Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Apr 16, 2026

• Fix: geometry disk cache skipped due to phase_cache_valid=0 on first run

  Show description for 4e88785
  NagusameCS

  committed
  4e88785

  Copy full SHA for 4e88785

• OTT Todo27: context-conditioned HS, KV snapshot, dynamic spec batch, remove per-token noise

  NagusameCS

  committed
  ed012c7

  Copy full SHA for ed012c7

• Fix rollout vel integration, skip tok_prev fwd pass, add per-request TPS timing

  NagusameCS

  committed
  e634fc7

  Copy full SHA for e634fc7

• OTT: trajectory-coherent multi-step geodesic rollout + vocab index in v2/rollout paths

  NagusameCS

  committed
  504e1af

  Copy full SHA for 504e1af

• OTT: full-vocab PCA nearest-neighbour index (ott_vocab_idx.bin)

  NagusameCS

  committed
  9a77c94

  Copy full SHA for 9a77c94

• OTT: GRC disk persistence + per-QKV head pullback metric refinement

  NagusameCS

  committed
  082ad8c

  Copy full SHA for 082ad8c

• OTT: GRC online feedback + Phase-3 geometry disk persistence (ott_geometry.bin)

  NagusameCS

  committed
  152a159

  Copy full SHA for 152a159

• OTT Speculative Decode: geodesic drafts + transformer verification (--ott-speculative)

  NagusameCS

  committed
  4a4f28f

  Copy full SHA for 4a4f28f

• GRC perfect: curvature-based rho + bucket ANN + AttnRes block summaries

  NagusameCS

  committed
  38286dd

  Copy full SHA for 38286dd

• GRC: implement Jacobi propagator + geodesic record library (OTT paper §4)

  Show description for 44bf412
  NagusameCS

  committed
  44bf412

  Copy full SHA for 44bf412

• OTT k4 Step 3 quality: raise confidence threshold + add piece quality gate

  Show description for 17188a2
  NagusameCS

  committed
  17188a2

  Copy full SHA for 17188a2

• OTT k4 Step 3: axiom_beta_geodesic_step_fast() wired into local hot path

  Show description for d8ecd56
  NagusameCS

  committed
  d8ecd56

  Copy full SHA for d8ecd56

• OTT k^4 step 1+4: pullback metric + RMSNorm connection correction

  Show description for 2644a65
  NagusameCS

  committed
  2644a65

  Copy full SHA for 2644a65

• perf: daemon mode, CUDA graph fix, 2048 ctx cap, adaptive GEMV blocks

  Show description for 054fd04
  NagusameCS

  committed
  054fd04

  Copy full SHA for 054fd04

Commits on Apr 14, 2026

• Axiom Beta: reuse phase5 probe pool and patch target row for faster scoring

  NagusameCS

  committed
  05b8ee7

  Copy full SHA for 05b8ee7

• Axiom Beta: add CUDA-backed phase5 scorer and GPU CLI controls

  NagusameCS

  committed
  4b1bc2c

  Copy full SHA for 4b1bc2c

• Axiom Beta: phase5 token metrics, probe CLI, and geodesic roadmap upgrades

  NagusameCS

  committed
  a32e0b4

  Copy full SHA for a32e0b4

• feat(beta): add autonomous axiomatic manifold subsystem

  NagusameCS

  committed
  96d2581

  Copy full SHA for 96d2581

• docs: rewrite README performance section with measured benchmarks

  NagusameCS

  committed
  3f7f201

  Copy full SHA for 3f7f201

• chore: harden release hygiene and internal-doc exclusions

  NagusameCS

  committed
  c2cb2e5

  Copy full SHA for c2cb2e5

• checkpoint: host runtime expansion and perf-path integration

  NagusameCS

  committed
  207bb53

  Copy full SHA for 207bb53

Commits on Apr 12, 2026

• P8: GPU GEMV offload + HyperTensor API rebrand

  Show description for 7682bc3
  NagusameCS

  committed
  7682bc3

  Copy full SHA for 7682bc3

• fix: CUDA kernel warnings

  NagusameCS

  committed
  f6dce3a

  Copy full SHA for f6dce3a

• P3: CUDA backend + BLT tokenizer + build updates

  Show description for d46579b
  NagusameCS

  committed
  d46579b

  Copy full SHA for d46579b

• P1-P7: Backend vtable, metadata normalization, tensor bridge, HTTP API server

  Show description for 35bc957
  NagusameCS

  committed
  35bc957

  Copy full SHA for 35bc957

Commits on Apr 10, 2026

• fix: first working inference on Windows host

  Show description for bc11085
  NagusameCS

  committed
  bc11085

  Copy full SHA for bc11085

• feat: host-mode runtime — HyperTensor v0.4.0 Axon

  Show description for 7358b69
  NagusameCS

  committed
  7358b69

  Copy full SHA for 7358b69

• docs: update all documentation for v0.3.0 SMP+JIT changes

  Show description for 1ab337d
  NagusameCS

  committed
  1ab337d

  Copy full SHA for 1ab337d

Commits on Mar 25, 2026

• perf: SMP + JIT optimizations, 454 -> 162 ms/tok (2.8x speedup)

  Show description for f5ba534
  NagusameCS

  committed
  f5ba534

  Copy full SHA for f5ba534

Commits on Mar 23, 2026

• Pass 11: Coherent LLM inference — fix Q4_0 dequant layout, RMSNorm epsilon, LongRoPE

  Show description for 6e6379e
  NagusameCS

  committed
  6e6379e

  Copy full SHA for 6e6379e

Commits on Mar 20, 2026

• Pass 10: AVX2 JIT GEMV kernel, VEX prefix encoding, fused Q8_0 inference

  Show description for e5e9e60
  NagusameCS

  committed
  e5e9e60

  Copy full SHA for e5e9e60

Commits on Mar 19, 2026

• Pass 9: Dynamic allocation — remove all compile-time model size caps

  Show description for 51a66b2
  NagusameCS

  committed
  51a66b2

  Copy full SHA for 51a66b2

Commits on Mar 18, 2026

• Pass 8: AVX2+FMA inference acceleration, larger model support

  Show description for 531970a
  NagusameCS

  committed
  531970a

  Copy full SHA for 531970a

Commits on Mar 17, 2026

• Pass 7: kernel/user separation, SMP MADT, E1000/AHCI drivers, TensorFS free-list, dynamic pools

  Show description for 69082b4
  NagusameCS

  committed
  69082b4

  Copy full SHA for 69082b4

• Pass 6: production features — PBKDF2, TCP retransmit, W^X VMM, TLS 1.3, batch inference queue

  Show description for ad6a181
  NagusameCS

  committed
  ad6a181

  Copy full SHA for ad6a181