Repository navigation

feat: ForgeRL v2 M0–M1 vector-batch runtime and formal acceptance - #1

Draft

MrLPF wants to merge 74 commits into

masterfrom

feat/forge-rl-v2-bootstrap

Draft

feat: ForgeRL v2 M0–M1 vector-batch runtime and formal acceptance#1
MrLPF wants to merge 74 commits into
masterfrom
feat/forge-rl-v2-bootstrap

Commits on Aug 21, 2026
- feat(forge-rl): bootstrap v2 distributed runtime
  MrLPF
  committed
  ae9275d
- feat(forge-rl): implement M1 local high-throughput runtime
  MrLPF
  committed
  36bf6de
- feat(forge-rl): add M1 acceptance harness
  MrLPF
  committed
  373b40e
- feat(forge-rl): add controlled M1 PPO learning gate
  MrLPF
  committed
  be84cce
- perf(forge-rl): optimize M1 node-local inference hot path
  MrLPF
  committed
  ce94906
- fix(forge-rl): restore UTF-8 README encoding
  MrLPF
  committed
  72a9d1a
- feat(forge-rl): add event-driven M1 descriptor collection
  MrLPF
  committed
  f58d8b0
- fix(forge-rl): preserve optimized inference constructor compatibility
  MrLPF
  committed
  fe81353
- perf(forge-rl): add persistent synchronous inference mailbox
  MrLPF
  committed
  a1a3b35
- perf(forge-rl): prebind mailbox views and align batch release
  MrLPF
  committed
  98cc31e
- fix(forge-rl): correct fast mailbox acceptance schema
  MrLPF
  committed
  d8c6e7b
- perf(forge-rl): greedily drain ready mailbox requests
  MrLPF
  committed
  d4432bf
- perf(forge-rl): isolate mailbox predictor in a dedicated process
  MrLPF
  committed
  7d16b4b
Commits on Aug 22, 2026
- perf(forge-rl): add actor-local CPU inference fast path
  MrLPF
  committed
  327ec2e
- perf(forge-rl): batch threaded actors without IPC
  MrLPF
  committed
  f4abe4b
- perf(forge-rl): add rendezvous-aware actor wave batching
  MrLPF
  committed
  6d180ad
- perf(forge-rl): make rendezvous batching the M1 in-process default
  MrLPF
  committed
  911e917
- test(forge-rl): verify staggered Actor rendezvous and timeout fallback
  MrLPF
  committed
  9105d8a
- perf(forge-rl): coalesce synchronous Actor waves before worker dispatch
  MrLPF
  committed
  4427afa
- fix(forge-rl): keep each rendezvous wave atomic in the worker
  MrLPF
  committed
  62ad8f2
- perf(forge-rl): use bounded sliding arrival windows for Actor batching
  MrLPF
  committed
  0448676
- perf(forge-rl): tighten rendezvous hard wait to the measured 1 ms budget
  MrLPF
  committed
  87d5faf
- perf(forge-rl): restore empirically better 2 ms hard batching budget
  MrLPF
  committed
  b39681b
- bench(forge-rl): add work-conserving double-buffer screening
  MrLPF
  committed
  664c3b7
- ci(forge-rl): run the focused M1 double-buffer screening
  MrLPF
  committed
  b4766e3
- perf(forge-rl): add direct vector-batch inference path
  MrLPF
  committed
  90c009c
- feat(forge-rl): export direct vector-batch inference runtime
  MrLPF
  committed
  10f5a60
- bench(forge-rl): add direct vector-batch M1 acceptance subject
  MrLPF
  committed
  663a0a7
- bench(forge-rl): export vector-batch acceptance subject
  MrLPF
  committed
  a64ee81
- bench(forge-rl): expose vector-batch M1 screening mode
  MrLPF
  committed
  f047d8d
- bench(forge-rl): support direct vector-batch double-buffer screening
  MrLPF
  committed
  8c23b56
- test(forge-rl): verify direct vector-batch inference contracts
  MrLPF
  committed
  f15ea14
- ci(forge-rl): screen direct vector-batch sampling path
  MrLPF
  committed
  0265198
- ci(forge-rl): repeat vector-batch screening for stability
  MrLPF
  committed
  1f1ac67
- feat(forge-rl): use vector-batch runtime for M1 learning gate
  MrLPF
  committed
  0c7741b
- feat(forge-rl): align M1 learning with vector-batch subject
  MrLPF
  committed
  e8b2ae0
- test(forge-rl): use FP32-appropriate mailbox tolerances
  MrLPF
  committed
  32d6674
- test(forge-rl): close the C++ vector-env M1 training loop
  MrLPF
  committed
  76a5607
- ci(forge-rl): make vector-batch the complete M1 formal subject
  MrLPF
  committed
  3e51953
- docs(forge-rl): finish the M1 vector-batch implementation checklist
  MrLPF
  committed
  94558fa
- docs(forge-rl): define one vector-batch M1 acceptance subject
  MrLPF
  committed
  d400c4c
- docs(forge-rl): make vector-batch the local production data path
  MrLPF
  committed
  ff71f01
Commits on Aug 23, 2026
- feat(forge-rl): preserve coordinator generations across lease expiry
  MrLPF
  committed
  cce9a69
- feat(forge-rl): add networked coordinator service and client
  MrLPF
  committed
  62891e1
- feat(forge-rl): export M2 network coordinator contracts
  MrLPF
  committed
  e08ce29
- test(forge-rl): validate M2 network coordinator fencing and discovery
  MrLPF
  committed
  ffea6ed
- feat(forge-rl): add M2 coordinator service launcher
  MrLPF
  committed
  523a3a5
- docs(forge-rl): define M2 network coordinator contract
  MrLPF
  committed
  ab48b90
- docs(forge-rl): enter M2 with M1 formal validation pending locally
  MrLPF
  committed
  eb55b58
- feat(forge-rl): add direct M2 experience and policy channels
  MrLPF
  committed
  b902ac0
- feat(forge-rl): export M2 direct data channels
  MrLPF
  committed
  805878d
- test(forge-rl): validate M2 direct experience and policy channels
  MrLPF
  committed
  e044534
- docs(forge-rl): specify M2 direct data channels
  MrLPF
  committed
  bb3520c
- docs(forge-rl): mark M2 direct channels complete
  MrLPF
  committed
  d2c68da
- fix(forge-rl): encode scalar policy tensors safely
  MrLPF
  committed
  25e6958
- feat(forge-rl): add topology-independent distributed checkpoints
  MrLPF
  committed
  521bdc8
- test(forge-rl): verify topology-independent checkpoint restore
  MrLPF
  committed
  1784468
- feat(forge-rl): export distributed checkpoint manager
  MrLPF
  committed
  5bfabf1
- docs(forge-rl): specify topology-independent checkpoint contract
  MrLPF
  committed
  7c360f9
- docs(forge-rl): mark topology-independent checkpoint complete
  MrLPF
  committed
  0295d1f
- feat(forge-rl): add restartable process supervisor for M2 recovery
  MrLPF
  committed
  c67ed98
- test(forge-rl): inject actor and inference replica failures
  MrLPF
  committed
  d590457
- test(forge-rl): restart failed learner group from distributed checkpoint
  MrLPF
  committed
  12f9a79
- feat(forge-rl): export M2 failure recovery supervisor
  MrLPF
  committed
  9433a25
- docs(forge-rl): document M2 component restart recovery
  MrLPF
  committed
  3cbe4b3
- docs(forge-rl): mark M2 failure injection complete
  MrLPF
  committed
  6c0daa7
Commits on Aug 24, 2026
- bench(forge-rl): add M2 weak-scaling acceptance harness
  MrLPF
  committed
  d8a9b9e
- bench(forge-rl): export M2 acceptance utilities
  MrLPF
  committed
  9faeb11
- bench(forge-rl): add M2 distributed worker entrypoint
  MrLPF
  committed
  f196ef5
- bench(forge-rl): add M2 acceptance report entrypoint
  MrLPF
  committed
  40084fe
- test(forge-rl): cover M2 weak-scaling acceptance
  MrLPF
  committed
  b66b172
- ci(forge-rl): add formal M2 two-node acceptance workflow
  MrLPF
  committed
  15aa3cf
- docs(forge-rl): define formal M2 acceptance
  MrLPF
  committed
  7f1be19
- docs(forge-rl): add M2 formal acceptance gate
  MrLPF
  committed
  1630775

Commits

Commits on Aug 21, 2026

Commits on Aug 22, 2026

Commits on Aug 23, 2026

Commits on Aug 24, 2026