# qwen36 CUDA chain, the MTP head's adaptive draft list, prose, the most likely token, one draft a step: raw evidence (2026-10-26, UTC)
# Google Colab L4 VM: Intel(R) Xeon(R) CPU @ 2.20GHz (6 cores / 12 threads, 38.5 MiB L3) + NVIDIA L4 (23034 MiB, sm_89), 52 GiB; Ubuntu 24.04.4 LTS, Linux 6.6.122+ x86_64 (Google Colab), CUDA 13.0 (nvcc V13.0.88), driver 580.82.07, gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0.
# Model: Kreuzzelg/qwen36-35b-a3b-colibri-i4-gs64 (CUDA_EXPERT_GB=auto: all 10240 experts in VRAM), with its
#   MTP head (mtp.* of Qwen/Qwen3.6-35B-A3B, convert_qwen36.py --mtp-only); Q36_MTP=1 Q36_MTP_DRAFTS=1.
# v0 = 16d9f670: origin/dev 7bbea341 + the CUDA chain's small copies + the MTP draft picked on the
#   device, the changes this one builds on; v1 = 16d9f670 + this change (the measured build; after it only comments and
#   c/tests/ changed). Harness: bench.c beside this file (sha256
#   edbe7ddcc146af34d10c3d8a871fb1d3b93404537e374f9bc1504cf2c6886c17), built against each version's c/ with:
#   cd c && make qwen36_tier.o backend_cuda.o cuda_chain.o cu_xb.o cu_tier.o CUDA=1 CUDA_ARCH=sm_89 && gcc -O3 -march=native -fopenmp -pthread -Wall -Wextra -Wno-unused-parameter -Wno-misleading-indentation -Wno-unused-function -DCOLI_CUDA -I. <bench.c> qwen36_tier.o backend_cuda.o cuda_chain.o cu_xb.o cu_tier.o -o qwen36-bench -lm -fopenmp -pthread -L/usr/local/cuda/lib64 -Wl,-rpath,/usr/local/cuda/lib64 -lcudart -lcublasLt -lcublas -lstdc++
#   (Its comments name the driver and scripts it came with; the runs below need only this file.)
# Text: promessi_sposi.txt, speed-bench/promessi_sposi.txt of antirez's ds4 at commit 0aaea5a (sha256
#   f53e0d80cb2d4492d24ebd63c7000c397b16ae70f9bf09b3763e5d8323ec209f); for each length its first N tokens from an empty state, then 32 decode
#   steps of the most likely token, never an end of text. One process a run, the two lengths in order.
# A warm-up run of v0 (not counted), then v0 v1 v1 v0 v0 v1 v1 v0, each once the card was under 70 C
#   with no thermal clock event, then 60 s of rest. Every run's CSV and stderr in full, in the order they ran;
#   decode tok/s = gen_steady_tps (steps 2 to 32), the summary at the end.

===== setup
bench 482e1711d589; engine qwen36; Linux (Intel(R) Xeon(R) CPU @ 2.20GHz, NVIDIA L4); 2026-10-26 00:45
model /content/q36mtp; text promessi_sposi.txt (sha256 f53e0d80cb2d4492); lengths 64,512; gen 32; free (most likely token, never the end)
runs: a warm-up of v0 (does not count), then 4 per version, the order turning at every round (ABBA); before every run: wait until cool (at most 900 s), then 60 s of rest
v0 c4: /content/bin-c4/qwen36-bench (commit 16d9f6703aeb716135d5df22092e742be9cc9b5b; harness qwen36_bench.c edbe7ddcc146; CUDA); env: COLI_CUDA=1 COLI_GPUS=0 CUDA_EXPERT_GB=auto OMP_WAIT_POLICY=ACTIVE OMP_PROC_BIND=close Q36_MTP=1 Q36_MTP_DRAFTS=1
   command: env -i PATH=/usr/bin:/bin HOME=/root LD_LIBRARY_PATH=/usr/lib64-nvidia COLI_CUDA=1 COLI_GPUS=0 CUDA_EXPERT_GB=auto OMP_WAIT_POLICY=ACTIVE OMP_PROC_BIND=close Q36_MTP=1 Q36_MTP_DRAFTS=1 /content/bin-c4/qwen36-bench --model /content/q36mtp --text /content/bench/promessi_sposi.txt --lengths 64,512 --gen 32 --steps /content/c6-prosa-libera-1/v0-runN.steps
v1 c6: /content/bin-c6/qwen36-bench (commit 195e7e5ec793ae8e34c033a1d4513276b095d7da; harness qwen36_bench.c edbe7ddcc146; CUDA); env: COLI_CUDA=1 COLI_GPUS=0 CUDA_EXPERT_GB=auto OMP_WAIT_POLICY=ACTIVE OMP_PROC_BIND=close Q36_MTP=1 Q36_MTP_DRAFTS=1
   command: env -i PATH=/usr/bin:/bin HOME=/root LD_LIBRARY_PATH=/usr/lib64-nvidia COLI_CUDA=1 COLI_GPUS=0 CUDA_EXPERT_GB=auto OMP_WAIT_POLICY=ACTIVE OMP_PROC_BIND=close Q36_MTP=1 Q36_MTP_DRAFTS=1 /content/bin-c6/qwen36-bench --model /content/q36mtp --text /content/bench/promessi_sposi.txt --lengths 64,512 --gen 32 --steps /content/c6-prosa-libera-1/v1-runN.steps

===== warmup.csv (sha256 1d4a61cd41ba0796e5cfdf8e1d313f2021c95af3f5f70552f66ffc8f1374d00a)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.610,104.93,32,27.5,72.750,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.312,1640.32,32,21.2,63.807,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== warmup.err (sha256 74063a5084e80c560d5a8bafe817e29902327e34c6faa4d4113fcadc1df05cba)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.3s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.113 ms on CUDA dev 0, 0.419 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.31 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4)
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 66.7% (12/18 drafts) | gate T(1) 20.9 ms, rows 2:1.07, rho 0.50, acceptance by position 0.69 0.60 0.50 0.50 0.50
[mtp] head rows 18 (3.08 ms each), fed rows 83 (K/V only, 0.06 ms each), drafts 18
[qwen36 spec gate] 64 tokens: T(1) 20.9 ms, rows 2:1.07, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 18 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.6 ms, rows 2:1.56, rho 0.50, acceptance by position 0.43 0.60 0.50 0.50 0.50
[mtp] head rows 40 (3.49 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 40
[qwen36 spec gate] 512 tokens: T(1) 14.6 ms, rows 2:1.56, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 0 declined, 0 probes
[bench] peak RSS 20.04 GB

===== v0-run1.csv (sha256 de44a816bdaa34a793edff01bfa08db58b586ba9af0bcae28b10f3394047d003)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.610,104.92,32,27.0,82.290,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.319,1606.44,32,21.6,70.611,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v0-run1.err (sha256 78a4203f6ffc2acf9aff20d8b11151ccdc2b039c73c0ff25b2af9d11b5ad96fa)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.2s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.114 ms on CUDA dev 0, 0.423 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.7 s
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.31 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4)
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 66.7% (12/18 drafts) | gate T(1) 14.7 ms, rows 2:1.39, rho 0.50, acceptance by position 0.69 0.60 0.50 0.50 0.50
[mtp] head rows 18 (2.44 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 18
[qwen36 spec gate] 64 tokens: T(1) 14.7 ms, rows 2:1.39, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 18 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.9 ms, rows 2:1.35, rho 0.50, acceptance by position 0.43 0.60 0.50 0.50 0.50
[mtp] head rows 40 (2.44 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 40
[qwen36 spec gate] 512 tokens: T(1) 14.9 ms, rows 2:1.35, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 0 declined, 0 probes
[bench] peak RSS 20.04 GB

===== v1-run1.csv (sha256 029c39060e4faf3d60f361f2bd0defab6d4960556b03036f549a4ce05eef10f2)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.624,102.50,32,24.8,88.796,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.311,1647.94,32,19.4,76.736,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v1-run1.err (sha256 12825a89b149b486a5be0eaddc12dc20ffac69882e1dbc02b8b1b71e71d44551)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.3s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.113 ms on CUDA dev 0, 0.422 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] draft vocabulary: top 16000 + 3/4 rule + conversation, no counts file at /content/q36mtp/.coli_vocab yet
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.32 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4), drafts score an adaptive vocabulary
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 63.2% (12/19 drafts) | gate T(1) 14.7 ms, rows 2:1.26, rho 0.50, acceptance by position 0.66 0.60 0.50 0.50 0.50
[mtp] head rows 19 (0.57 ms each), fed rows 83 (K/V only, 0.06 ms each), drafts 19
[mtp] draft list 16003 tokens on average, 75.0% of picks inside, 8 ids appended, 0 rebuilds
[qwen36 spec gate] 64 tokens: T(1) 14.7 ms, rows 2:1.26, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 19 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.7 ms, rows 2:1.25, rho 0.50, acceptance by position 0.44 0.60 0.50 0.50 0.50
[mtp] head rows 41 (0.57 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 41
[mtp] draft list 16001 tokens on average, 87.5% of picks inside, 4 ids appended, 0 rebuilds
[qwen36 spec gate] 512 tokens: T(1) 14.7 ms, rows 2:1.25, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 1 declined, 0 probes
[bench] peak RSS 20.10 GB

===== v1-run2.csv (sha256 26dd482184ea6850f12753353a7f8e51bd8b139a065de6163fa5b1c8971bc5bc)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.620,103.16,32,24.6,88.691,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.310,1653.96,32,18.7,77.994,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v1-run2.err (sha256 090cafc27743cd05498fcfdb67b622f8a6acbfd65fe0c049d4a20240b3e74e97)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.3s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.113 ms on CUDA dev 0, 0.420 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] draft vocabulary: top 16000 + 3/4 rule + conversation, no counts file at /content/q36mtp/.coli_vocab yet
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.34 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4), drafts score an adaptive vocabulary
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 63.2% (12/19 drafts) | gate T(1) 14.8 ms, rows 2:1.26, rho 0.50, acceptance by position 0.66 0.60 0.50 0.50 0.50
[mtp] head rows 19 (0.56 ms each), fed rows 83 (K/V only, 0.06 ms each), drafts 19
[mtp] draft list 16003 tokens on average, 75.0% of picks inside, 8 ids appended, 0 rebuilds
[qwen36 spec gate] 64 tokens: T(1) 14.8 ms, rows 2:1.26, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 19 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.7 ms, rows 2:1.23, rho 0.50, acceptance by position 0.44 0.60 0.50 0.50 0.50
[mtp] head rows 41 (0.55 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 41
[mtp] draft list 16001 tokens on average, 87.5% of picks inside, 4 ids appended, 0 rebuilds
[qwen36 spec gate] 512 tokens: T(1) 14.7 ms, rows 2:1.23, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 1 declined, 0 probes
[bench] peak RSS 20.10 GB

===== v0-run2.csv (sha256 13706020147520e426c4df08c50f8126e02a719a6fe93407bc163bc9719acfc2)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.611,104.71,32,27.0,82.739,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.309,1658.98,32,21.1,71.248,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v0-run2.err (sha256 50816278af6813ed0bd55c74a7edbc6f07b7a83eac43ee0c69fbf2a1cee1699e)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.2s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.113 ms on CUDA dev 0, 0.417 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.25 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4)
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 66.7% (12/18 drafts) | gate T(1) 14.7 ms, rows 2:1.38, rho 0.50, acceptance by position 0.69 0.60 0.50 0.50 0.50
[mtp] head rows 18 (2.44 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 18
[qwen36 spec gate] 64 tokens: T(1) 14.7 ms, rows 2:1.38, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 18 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.7 ms, rows 2:1.35, rho 0.50, acceptance by position 0.43 0.60 0.50 0.50 0.50
[mtp] head rows 40 (2.44 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 40
[qwen36 spec gate] 512 tokens: T(1) 14.7 ms, rows 2:1.35, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 0 declined, 0 probes
[bench] peak RSS 20.04 GB

===== v0-run3.csv (sha256 45c7fd759ad3adce78b36b64e21ccca1254cf697cb64ce019ceded3bc9fefc40)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.613,104.41,32,26.8,82.027,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.312,1641.89,32,21.2,70.579,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v0-run3.err (sha256 a70077e10227306bf95953c0b2f097497c69ac9d9e5cfebf7bc223823fe7a8c5)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.2s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.114 ms on CUDA dev 0, 0.416 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.18 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4)
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 66.7% (12/18 drafts) | gate T(1) 14.7 ms, rows 2:1.40, rho 0.50, acceptance by position 0.69 0.60 0.50 0.50 0.50
[mtp] head rows 18 (2.44 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 18
[qwen36 spec gate] 64 tokens: T(1) 14.7 ms, rows 2:1.40, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 18 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.5 ms, rows 2:1.38, rho 0.50, acceptance by position 0.43 0.60 0.50 0.50 0.50
[mtp] head rows 40 (2.43 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 40
[qwen36 spec gate] 512 tokens: T(1) 14.5 ms, rows 2:1.38, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 0 declined, 0 probes
[bench] peak RSS 20.03 GB

===== v1-run3.csv (sha256 04e6c2c9f56873c55860eec27dd42b8de937f5d6696b777fb55b3ac1f1ee62b7)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.621,103.04,32,24.3,91.063,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.312,1640.79,32,18.8,78.681,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v1-run3.err (sha256 1ad1854dc393feecebcd26b13d5c0aa136147b9409dea453487b32fe6c5f1868)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.3s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.114 ms on CUDA dev 0, 0.417 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.7 s
[mtp] draft vocabulary: top 16000 + 3/4 rule + conversation, no counts file at /content/q36mtp/.coli_vocab yet
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.35 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4), drafts score an adaptive vocabulary
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 63.2% (12/19 drafts) | gate T(1) 14.7 ms, rows 2:1.24, rho 0.50, acceptance by position 0.66 0.60 0.50 0.50 0.50
[mtp] head rows 19 (0.55 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 19
[mtp] draft list 16003 tokens on average, 75.0% of picks inside, 8 ids appended, 0 rebuilds
[qwen36 spec gate] 64 tokens: T(1) 14.7 ms, rows 2:1.24, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 19 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.7 ms, rows 2:1.22, rho 0.50, acceptance by position 0.44 0.60 0.50 0.50 0.50
[mtp] head rows 41 (0.54 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 41
[mtp] draft list 16001 tokens on average, 87.5% of picks inside, 4 ids appended, 0 rebuilds
[qwen36 spec gate] 512 tokens: T(1) 14.7 ms, rows 2:1.22, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 1 declined, 0 probes
[bench] peak RSS 20.10 GB

===== v1-run4.csv (sha256 9d825d760405d7fab379adb7deb8fe93e16804ce777129eb3ee889ab214d570a)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.619,103.42,32,25.5,89.373,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.313,1638.26,32,19.5,77.804,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v1-run4.err (sha256 3409b4f4ef821935d29035d17be173984cc28c460b92885df485c4845e2ae12e)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.3s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.114 ms on CUDA dev 0, 0.419 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.5 s
[mtp] draft vocabulary: top 16000 + 3/4 rule + conversation, no counts file at /content/q36mtp/.coli_vocab yet
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.13 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4), drafts score an adaptive vocabulary
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 63.2% (12/19 drafts) | gate T(1) 14.8 ms, rows 2:1.25, rho 0.50, acceptance by position 0.66 0.60 0.50 0.50 0.50
[mtp] head rows 19 (0.55 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 19
[mtp] draft list 16003 tokens on average, 75.0% of picks inside, 8 ids appended, 0 rebuilds
[qwen36 spec gate] 64 tokens: T(1) 14.8 ms, rows 2:1.25, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 19 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.6 ms, rows 2:1.24, rho 0.50, acceptance by position 0.44 0.60 0.50 0.50 0.50
[mtp] head rows 41 (0.54 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 41
[mtp] draft list 16001 tokens on average, 87.5% of picks inside, 4 ids appended, 0 rebuilds
[qwen36 spec gate] 512 tokens: T(1) 14.6 ms, rows 2:1.24, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 1 declined, 0 probes
[bench] peak RSS 20.10 GB

===== v0-run4.csv (sha256 683661f6da39c329d8d062613dafff59c7cefca732fdbd00e3cb03a4652fd744)
ctx_tokens,prefill_tokens,prefill_s,prefill_tps,gen_tokens,gen_first_ms,gen_steady_tps,expert_miss,disk_s,fingerprint,tokens,nll,top1
64,64,0.608,105.23,32,26.7,81.848,0,0.00,d9d3351b46e6980d,1190b702cc495428,0.00000,0.000
512,512,0.313,1636.77,32,21.0,70.424,0,0.00,14c3a19d195d62a3,f57099df7a2d1304,0.00000,0.000
===== v0-run4.err (sha256 1e79acdca0632f6d17924d56cbb780dcfd4f3751197141a146c86f490547ac39)
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[OMP] qwen36: 6 physical-core threads instead of 12 logical CPUs; SMT can halve decode throughput on some CPUs (#718); set OMP_NUM_THREADS=<n> to override
== qwen36 Phase-2 engine | cache=256/layer bits=4 ctx=8192 pilot=0 wide=1 hot=0 smooth=0.30 conf=0.92 ==
[tok] loaded 248044 pieces (max id 248069) from /content/q36mtp/tokenizer.json
[meta] loaded: q_heads=16 kv_heads=2 head_dim=256 q_head_dim=512 k_head_dim=256 v_head_dim=256 o_in=4096 rotary_dim=64 n_experts=256 topk=8 inter=512 shared_inter=512 n_group=1 topk_group=1 attn_output_gate=1 n_active=10
[meta] DeltaNet: vheads=32 kheads=16 kdim=128 vdim=128 convk=4 conv_dim=8192
[qwen36] embedding: rows read in place from the checkpoint, COLI_EMBED_MMAP=0 builds the f32 table
[dense-i8] 291 matrices quantized during load, 7.2 GB f32 freed
[qwen36] group-scaled experts: gs=64
resident weights loaded in 9.2s | RSS after load: 3.03 GB
[qwen36] expert format on disk: int4 packed (tier fmt=4)
[qwen36] expert kernel: planar int4 in RAM (expert_ffn.h), QWEN_EXPERT_KERNEL=0 restores int8 unpack
[CUDA] device 0: NVIDIA L4, 23.7 GB VRAM, sm_89
[qtier] dev 0: 21.8 GB free, allowance 20.8 GB
[place] auto: dev 0 holds 1828.5 MB of trunk (lmhead yes, 30 dnproj layers), 19.02 GB left for experts
[qtier] dev 0: budget 19.02 GB for experts (~10794 experts) after trunk
[qtier] CUDA VRAM expert tier active: 1 device(s), 1.80 MB/expert
[CUDA] mode: routed experts (qwen36 VRAM tier)
[gpu] MoE experts -> CUDA VRAM tier
[place] probe: one [8192 x 2048] int8 GEMV takes 0.113 ms on CUDA dev 0, 0.415 ms on the CPU -> trunk to VRAM
[lmh] lm_head [248320 x 2048] int8 resident on CUDA dev 0 (0.47 GB)
[dnp] 30 DeltaNet-Projektionen auf GPU (0.70 GB VRAM)
[dense] 190 trunk matrices on GPU (30 dnout, 40 attnproj, 120 shexp; 0.61 GB VRAM)
[chain] qwen36: a decode step's routed experts in the frame (the tier's residents by address, format 4, group 64; COLI_CUDA_CHAIN_MOE=0: the host's step)
[chain] qwen36: 40 layers on CUDA device 0 (10 attention), lm_head too; 4.4 MiB of parameters, 35.3 MiB of router, b|a and gate rows, 66 MB of DeltaNet state
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qwen36] int4 packed weights detected — kept int4, repacked planar for expert_ffn.h
[qtier] warmstart (parallel): all 10240 experts in RAM (planar int4, the only copy), 10240 in VRAM -- 14.6 s
[mtp] Qwen3.6 checkpoint head: 256 experts
[bench] CUDA ready in 26.19 s; text 5871 tokens; lengths 64,512; gen 32; each length from an empty state
[qwen36] KV cache: float16, 0.00 GiB at 96 positions (half of float32; Q36_KV=f32 keeps float32)
[chain] qwen36: 4096 rows per prompt chunk (286 KiB a row, 3.1 GB free on the device; COLI_CUDA_CHAIN_ROWS sets it)
[chain] qwen36: up to 4 experts a step promoted into the card by this conversation's routes, the coldest residents giving way, none while the last 64 steps hit 75 % or more (COLI_CUDA_CHAIN_PROMOTE=0 keeps the warm start's set)
[mtp] the head runs on the CUDA chain's device, its experts in the frame (format 4)
[qwen36 mtp] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 66.7% (12/18 drafts) | gate T(1) 14.6 ms, rows 2:1.41, rho 0.50, acceptance by position 0.69 0.60 0.50 0.50 0.50
[mtp] head rows 18 (2.45 ms each), fed rows 83 (K/V only, 0.05 ms each), drafts 18
[qwen36 spec gate] 64 tokens: T(1) 14.6 ms, rows 2:1.41, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 64 tokens: 1.60 tokens/forward (20 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 18 verifies) | gate on, 0 declined, 0 probes
[qwen36 mtp] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 40.9% (9/22 drafts) | gate T(1) 14.5 ms, rows 2:1.39, rho 0.50, acceptance by position 0.43 0.60 0.50 0.50 0.50
[mtp] head rows 40 (2.44 ms each), fed rows 617 (K/V only, 0.03 ms each), drafts 40
[qwen36 spec gate] 512 tokens: T(1) 14.5 ms, rows 2:1.39, rho 0.50, acceptance by position 0.50 0.50 0.50 0.50 0.50
[qwen36 lookup] 512 tokens: 1.39 tokens/forward (23 forwards per 32 tokens) | acceptance 0.0% (0/0 drafts in 22 verifies) | gate on, 0 declined, 0 probes
[bench] peak RSS 20.02 GB

===== summary (bench's report of the runs above)
# bench: c6-prosa-libera-1

- bench 482e1711d589; engine qwen36; Linux (Intel(R) Xeon(R) CPU @ 2.20GHz, NVIDIA L4); 2026-10-26 00:45
- model q36mtp; text promessi_sposi.txt (sha256 f53e0d80cb2d4492); lengths 64,512; gen 32; free (most likely token, never the end)
- runs: a warm-up of v0 (does not count), then 4 per version, the order turning at every round (ABBA); before every run: wait until cool (at most 900 s), then 60 s of rest

## v0 c4: 4 runs (median, min–max)

| tokens | new | prompt tok/s | first step ms | decode tok/s | experts read | reads' s | fingerprint |
|---|---|---|---|---|---|---|---|
| 64 | 64 | 104.81 (104.41–105.23) | 27 (27–27) | 82.159 (81.848–82.739) | 0 | 0.00 | d9d3351b46e6980d |
| 512 | 512 | 1639.33 (1606.44–1658.98) | 21 (21–22) | 70.595 (70.424–71.248) | 0 | 0.00 | 14c3a19d195d62a3 |

peak memory 20516 MiB, model ready in 26.22 s (medians)

decode by steps (step 1 gives the second new word; medians of 4 runs):

| tokens | steps | experts read | decode tok/s | reads' s |
|---|---|---|---|---|
| 64 | 2 to 32 | 0 | 82.158 | 0.000 |
| 512 | 2 to 32 | 0 | 70.595 | 0.000 |

## v1 c6: 4 runs (median, min–max)

| tokens | new | prompt tok/s | first step ms | decode tok/s | experts read | reads' s | fingerprint |
|---|---|---|---|---|---|---|---|
| 64 | 64 | 103.10 (102.50–103.42) | 25 (24–26) | 89.085 (88.691–91.063) | 0 | 0.00 | d9d3351b46e6980d |
| 512 | 512 | 1644.37 (1638.26–1653.96) | 19 (19–20) | 77.899 (76.736–78.681) | 0 | 0.00 | 14c3a19d195d62a3 |

peak memory 20582 MiB, model ready in 26.33 s (medians)

decode by steps (step 1 gives the second new word; medians of 4 runs):

| tokens | steps | experts read | decode tok/s | reads' s |
|---|---|---|---|---|
| 64 | 2 to 32 | 0 | 89.085 | 0.000 |
| 512 | 2 to 32 | 0 | 77.899 | 0.000 |

## v1 c6 against v0 c4 (medians; positive = faster)

| tokens | prompt | first step | decode | results |
|---|---|---|---|---|
| 64 | -1.6%, every run slower | +8.9%, every run faster | +8.4%, every run faster | same results |
| 512 | +0.3%, ranges overlap | +10.7%, every run faster | +10.3%, every run faster | same results |

| tokens | steps | experts read | decode | reads' s |
|---|---|---|---|---|
| 64 | 2 to 32 | 0 → 0 | +8.4%, every run faster | 0.000 → 0.000 s |
| 512 | 2 to 32 | 0 → 0 | +10.3%, every run faster | 0.000 → 0.000 s |

warm-up (does not count): 64 tokens: prompt 104.93 tok/s, decode 72.750 tok/s; 512 tokens: prompt 1640.32 tok/s, decode 63.807 tok/s
runs not cool: none
pages written to swap during the measurement: 0, 0 of them in the runs that count
