cortex

fix(gateway): full observability + stop leaking upstream bodies

#160:Commit aa88d37509 pushed by grenade

main

2026-05-22 04:37:03 +00:00

19m27s

fix(router): rewrite loopback inference URLs to use neuron's host

#156:Commit 9b0ed0b57f pushed by grenade

main

2026-05-22 03:43:50 +00:00

19m49s

feat(stage-8e-3): quantize lm_head in TP Qwen3-Next

#152:Commit e71181499e pushed by grenade

main

2026-05-21 19:14:09 +00:00

20m42s

feat(stage-8e-2d): route quantized matmul by M (prefill vs decode)

#148:Commit 34f9b77d9d pushed by grenade

main

2026-05-21 18:36:26 +00:00

20m40s

fix(stage-8e-2c): cast bf16/f16 activations to f32 around QMatMul

#146:Commit f084aaab8e pushed by grenade

main

2026-05-21 17:25:24 +00:00

19m56s

fix(stage-8e-2b): allow quant on the TP load path

#144:Commit 68a606a79c pushed by grenade

main

2026-05-21 16:46:09 +00:00

28m46s

feat(stage-8e-2): plumb quant config from ModelSpec to TP load path

#142:Commit 4aa71902d0 pushed by grenade

main

2026-05-21 15:44:17 +00:00

40m32s

diag(stage-8d-6): per-layer VRAM logging in TP load path

#136:Commit 89d98d1fb2 pushed by grenade

main

2026-05-21 10:14:05 +00:00

19m51s

feat(stage-8d-5b): wire fused_gdn_gating CUDA kernel

#134:Commit cc95fe28d9 pushed by grenade

main

2026-05-21 09:14:29 +00:00

21m43s

feat(tp): cancellation-safe inference + structured tracing

#122:Commit 70eb6af42b pushed by grenade

main

2026-05-21 05:41:50 +00:00

19m41s

fix(tp): always drain worker responses on leader failure

#120:Commit d1a4aad91d pushed by grenade

main

2026-05-21 04:58:46 +00:00

18m53s

feat(stage-8c): TP-aware Qwen3-Next (tp_qwen3_5)

#118:Commit 95dc8745eb pushed by grenade

main

2026-05-20 19:24:58 +00:00

22m6s

fix(qwen3_5): promote beta to F32 alongside q/k/v in delta rule

#116:Commit 495d3f7c05 pushed by grenade

main

2026-05-20 18:39:29 +00:00

25m57s

fix(qwen3_5): tensor names are under `model.language_model.*`, not `model.*`

#114:Commit 5c4c8e0eba pushed by grenade

main

2026-05-20 14:12:06 +00:00

23m45s

fix(qwen3_5): nested rope_parameters + partial_rotary_factor=0.25

#110:Commit 07c44d5db1 pushed by grenade

main

2026-05-20 13:39:23 +00:00

20m21s

feat(stage-8c): full-attention layer + decoder + Model + ForCausalLM for qwen3_5

#108:Commit e7eb3dab6a pushed by grenade

main

2026-05-20 13:13:12 +00:00

20m25s

feat(stage-8c): linear-attention layer (Qwen3-Next GatedDeltaNet)

#106:Commit 180274548d pushed by grenade

main

2026-05-20 06:49:42 +00:00

19m40s

feat(stage-8c): scaffold qwen3_5 (Qwen3.6) — dispatch + stubs + TP gate

#104:Commit a70f317729 pushed by grenade

main

2026-05-20 06:19:33 +00:00

21m22s

feat(stage-8b): Llama + Qwen3 MoE families on the candle harness

#102:Commit c6022aa6b9 pushed by grenade

main

2026-05-20 05:55:49 +00:00

19m18s

feat(neuron): honour HF_HUB_CACHE / HF_HOME for the candle harness cache

#98:Commit b400e8b704 pushed by grenade

main

2026-05-20 05:15:27 +00:00

22m26s

feat(tp): Stage 7b-iv — RPC + orchestration for TP load/inference

#90:Commit d46d8d4f6c pushed by grenade

main

2026-05-20 04:20:53 +00:00

26m44s

feat(tp): --tp-smoke CLI subcommand + remote validation script

#88:Commit 9b8bd146f6 pushed by grenade

main

2026-05-19 17:00:10 +00:00

19m35s

fix(tp): add half dep + drop double-wrapped .w() on CudaDevice::alloc

#86:Commit 96d8755245 pushed by grenade

main

2026-05-19 16:32:03 +00:00

19m54s

fix(neuron/candle): dense Qwen3 returns rank-3 logits, double-squeeze

#80:Commit 5436af9c73 pushed by grenade

main

2026-05-19 15:09:06 +00:00

19m11s

fix(neuron/tp): NcclError {e:?} + cudarc 0.19 deprecation cleanup

#78:Commit 8e882c0757 pushed by grenade

main

2026-05-19 14:43:33 +00:00

19m9s

Stage 7a-i: TP worker lifecycle scaffolding

#70:Commit 2a7ede0232 pushed by grenade

main

2026-05-19 13:13:08 +00:00

19m53s

post-validation cleanup: cuDNN runtime + repetition penalty

#68:Commit 18ae3c30ee pushed by grenade

main

2026-05-19 12:12:41 +00:00

24m22s

fix(deploy): use dnf upgrade for stale installs, install only when absent

#66:Commit 1a0400131e pushed by grenade

main

2026-05-19 11:30:40 +00:00

19m41s

fix(validate-neuron): jq for JSON, say→stderr, sane max_tokens

#64:Commit 1866b99a89 pushed by grenade

main

2026-05-19 11:02:08 +00:00

18m54s

fix(neuron/candle): source tokenizer.json from base repo when GGUF

#62:Commit 602e8e1471 pushed by grenade

main

2026-05-19 10:36:38 +00:00

19m49s