpjk4XA11jpwEdQ6Cy

docs/STATUS.md has recorded SPEC-DFLASH DONE since D14 (2026-07-27:
warp-scoped draft attention, c1 our-on 29.32 vs vLLM-on 29.24 tok/s,
non-overlapping 3-rep bands, 1.003x; spec-off 10.16). BENCHMARKS.md and
FEATURES.md still said 'below vLLM throughput' / partial. Update the
keyed rows in place. No engine change, no new measurement.

FOLLOWING_AGENTS_PROTOCOL
Assisted-by: Claude Code:claude-fable-5 [ClaudeCode]
Claude-Session: https://claude.ai/code/session_01UtefFpjk4XA11jpwEdQ6Cy
@mudler
mudler merged commit 1153927 into main Aug 5, 2026
8 of 9 checks passed
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.

2 participants