Repository navigation

feat: server-reported token counts for ISL/OSL/TPS/TPOT (#540) - #552

Merged

arekay-nv merged 27 commits into

mainfrom

feat/server-token-counts

Nov 14, 2026

Merged

feat: server-reported token counts for ISL/OSL/TPS/TPOT (#540)#552
arekay-nv merged 27 commits into
mainfrom
feat/server-token-counts

Commits on Nov 10, 2026
- test: pin an archived pre-usage run against post-run consumers
  
  arekay-nv
  and
  claude
  committed
  1f1c175
- feat(core): token-count basis type, naming rule and server-usage helpers
  
  arekay-nv
  and
  claude
  committed
  8890138
- feat(config): token_count_source and stream_include_usage settings
  
  arekay-nv
  and
  claude
  committed
  db3bfd7
- feat(endpoint_client): request and capture server token usage
  
  arekay-nv
  and
  claude
  committed
  5848596
- test(testing): stub servers report token usage when the request asks
  
  arekay-nv
  and
  claude
  committed
  90e7a6f
- feat(metrics): record server token counts beside the tokenizer's
  
  arekay-nv
  and
  claude
  committed
  260d975
- feat(report): report the other token basis and gate completeness on the primary
  
  arekay-nv
  and
  claude
  committed
  bb4e2b2
- feat(benchmark): post-run tools count tokens on the run's basis
  
  arekay-nv
  and
  claude
  committed
  78d7b8b
- docs: server-reported token counts, the naming rule and the usage rule
  
  arekay-nv
  and
  claude
  committed
  6d5229e
- test(perf): roofline option to opt out of stream usage
  
  arekay-nv
  and
  claude
  committed
  c68ad96
- fix(metrics): server TPOT records zero windows, as the client basis does
  
  arekay-nv
  and
  claude
  committed
  4b7153b
- fix(accuracy): mark the server OSL block partial on either primary
  
  arekay-nv
  and
  claude
  committed
  4c1799e
- docs: say stream_include_usage changes the request body on either basis
  
  arekay-nv
  and
  claude
  committed
  b0fd42c
- perf(worker): read stream_include_usage once per worker
  
  arekay-nv
  and
  claude
  committed
  2f9eac8
- fix(session): drop a non-count left under the usage metadata keys
  
  arekay-nv
  and
  claude
  committed
  7e09fe9
- fix(report): withhold TPS and interactivity when the primary basis fell short
  
  arekay-nv
  and
  claude
  committed
  f48c068
- fix(report): name tokenization only when it is pending on a client primary
  
  arekay-nv
  and
  claude
  committed
  23882ca
- docs(load_generator): pass token_count_source to Report.from_snapshot
  
  arekay-nv
  and
  claude
  committed
  e4f0e60
- fix(scripts): read --summary before the early-stopping replay
  
  arekay-nv
  and
  claude
  committed
  2490917
- refactor(token-basis): keyword-only counts in missing_usage_message, and say why replay_basis fell back
  
  arekay-nv
  and
  claude
  committed
  eb74650
- docs(steady-state): explain _is_complete's skip of an undecodable line
  
  arekay-nv
  and
  claude
  committed
  e4e4920
- fix(full-run-osl): count an unexpected payload as server missing usage
  
  arekay-nv
  and
  claude
  committed
  e61d68e
- fix(openai): treat numpy scalars as n in requests_multiple_choices
  
  arekay-nv
  and
  claude
  committed
  8e8ff3c
- refactor(scoring): drop get_raw_outputs' unused wanted_uuids
  
  arekay-nv
  and
  claude
  committed
  d82db67
- docs: client is the default basis; accuracy warnings name the client blocks
  
  arekay-nv
  and
  claude
  committed
  704186e
- test(perf): benchmark the streaming encode path that requests usage
  
  arekay-nv
  and
  claude
  committed
  c805b32
Commits on Nov 14, 2026
- Merge branch 'main' into feat/server-token-counts
  arekay-nv
  authored
  1b180f0

Commits

Commits on Nov 10, 2026

Commits on Nov 14, 2026