feat(frank-zealot): Bonsai-2-9B PQ2_0 with the MTP head at 3 x 262,144
Swap the 3090 endpoint from Ternary-Bonsai-2-27B PQ2_0 (2 x 200,704,
fork build 10754) to Ternary-Bonsai-2-9B PQ2_0 with the MTP head IN the
file, three conversations at the trained 262,144 window
(--ctx-size 786432 --parallel 3), on vendor fork build 10777 -- the
minimum build that pairs MTP with a quantized V cache. q8_0 K + q8_0 V
is the only KV that fits: 13,056 MiB against a 26,112 MiB f16 pair.
Settled boot 17,876 used / 6,249 free.
MTP is armed and then DECLINED by measurement: the draft context bills
its own f16 KV over the whole pool (3,072 MiB) and at UBATCH_SIZE=2048
that halves decode (167.2 -> 109.2 t/s). The documented MTP arm
(SPEC_TYPE=draft-mtp + UBATCH_SIZE=1024, 213.5 t/s) is recorded in the
env header rather than left to rediscovery.
Two bugs the trial caught: ${SPEC_PRESENCE_PENALTY:=0.0} was a
default-assignment, so the env value always won and the no-presence-penalty
MTP arm was unreachable (now ${VAR=0.0}); and the Q8_0 mmproj is the
publisher's requirement for concurrent image requests, which matters at
--parallel 3.
contextWindow is now 262144 -- the widest in the fleet -- so :7771 is the
compaction hub. 27B runbook re-banned RETIRED with its bench-window
section closed; indexes, homepage card, OpenWebUI doc and the worklog
re-keyed.
This commit is contained in:
Frank committed
1 parent
2b127163c8
commit
466fd477d8
14 files changed
+1001
-314
No files matched your search
+104
-51
@@ -1,29 +1,39 @@
|
||||
# frank-zealot — llama.cpp (PrismML fork build, CUDA) serving
|
||||
# prism-ml/Ternary-Bonsai-2-27B-gguf at PQ2_0 on the RTX 3090, :7771.
|
||||
# frank-zealot — llama.cpp (PrismML fork build 10777, CUDA) serving
|
||||
# prism-ml/Ternary-Bonsai-2-9B-gguf at PQ2_0 **with the MTP head in the file**
|
||||
# on the RTX 3090, :7771.
|
||||
#
|
||||
# 2026-10-17 (Frank): swap from Qwen3.5-9B-Uncensored Q6_K to the Bonsai-2-27B
|
||||
# ternary checkpoint at the SAME shape Frank asked for — TWO 200,704-token
|
||||
# conversations — with q8_0 K + q8_0 V (f16 is arithmetically impossible here,
|
||||
# see the KV law), MTP off (the file carries no nextn/mtp tensors), verbosity 4.
|
||||
# 2026-10-18 (Frank): swap from the Bonsai-2-**27B** PQ2_0 to the 9B MTP export
|
||||
# at THREE 262,144-token conversations (Frank: "if you can possibly fit a third
|
||||
# context, do it"). The 9B bills HALF the KV of the 27B per token, which is what
|
||||
# funds 3 x the native window on one 24,123 MiB card. Speculative decoding ships
|
||||
# OFF — measured, not assumed; see the MTP law below, which is the whole story
|
||||
# and the reason the publisher's "depth 2 works best" line does not transfer.
|
||||
#
|
||||
# THE RUNTIME IS A LOCK, NOT A PREFERENCE. The file is general.file_type 141
|
||||
# with 402 tensors of the CUSTOM GGML type 142 (pq2_0) plus prism.hadamard.*
|
||||
# metadata (normalized-Sylvester-Walsh transform, block 1024, 401 named
|
||||
# weights, inverse on token_embd, gdn_v_grouped). Stock llama.cpp refuses type
|
||||
# 142 at load; only the PrismML fork registers the type AND reads the Hadamard
|
||||
# metadata. The vendor prebuilt build 10754 (commit 2459f68b5) is what every
|
||||
# number in this file was measured on; sha256 12ae5baba033d8cd…
|
||||
LLAMA_BIN=/home/dank/GitPackages/llama-prism/llama-server
|
||||
MODEL_FILE=/home/dank/models/Bonsai2-PQ2_0/Ternary-Bonsai-2-27B-PQ2_0.gguf
|
||||
MMPROJ_FILE=/home/dank/models/Bonsai2-PQ2_0/Ternary-Bonsai-2-27B-mmproj-Q8_0.gguf
|
||||
# The CRACK re-export (dealignai/Bonsai-2-27B-Ternary-CRACK-GGUF, uncensored,
|
||||
# already on disk at /home/dank/models/Bonsai2-CRACK/) is a ONE-LINE swap: the
|
||||
# header census proves byte-identical geometry (851 tensors, same
|
||||
# {pq2_0:402, f32:353, bf16:96} histogram, 64 blocks, vocab 248,320, same
|
||||
# 8,952-char chat template) and the official file's sha256 is the Hub LFS oid.
|
||||
# Bonsai-2 is the BASE (censored) Qwen3.8 family; if a compaction or subagent
|
||||
# request ever comes back with a refusal, point MODEL_FILE at the CRACK file
|
||||
# and restart — the VRAM ledger below does not move.
|
||||
# carrying custom GGML type 142 tensors plus prism.hadamard.* metadata
|
||||
# (normalized-Sylvester-Walsh transform); stock llama.cpp refuses type 142 at
|
||||
# load and a loader that ignored the Hadamard metadata would dequant WRONG. On
|
||||
# top of that lock, MTP beside a quantized V cache needs fork build >= b10776,
|
||||
# so the vendor prebuilt moved from 10754 (commit 2459f68b5, the 27B era) to
|
||||
# 10777 (commit e3de71522). The 10754 tree stays on disk as the rollback
|
||||
# runtime for the 27B payload.
|
||||
LLAMA_BIN=/home/dank/GitPackages/llama-prism-b10777/llama-server
|
||||
# 2,675,410,720 B, Hub LFS sha256 c79fcdfaf6b7… (verified at fetch by
|
||||
# scripts/download-frank-zealot-bonsai9.sh). This IS the MTP build: the draft
|
||||
# head ships INSIDE the checkpoint, so MTP availability is a property of the
|
||||
# FILE, never of a flag. The plain PQ2_0 sibling could not run it at any flag
|
||||
# value; loading this file instead of that one is the only change that matters.
|
||||
MODEL_FILE=/home/dank/models/Bonsai2-9B-MTP/Ternary-Bonsai-2-9B-PQ2_0-MTP-Q8_0.gguf
|
||||
# 624,229,664 B, Hub LFS sha256 e0cb6f69… — Q8_0 IS MANDATORY, not a preference:
|
||||
# the publisher's card says the BF16 mmproj "can crash llama-server when several
|
||||
# image requests run at once; Q8_0 is not affected", and this endpoint runs
|
||||
# --parallel 3, i.e. exactly that shape. The projector's worst case lands AFTER
|
||||
# the engine's own fit check, so it is never inside a projection number.
|
||||
MMPROJ_FILE=/home/dank/models/Bonsai2-9B-MTP/Ternary-Bonsai-2-9B-mmproj-Q8_0.gguf
|
||||
# Base alignment: Bonsai-2 is the censored BASE lineage. The 27B CRACK re-export
|
||||
# (/home/dank/models/Bonsai2-CRACK/) is byte-identical geometry to the 27B it
|
||||
# matched, NOT to this 9B file — if a refusal ever surfaces there is no
|
||||
# one-line swap here; the 9B dealigned export would have to be fetched first.
|
||||
#
|
||||
# Served model id — the harness/family name. The SERVICE, unit, scripts and
|
||||
# this env file keep the frank-zealot name; only the alias on the wire is the
|
||||
@@ -32,40 +42,83 @@ MMPROJ_FILE=/home/dank/models/Bonsai2-PQ2_0/Ternary-Bonsai-2-27B-mmproj-Q8_0.ggu
|
||||
ALIAS=frank-zealot
|
||||
BIND_ADDR=0.0.0.0
|
||||
LISTEN_PORT=7771
|
||||
# CTX_SIZE is the POOL llama.cpp divides across slots: 401,408 / 2 = 200,704
|
||||
# per conversation. MUST stay divisible by PARALLEL (the seam enforces it) and
|
||||
# the harness contextWindow quotes the PER-SLOT number — a longer prompt is
|
||||
# answered with a 400, never truncated. The model's native window is 262,144;
|
||||
# 2 × 262,144 at q8_0 projects 25,758 MiB and does NOT fit this card (measured
|
||||
# f16 analogue: 41,118). One 262,144 slot fits with 5,995 MiB free but
|
||||
# serializes compaction, so the pair stays.
|
||||
CTX_SIZE=401408
|
||||
PARALLEL=2
|
||||
# CTX_SIZE is the POOL llama.cpp divides across slots: 786,432 / 3 = 262,144 per
|
||||
# conversation — each slot at the model's TRAINED window, which is the point of
|
||||
# the swap (the 27B era had to cut each conversation to 200,704 to fit two).
|
||||
# MUST stay divisible by PARALLEL (the seam enforces it) and the harness
|
||||
# contextWindow quotes the PER-SLOT number — a longer prompt is answered with a
|
||||
# 400, never truncated. The ceiling cannot go past 3 x 262,144: that is the
|
||||
# native wall (no YaRN in this path) and the arithmetic agrees — a 4th slot
|
||||
# would push each below the compaction contract.
|
||||
CTX_SIZE=786432
|
||||
PARALLEL=3
|
||||
BATCH_SIZE=4096
|
||||
# Ubatch is a prefill knob (decode never moves) with NO interior optimum on this
|
||||
# CUDA card up to 2048 — 2048 is the measured optimum (+4.6 % prefill). It has a
|
||||
# second job here: it sets the compute-buffer size, and the compute buffers are
|
||||
# what decide whether MTP fits (see the MTP law). 1024 is the documented MTP arm.
|
||||
UBATCH_SIZE=2048
|
||||
# KV law (from the GGUF header, then confirmed by the boot ledger): 16
|
||||
# full-attention layers (full_attention_interval 4 over 64 blocks) × 4 KV heads
|
||||
# × 256 dim = 16,384 elements per token per cache. q8_0 bills 17,408 B per
|
||||
# cache ⇒ 34,816 B/token = 34 KiB, so the 401,408 pool is 13,328.00 MiB — the
|
||||
# ledger said exactly that. f16 would bill 25,088 MiB and the boot DIES:
|
||||
# `cudaMalloc failed: out of memory` on the 25,088.00 MiB request, engine fit
|
||||
# projection 33,198 vs 23,558 free. The 48 DeltaNet layers bill a fixed
|
||||
# recurrent state per SLOT — ~150 MiB each (context 13,627 − KV 13,328 = 299
|
||||
# for two slots), which is why pool arithmetic alone under-prices the shape.
|
||||
# KV law (header + boot ledger, both confirmed): qwen35 hybrid, 32 layers with
|
||||
# qwen35.full_attention_interval = 4 ⇒ 8 full-attention + 24 Gated DeltaNet;
|
||||
# 4 KV heads × 256 dim = 8,192 elements per token per cache. q8_0 bills 8,704 B
|
||||
# per cache ⇒ 17,408 B/token = 17 KiB, so the 786,432 pool is 13,056.00 MiB —
|
||||
# the ledger said exactly that. HALF the 27B's 34,816 B/token, which is the
|
||||
# whole reason three native windows fit. f16 would bill 26,112 MiB at this pool
|
||||
# (against a 24,123 MiB card) and dies at cudaMalloc; the 24 DeltaNet layers
|
||||
# bill a fixed recurrent state per SLOT instead of per token — the ledger's
|
||||
# "RS buffer size = 452.25 MiB (3 cells, 32 layers, 3 seqs)" = 150.75 MiB a
|
||||
# slot — which is why pool arithmetic alone under-prices the shape.
|
||||
CACHE_TYPE_K=q8_0
|
||||
CACHE_TYPE_V=q8_0
|
||||
# Boot ledger (2026-10-17, this exact shape, nvidia-smi settled):
|
||||
# 22,664 MiB used / 1,461 MiB free of 24,123
|
||||
# fit projection 21,439 = 6,539 model + 13,627 context + 1,272 compute
|
||||
# the mtmd projector's 873.10 MiB worst-case is NOT inside that projection —
|
||||
# it lands after the fit check, which is the whole difference between
|
||||
# "2.6 GiB free on paper" and 1.4 GiB free in reality. Watch it.
|
||||
# Fallback rung if the margin is ever eaten (desktop, co-tenant): q4_0 KV
|
||||
# measured 16,392 used / 7,733 free at the same 2 × 200,704.
|
||||
# 1024 is the ubatch fallback (−616 MiB, measured on the 9B predecessor at
|
||||
# this card/arch; 2048 was the measured CUDA optimum, +4.6 % prefill).
|
||||
# SPECULATIVE DECODING: OFF, MEASURED. The head is in the file and the drafter
|
||||
# is GOOD (acceptance 0.798/0.617 per position, mean accepted length 2.41) — it
|
||||
# still loses, because the draft context bills its OWN f16 KV over the WHOLE
|
||||
# pool: 4,096 B/position × 786,432 = 3,072.00 MiB, plus its compute buffers,
|
||||
# 4,830 MiB all told. Booted arms, same card, same traffic (2026-10-18):
|
||||
# 3 × 262,144 spec OFF ub 2048 → 17,876 used / 6,249 free, decode 167.2 t/s
|
||||
# 3 × 262,144 spec ON ub 2048 → 22,706 used / 1,419 free, decode 109.2 t/s
|
||||
# 3 × 262,144 spec ON ub 1024 → 21,400 used / 2,725 free, decode 213.5 t/s
|
||||
# 2 × 262,144 spec ON ub 2048 → 17,178 used / 6,947 free, decode 213.2 t/s
|
||||
# 1 × 262,144 spec ON ub 2048 → 11,652 used / 12,473 free, decode 214.2 t/s
|
||||
# i.e. MTP is worth +28 % decode whenever the card has room, and at the 3-slot
|
||||
# pool it costs 1.3 GiB of compute buffers that ubatch 2048 refuses to give up —
|
||||
# the squeeze, not the drafter, is what halves decode. The arm that WOULD run
|
||||
# MTP at three windows is SPEC_TYPE=draft-mtp + UBATCH_SIZE=1024 (decode 213.5,
|
||||
# 3-stream 250.8 aggregate) and pays prefill −17 % and 2.7 GiB of headroom on a
|
||||
# card whose projector worst-case lands after the fit check. This endpoint's
|
||||
# load is the fleet's COMPACTION HUB (prefill-bound, ~170k in) plus 2-3
|
||||
# concurrent subagents, so the deployed arm takes the +17 % prefill
|
||||
# (2,877 vs 2,395 t/s at 106,418 tokens) and the +12 % three-stream aggregate
|
||||
# (280.2 vs 250.8) over single-stream speed, and keeps 6.2 GiB free.
|
||||
# SPEC_TYPE=none strips the WHOLE spec set (--spec-type, every --spec-draft-*):
|
||||
# flags move as a set or the arm measures a crash instead of an absence.
|
||||
SPEC_TYPE=none
|
||||
SPEC_DRAFT_N_MAX=2
|
||||
SPEC_DRAFT_N_MIN=0
|
||||
SPEC_DRAFT_P_MIN=0
|
||||
# The card's sampler recipe lives in the checkpoint (general.sampling.* = temp
|
||||
# 1.0 / top_p 0.95 / top_k 20 / min_p 0 / repeat_penalty 1.0 — confirmed in the
|
||||
# engine's own /props defaults) and llama.cpp reads it itself. Presence penalty
|
||||
# is the ONE recipe value with no GGUF metadata key, so it must reach the argv.
|
||||
# PRESENCE_PENALTY="" is the documented opt-out.
|
||||
PRESENCE_PENALTY=1.0
|
||||
NGL=99
|
||||
# Boot ledger (2026-10-18, this exact shape, nvidia-smi settled):
|
||||
# 17,876 MiB used / 6,249 MiB free of 24,123
|
||||
# CUDA0 KV 13,056.00 MiB (q8_0 K + q8_0 V, 786,432 positions)
|
||||
# CUDA0 RS 452.25 MiB (3 recurrent-state cells)
|
||||
# compute 1,409 MiB CUDA + 1,201 MiB CUDA_Host, projector 595 + 248 MiB
|
||||
# no clamp line, no cudaMalloc error
|
||||
# Measured on the same boot: 167.2 t/s greedy decode (/completion, 512 tokens),
|
||||
# 158.9 t/s single-stream chat, 280.2 t/s aggregate across THREE concurrent
|
||||
# streams (93.8 each), 2,877 t/s prefill at 106,418 tokens. The 27B record it
|
||||
# replaces: 76.6 decode, 109.4 aggregate across two streams, 691.7 prefill at
|
||||
# 166k — beaten on every axis. The FIRST request after a boot is a warmup
|
||||
# artifact (37.7 t/s class); never report it.
|
||||
# Fallback rung if the margin is ever eaten (desktop, co-tenant): q4_0 KV on the
|
||||
# 27B era measured 16,392 used / 7,733 free at 2 × 200,704; the same drop here
|
||||
# frees ~6.5 GiB at three windows. UBATCH_SIZE=1024 is the cheaper lever
|
||||
# (−1,306 MiB measured on this shape).
|
||||
# Log verbosity threshold (llama.cpp -lv). 4 = debug: KV layout, buffer sizes
|
||||
# and graph capture land in the journal, which is what the VRAM-fit forensics
|
||||
# read. Levels above 4 flood the journal.
|
||||
|
||||
Reference in new issue
Block a user