== 167d5034f946 2026-09-29T07:05:06Z
07:05:19 INFO    TKV recent ring: w=128 slack=2 for 4 sequences (1024 rows); 16 attention layer(s) read it through 16 ring layer(s); 32.53 MiB in scratch.attn_codec, 8.132 MiB per sequence.
07:05:24 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:05:24 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:05:24 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:05:24 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.2s
07:05:37 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:05:37 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:05:37 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:05:37 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.0s
07:05:49 INFO    engine ready, accepting requests on 0.0.0.0:8000
== b5eeed4259e6 2026-09-29T07:06:36Z
07:06:46 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:06:46 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:06:46 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:06:46 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (7 kernel variants), 0 swept, 1 cache-hit, 0.1s
07:06:58 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:06:58 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:06:58 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:06:58 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (7 kernel variants), 0 swept, 1 cache-hit, 0.0s
07:07:09 INFO    engine ready, accepting requests on 0.0.0.0:8000
== c90a17e56112 2026-09-29T07:24:18Z
07:24:23 INFO    TKV recent ring: w=128 slack=2 for 4 sequences (1024 rows); 16 attention layer(s) read it through 16 ring layer(s); 32.53 MiB in scratch.attn_codec, 8.132 MiB per sequence.
07:24:29 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:24:29 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:24:29 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:24:29 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.2s
07:24:42 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:24:42 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:24:42 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:24:42 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.0s
07:24:54 INFO    engine ready, accepting requests on 0.0.0.0:8000
== 4f975c8f6a56 2026-09-29T07:58:41Z
07:58:46 INFO    TKV recent ring: w=128 slack=2 for 4 sequences (1024 rows); 16 attention layer(s) read it through 16 ring layer(s); 32.53 MiB in scratch.attn_codec, 8.132 MiB per sequence.
07:58:52 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:58:52 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:58:52 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:58:52 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.2s
07:59:04 INFO    TKV warm-autotune: 1 fp × 16 layers, MTP verify block_m=none (MTP off) (miss → 1-time nvcc JIT, GPU idle, NOT a hang → /tkv-jit-cache)
07:59:04 INFO    TKV warm-autotune fingerprint 1/1: K4/V4 bits, codec=Klloyd/Vlloyd, head_dim=256, kv_heads=4, q_heads=24 (gqa=6), 16 layer(s)
07:59:04 INFO    [TKV warm-autotune] arbi-serve 1 fingerprint(s): 0 swept here, 0 trusted from a same-surface prior (no kernel launched), 1 cache-hit (measured against sweep-src=b5332b8dfc1c8389)
07:59:04 INFO    TKV warm-autotune complete: 1 fp × 16 layers, MTP verify block_m=none (MTP off) — loaded 1/1 combo(s) (13 kernel variants), 0 swept, 1 cache-hit, 0.0s
07:59:16 INFO    engine ready, accepting requests on 0.0.0.0:8000
