Update README typo#9
Merged
0hq merged 1 commit intoopenai:mainfrom Mar 18, 2026
Merged
Conversation
kxddry
pushed a commit
to kxddry/parameter-golf
that referenced
this pull request
Mar 19, 2026
Update README typo
taka6745
pushed a commit
to taka6745/parameter-golf
that referenced
this pull request
Apr 7, 2026
…g for Muon optimizer From PR openai#1440 + arxiv:2603.09697 "Mousse: Rectifying the Geometry of Muon with Curvature-Aware Preconditioning" (Feb 2026). Inserts ~5 lines of diagonal preconditioning before zeropower_via_newtonschulz5 in the Muon optimizer step. Normalizes momentum gradient by row/col norms before spectral orthogonalization, trace-normalizing the matrix: G_pre = G / (||row||_2 * ||col||_2) Gated by USE_MOUSSE=1, falls back to vanilla Muon when unset. Idempotent via MOUSSE_MARKER. Anchored on the unique zeropower call which is invariant under all existing 22 patches. This is the FIRST shippable finding in 5 research fires that fits our train_loss metric (optimizer-side change affects training directly, unlike EMA/Tilt/GPTQ which only affect eval). Subagent recommended PASS due to medium effort estimate; overrode after confirming PR openai#1440 ships only the SIMPLIFIED diagonal preconditioning version (5 LOC, not 50-80). 4 MS experiments queued for validation: MS0_mousse_alone, MS1_mousse_plus_leaky_ng, MS2_mousse_seed42, MS3_mousse_plus_engram Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
taka6745
pushed a commit
to taka6745/parameter-golf
that referenced
this pull request
Apr 7, 2026
… confirmation), MR2 promising, PR openai#1430 MERGED at 0.39642 BPB Subagent reports PR openai#1430 (Per-Sample SLOT + Causal Backoff N-gram Mixer + TTT) has been MERGED at claimed 0.39642 BPB — 65% below public SOTA. If real, this fundamentally changes the competitive landscape. Audit fires openai#1-3 all flagged this PR as likely illegal under issue openai#677. Now MERGED. NEXT RESEARCH FIRE PRIORITY: deep-dive PR openai#1430 to verify legality and extract implementation. If real, port it. If leak-based, document it. Patches 17 (Mousse) and 18 (MuonEq-R) confirmed as known PORTS, not novel-to-comp. They were always documented as ports in research fires openai#9 and openai#10. Patches 15/16/21 still uncontested in 120+ open + 10 closed PRs (4 audits in a row). Pod healthy, ~$2.30/$36 spend. MR2_seed42 = 3.3004 (better than MS2 = 3.3358), suggesting MuonEq-R may slightly beat Mousse at L5 stack. Falsification of Patches 17 and 18 proceeding rapidly. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
taka6745
pushed a commit
to taka6745/parameter-golf
that referenced
this pull request
Apr 7, 2026
…util After 5 emergency interventions in 2 hours, the speed fix is finally working: GPU Memory: 744 MB -> 3370 MB (4.5x) GPU Util: 34% -> 100% (3x, FULLY MAXED) Power: 149W -> 218W Total compute/step: 270 GFLOP -> 17 TFLOP (64x) Total tokens/experiment: 1.5M -> 24M (16x) CHAMP_L5_seed42 currently running successfully: step:100 train_loss:3.6128 step_avg:861ms The actual root cause was Patch 22 EngramLite init anchor mismatch. The torch.compile crashes were a red herring — every experiment was crashing with AttributeError on self._engram_lite_enabled because the forward apply ran but the init didn't. getattr wrap fixed it. All prior "neutrality plateau" verdicts are now CONFIRMED INVALID: Mousse/MuonEq-R/NorMuon/Depth Recurrence/Coprime/EngramLite/QK_GAIN were all measured on 0.75% of intended data volume. Need re-validation. PR openai#1430 still OPEN, 24h no activity. Patches 15/16/20/21/25 still novel (9th consecutive audit confirmation). NEW finding: TMA Megakernel in 5 PRs (custom Triton kernel, hardware-side). We have ZERO hardware-side patches. Highest-leverage missing technique. Spend ~$6.33/$36 (17.6%). Far below $25 flag threshold. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag
pushed a commit
to gHashTag/parameter-golf
that referenced
this pull request
Apr 30, 2026
- Cargo workspace with 3 crates + bin/tri-railway - trios-railway-core: ProjectId/EnvironmentId/ServiceId/DeployId newtypes, RailwayHash::seal (R7 audit triplet), Client over Railway GraphQL v2 - trios-railway-audit: DriftCode D1..D7 + DriftEvent + verdict (Gate-2 PASS criterion), idempotent Neon DDL (railway_projects, railway_services, railway_audit_runs, railway_audit_events, v_railway_drift_open) - trios-railway-experience: append-only L7 writer to .trinity/experience/<YYYYMMDD>.trinity (L21-safe, no truncation) - bin/tri-railway: clap CLI with 'version', 'audit migrate-sql', 'experience append' (mutating verbs deferred to issues openai#4..openai#9) - LICENSE Apache-2.0, README, AGENTS.md, TASK.md per crate - CI: fmt --check + clippy -D warnings + build + test - Neon DDL applied to neondb (5 objects verified) - 16 unit tests passing (6 audit + 8 core + 2 experience) - ascii-only sources, R1 (no .sh / no Python in scripts/) Anchor: phi^2 + phi^-2 = 3. Closes #1 Agent: GENERAL
gHashTag
added a commit
to gHashTag/parameter-golf
that referenced
this pull request
Apr 30, 2026
openai#143) (openai#29) * feat(audit): tri-railway audit run + verdict CLI (closes openai#9, refs openai#143) Anchor: phi^2 + phi^-2 = 3. Adds the online-audit subcommands that close L-R14 (Gate-2 verdict) formally: tri-railway audit run --project <UUID> --target <BPB> [--ledger PATH] [--json] tri-railway audit verdict --ledger PATH --target <BPB> Behaviour: - audit run lists Railway services for a project (Q::project_view), converts them to RealService (with seed parsed from name like 'trios-train-seed-43' or 'igla-final-seed-44'), optionally loads a JSONL ledger, calls trios_railway_audit::detect to produce the full D1..D7 drift event set, runs verdict() to compute Gate2Pass / NotYet / Drift, prints a text summary, optionally JSON, then seals one R7 audit triplet to .trinity/experience via the existing experience writer. Exit codes: 0 = GATE-2 PASS (>= 3 services with bpb < target, no error drift) 1 = DRIFT (any error-severity event) 2 = NOT YET (no errors, target not yet met) - audit verdict is the offline form for cron/CI: takes a JSONL ledger snapshot already serialized from Neon, computes the same verdict against synthetic services, prints one line, exits with the same codes. Auth fix in trios-railway-core: RAILWAY_TOKEN_AUTH env var allows forcing 'team' (Bearer) vs 'project' (Project-Access-Token) when the UUID-shape heuristic guesses wrong. Personal API tokens are also UUID-shaped but require Bearer; without this override, authenticating to backboard.railway.com returned 'Not Authorized'. Verified with both curl variants against the live IGLA project. R5-honest verification (logs in PR body): cargo build --bin tri-railway --locked : OK cargo test --workspace --locked : 22 passed, 0 failed cargo clippy -D warnings : 0 Live smoke against IGLA (e4fe33bb-...): 18 services, 16 D1_ORPHAN warnings, NOT YET, exit=2, R7 triplet sealed at /tmp/audit-smoke/.trinity/experience/<date>.trinity Synthetic ledger smoke: 3 seeds bpb<1.85 -> GATE-2 PASS, exit=0 3 seeds bpb<1.85 vs target=1.50 -> NOT YET, exit=2 1 seed bpb=3.5e38 -> DRIFT (D5_OVERFLOW), exit=1 Closes openai#9. Refs openai#143 (IGLA RACE Gate-2 / L-R14). * style(audit): cargo fmt --all (CI format-check fix) --------- Co-authored-by: Perplexity Computer <computer@perplexity.ai>
jzmyres
pushed a commit
to jzmyres/parameter-golf
that referenced
this pull request
May 3, 2026
Triggered by grad_norm=0.07 mid-flight obs in iter 112+122 (well below
grad_clip=1.0 → headroom for deeper backward gradient signal). Backward
coverage at K=16 increases 12.5%→19%.
Iter 85 (H63) PROMOTED stochastic {2,3,4} earlier but at WD=0.30 +
K-jitter (4,6,10) era — different regime. This is fixed=3 retest under
WD=0.01 + K-jitter (16,24) + iter 112+122 (gram=0.1 + softcap=30) baseline.
Expected ~+10% wallclock cost from extra TBPTT step.
Also: Tier 1 reordering (user directive 2026-05-02) — iter 108 (deq_k 16→10)
deprioritized to END of Tier 1 (was launched 04:08, killed at ~step 10,
commit fa22310 reverted). Throughput-only iter with no expected val_bpb
improvement; lower priority than val_bpb-improving + simplification iters.
Files:
train_gpt.py: Hyperparameters.deq_bptt_k 2 → 3
CLAUDE.md §5: architecture table mirror
experiments/hypotheses.md: Tier 1 reordered, iter 108 moved to position openai#9
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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
2 participants
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.
Think NanoGPT Speedrun is optimizing time, not loss