3 people 5b3793b85e feat(kv-calc): opt-in --kv-breakdown architecture cache planning layer (#213) ...

Additive + reporting-only. Default text and --json output are byte-identical
unless --kv-breakdown is passed; predict()/raw_verdict()/--calibration remain
the calibrated fit authority. Adds --gpus N (informational), a conflict-guarded
--sequences alias, and opt-in compressed-KV / indexer-cache / draft-KV estimate
buckets scoped to single-node 1-8 GPU home/workstation planning.

Implemented via Codex collab; independently validated on-rig by Claude:
- default output IDENTICAL (HEAD vs change) across text / --json / --solve-max-ctx
- --calibration 22/22, both kv-calc test scripts pass, py_compile clean, diff --check clean
- 4 error cases exit 2 (sequences/max-num-seqs conflict, --gpus bounds,
  indexer/compressed missing head-dim)

Co-authored-by: noonghunna <[email protected]>
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>

2026-05-24 08:57:05 +05:00

..

lib

docs+registry: surface ik-llama on the single-card front door + fix stale max_ctx

2026-05-23 23:13:49 +00:00

maintenance

…

tests

feat(kv-calc): opt-in --kv-breakdown architecture cache planning layer (#213)

2026-05-24 08:57:05 +05:00

bench.sh

feat(eval): expose request-level thinking toggles (#196)

2026-05-23 08:24:58 +05:00

detect_nvlink.sh

feat(nvlink): auto-detect NVLink on N-GPU topologies; add detection to multi4 + gemma-4-26b dual

2026-05-18 19:31:39 +00:00

diagnose-estate.sh

…

diagnose-profile.sh

…

generate-compose.sh

v0.8.0 STEP 3+4: compose generator + 5-triple golden-parity test (#141)

2026-05-16 15:49:37 +00:00

gpu-mode.sh

fix(#169): distinct default container_name per llama.cpp/ik single variant

2026-05-23 22:14:58 +00:00

health.sh

…

launch.sh

fix(#169): distinct default container_name per llama.cpp/ik single variant

2026-05-23 22:14:58 +00:00

power-cap-sweep.sh

…

preflight.sh

fix(soak,preflight): recognize llama-cpp / ik-llama containers in autodetect (#403)

2026-05-23 16:03:22 +00:00

pull.sh

feat(pull): v0.8.2 STEP V2 — surface pointer + --submit-last/--submit (gh + gh-less, consented, F5 reuse)

2026-05-18 17:37:41 +00:00

quality-test.sh

feat(eval): expose request-level thinking toggles (#196)

2026-05-23 08:24:58 +05:00

rebench-full.sh

fix(rebench): basename model id for the aider/litellm step (ik_llama full-path id → 0/30)

2026-05-23 18:55:25 +00:00

rebench-report.py

…

rebench-runtime.sh

feat(models): add ik_llama Qwen3.6-27B IQ4_KS composes — text 262K + vision 160K (#180)

2026-05-22 04:45:44 +05:00

report.sh

fix(#168): scope report.sh kv-calc calibration to the running model

2026-05-23 22:19:37 +00:00

setup.sh

feat(setup): WEIGHTS=gguf to fetch the llama.cpp GGUF (not just the vLLM model) (#191)

2026-05-23 02:38:15 +05:00

soak-helper.py

docs(soak-test): clarify PASS verdict semantics — closes #140

2026-05-15 22:18:09 +05:00

soak-test.sh

fix(soak,preflight): recognize llama-cpp / ik-llama containers in autodetect (#403)

2026-05-23 16:03:22 +00:00

submit-bench.sh

…

switch.sh

docs: correct single-card llama.cpp/ik_llama ctx 262K -> 200K (shipped default)

2026-05-23 21:55:35 +00:00

update.sh

…

verify-full.sh

Fix verify-full to accept reasoning_content

2026-05-22 19:15:26 +00:00

verify-stress.sh

fix(verify-stress): three live-caught bugs in ceiling ladder (#199)

2026-05-23 04:42:18 +00:00

verify.sh

…