Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Apr 16, 2026

• perf: daemon TTFT, CUDA graph fix, GRC+OTT plan, benchmark analysis

  Show description for 1c63d38
  NagusameCS

  committed
  1c63d38

  Copy full SHA for 1c63d38

Commits on Apr 14, 2026

• feat: advance token-native runtime and tooling stack

  NagusameCS

  committed
  1e567b9

  Copy full SHA for 1e567b9

Commits on Apr 12, 2026

• P8: GPU GEMV offload — selective CUDA dispatch for large matrices

  Show description for 25accf0
  NagusameCS

  committed
  25accf0

  Copy full SHA for 25accf0

• fix: CUDA kernel warnings (use NEG_INF, remove dup define)

  NagusameCS

  committed
  d5ebd93

  Copy full SHA for d5ebd93

• P3: CUDA backend with real GPU kernels + BLT integration

  Show description for f4859f2
  NagusameCS

  committed
  f4859f2

  Copy full SHA for f4859f2

• P5/P6: Tensor bridge + BLT tokenizer

  Show description for a32323b
  NagusameCS

  committed
  a32323b

  Copy full SHA for a32323b

• P1/P2: Backend vtable + metadata normalization

  Show description for 23d71d0
  NagusameCS

  committed
  23d71d0

  Copy full SHA for 23d71d0

• perf: re-enable Q4xQ8 integer GEMV path - fix JIT struct offset bug

  Show description for fa986ea
  NagusameCS

  committed
  fa986ea

  Copy full SHA for fa986ea

• tools: emit normalized manifest for downloaded models

  NagusameCS

  committed
  eb5ebc0

  Copy full SHA for eb5ebc0

• perf: add host correctness-speed benchmark harness

  NagusameCS

  committed
  ee45c24

  Copy full SHA for ee45c24

• llm: chat streaming, backend scaffolding, universal model intake

  NagusameCS

  committed
  9d6ced7

  Copy full SHA for 9d6ced7

Commits on Apr 10, 2026

• 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

• harden: Pass 5 — 5 fixes across 3 files

  Show description for 561ece1
  NagusameCS

  committed
  561ece1

  Copy full SHA for 561ece1

• harden: Pass 4 — 17 fixes across 11 files (final production sweep)

  Show description for fa75efd
  NagusameCS

  committed
  fa75efd

  Copy full SHA for fa75efd

• braniac: harden header + impl for production

  Show description for 6df310b
  NagusameCS

  committed
  6df310b

  Copy full SHA for 6df310b

• Pass 3: final production hardening — 15 fixes across 12 files

  Show description for 3552fca
  NagusameCS

  committed
  3552fca

  Copy full SHA for 3552fca

• hardening pass 2: 15 fixes across CRITICAL/HIGH/MEDIUM/LOW

  Show description for e6bc644
  NagusameCS

  committed
  e6bc644

  Copy full SHA for e6bc644

• production hardening: security, bounds checking, auth, selftests

  Show description for 4b63ba3
  NagusameCS

  committed
  4b63ba3

  Copy full SHA for 4b63ba3

• feat: Braniac brain-inspired predictive coding module with JIT

  Show description for 5485780
  NagusameCS

  committed
  5485780

  Copy full SHA for 5485780

Commits on Mar 16, 2026

• Add SSH-2 server, crypto library, and unified security framework

  Show description for ff9eca6
  NagusameCS

  committed
  ff9eca6

  Copy full SHA for ff9eca6

Commits on Mar 3, 2026

• perf: SSE2 SIMD hot paths, JIT Q8_0 GEMV, top-k/p sampling, hash tokenizer, multi-turn cache

  Show description for ee9cb4c
  NagusameCS

  committed
  ee9cb4c

  Copy full SHA for ee9cb4c

• ARM64: HDMI framebuffer driver for RPi4 VideoCore mailbox

  Show description for 7e67097
  NagusameCS

  committed
  7e67097

  Copy full SHA for 7e67097

• SMP: fix IPI dispatch, AP GDT/SSE2, stack overflow; add LAPIC timer, VM demand paging

  Show description for e11377e
  NagusameCS

  committed
  e11377e

  Copy full SHA for e11377e

Commits on Mar 2, 2026

• prod: real LLM inference, kprintf %f, AVX2 4x k-unroll, SMP 4-CPU verified

  Show description for 75a89ee
  NagusameCS

  committed
  75a89ee

  Copy full SHA for 75a89ee

• Update README.md

  NagusameCS

  authored
  7a4ae41

  Copy full SHA for 7a4ae41

• Implement BCM43455 firmware upload for working Bluetooth SPP

  Show description for ac33cf5
  NagusameCS

  committed
  ac33cf5

  Copy full SHA for ac33cf5

• README: Mermaid architecture diagrams, shields, updated roadmap & tree

  Show description for d378237
  NagusameCS

  committed
  d378237

  Copy full SHA for d378237

• Fix all warnings: zero -Wall warnings across 42 source files

  Show description for ef0543d
  NagusameCS

  committed
  ef0543d

  Copy full SHA for ef0543d