frank-default-small: ncmoe 35->36 for prefill stability

Fix a live crash: cuBLAS cudaErrorInvalidValue in ggml_cuda_mul_mat_cublas
killed the service during a 145k-token prefill burst at ncmoe35/ub3072.

ncmoe 36 moves one more expert layer to CPU: frees ~0.75 GiB peak VRAM
(grid: peak 8606 vs 9360 MiB at ub3072) for -2% prefill / -2% decode.
Keeps prefill priority (top ubatch tier). Docs synced; noted fallback
GGML_CUDA_FORCE_MMQ=1 if pressure persists.

docs/Dankserver System Specs.md

@@ -51,11 +51,11 @@ Dankserver is an NVIDIA GPU server for vLLM inference. The host runs Ubuntu 24.0
 
 ### Active Services
 
-The active `frank-default-small` service (llama.cpp, `b10108+`) serves the HauhauCS Qwen3.6-35B-A3B uncensored `Q6_K_P` model on port 7770 with **ngram-mod** speculative decoding. It runs **4 parallel slots** at 200000 context each (pool 800000). Of the 40 layers' MoE expert weights, **35 sit on CPU** (`--n-cpu-moe 35`) while the remaining **5 layers' expert weights** are resident on GPU along with all non-expert tensors (`--gpu-layers all`). KV stays on GPU with K=`q8_0`, V=`f16`. Batch 8192 / ubatch 3072, `--ctx-checkpoints 4`, `--no-mmap`. At idle ~22.8 GB VRAM used, leaving an estimated **~1.2 GiB** headroom (chosen by benchmark for prefill speed, see [[frank-default-small-benchmark]]). Idle RAM ~32 GB of 62 GB.
+The active `frank-default-small` service (llama.cpp, `b10108+`) serves the HauhauCS Qwen3.6-35B-A3B uncensored `Q6_K_P` model on port 7770 with **ngram-mod** speculative decoding. It runs **4 parallel slots** at 200000 context each (pool 800000). Of the 40 layers' MoE expert weights, **36 sit on CPU** (`--n-cpu-moe 36`) while the remaining **4 layers' expert weights** are resident on GPU along with all non-expert tensors (`--gpu-layers all`). KV stays on GPU with K=`q8_0`, V=`f16`. Batch 8192 / ubatch 3072, `--ctx-checkpoints 4`, `--no-mmap`. At idle ~21.05 GB VRAM used, leaving an estimated **~3.0 GiB** headroom (chosen for prefill stability, see [[frank-default-small-benchmark]]). Idle RAM ~32 GB of 62 GB.
 
 ## Hardware Relevance
 
-- 24 GB VRAM is close to saturated with this config. At idle ~22.8 GB used, leaving an estimated **~1.2 GiB** headroom at `--n-cpu-moe 35` (5 layers' experts on GPU) with `ubatch 3072`. A llama-bench sweep ([[frank-default-small-benchmark]]) chose this state because prefill is the priority: `ubatch 3072` reaches ~1458 t/s prefill, the top tier, and generation is independent of ubatch. If decode speed matters more, use `ncmoe 32 - ubatch 1024` (gen ~48.7 t/s). By editing `scripts/frank-default-small.env`.
+- 24 GB VRAM is close to saturated with this config. At idle ~21.05 GB used, leaving an estimated **~3.0 GiB** headroom at `--n-cpu-moe 36` (4 layers' experts on GPU) with `ubatch 3072`. A llama-bench sweep ([[frank-default-small-benchmark]]) guided this state: prefill is the priority, so `ubatch 3072` stays (top prefill tier), and `ncmoe 36` was raised from 35 to fix a cuBLAS OOM-class crash during big prefills by freeing ~0.75 GiB peak VRAM at only -2% prefill. If it still crashes under load, add `GGML_CUDA_FORCE_MMQ=1`. If decode speed matters more, use `ncmoe 32 - ubatch 1024` (gen ~48.7 t/s). By editing `scripts/frank-default-small.env`.
 - 62 GB system RAM holds the ~29 GB of MoE expert weights with room to spare (~57 GB available after load).
 - 127 GB swap is available for CPU KV cache paging if GPU KV offload is disabled.
 - Root disk has approximately 140 GB free. Store the GGUF models under `/home/dank/models` (about 32 GB total).

docs/deployment.md

@@ -143,7 +143,7 @@ The active service on Dankclaw runs llama.cpp serving the HauhauCS Qwen3.6-27B u
 
 ### Dankserver — frank-default-small
 
-The active service on Dankserver runs llama.cpp (upstream, `b10108+`) and serves the HauhauCS Qwen3.6-35B-A3B uncensored `Q6_K_P` model under the alias `frank-default-small` on port 7770. It uses **ngram-mod** speculative decoding (`--spec-ngram-mod-n-min 48 / n_max 64 / n_match 24`), **4 parallel slots** at 200000 context each (pool 800000), and mmproj vision. **35 of 40 MoE expert layers** stay on CPU (`--n-cpu-moe 35`); 5 expert layers' weights are back on GPU alongside all non-expert tensors (`--gpu-layers all`). KV stays on GPU (K=`q8_0`, V=`f16`). Batch 8192 / ubatch 3072, `--ctx-checkpoints 4`, `--no-mmap`. The unit file is `scripts/frank-default-small.service`, symlinked to `/etc/systemd/system/`. It launches through `scripts/serve-llama-frank-small.sh`, which reads tunables from `scripts/frank-default-small.env`.
+The active service on Dankserver runs llama.cpp (upstream, `b10108+`) and serves the HauhauCS Qwen3.6-35B-A3B uncensored `Q6_K_P` model under the alias `frank-default-small` on port 7770. It uses **ngram-mod** speculative decoding (`--spec-ngram-mod-n-min 48 / n_max 64 / n_match 24`), **4 parallel slots** at 200000 context each (pool 800000), and mmproj vision. **36 of 40 MoE expert layers** stay on CPU (`--n-cpu-moe 36`); 4 expert layers' weights are back on GPU alongside all non-expert tensors (`--gpu-layers all`). KV stays on GPU (K=`q8_0`, V=`f16`). Batch 8192 / ubatch 3072, `--ctx-checkpoints 4`, `--no-mmap`. The unit file is `scripts/frank-default-small.service`, symlinked to `/etc/systemd/system/`. It launches through `scripts/serve-llama-frank-small.sh`, which reads tunables from `scripts/frank-default-small.env`.
 
 > Note: the Dankserver process currently observed runs `--spec-type ngram-mod` with `--parallel 4`, batch 8192 / ubatch 2048, and `--n-cpu-moe 34` (this is being bench-tuned by the operator; reconcile against `scripts/frank-default-small.env`).
 

docs/frank-default-small-benchmark.md

@@ -101,24 +101,31 @@ and fails, which matches the recorded OOMs at `ncmoe 32` and `33`.
 
 ## Chosen State
 
-Prefill speed is the priority for this workload, so the final state takes
-the top prefill tier while keeping a safe headroom margin:
+Prefill speed is the priority, so the state keeps the top prefill tier and
+moves one expert layer to CPU for stability:
 
 ```
---n-cpu-moe 35     # 5 expert layers on GPU
+--n-cpu-moe 36     # 4 expert layers on GPU
 --batch-size   8192
 --ubatch-size  3072
 ```
 
-This gives prefill about 1458 t/s at 16384 prompt tokens, generation about
-45.4 t/s, and about 1.2 GiB of headroom at the 800000-token pool. Both are
-chosen because generation does not depend on `ubatch`: `ubatch 3072` reaches
-the best prefill speed with no cost to decode.
+This gives prefill about 1427 t/s at 16384 prompt tokens, generation about
+44.4 t/s, and about 3.0 GiB of headroom at the 800000-token pool.
+
+`ncmoe 36` came from a live crash rather than this sweep. At `ncmoe 35` with
+`ubatch 3072`, a big prefill burst failed a cuBLAS call under memory
+pressure (a `cudaErrorInvalidValue` in `ggml_cuda_mul_mat_cublas`), which
+killed the service. Raising `ncmoe` one step frees about 0.75 GiB of peak
+VRAM (the grid shows `ncmoe 36 - ubatch 3072` peak 8606 MiB versus 9360 MiB
+at `ncmoe 35`) for about 2 percent less prefill and 2 percent less decode.
+
+If a crash still occurs under load, set `GGML_CUDA_FORCE_MMQ=1` to bypass
+the cuBLAS path for quantized matmuls, or drop `ubatch`, or shrink the
+context pool.
 
 If decode speed matters more than prefill, swap to `ncmoe 32 - ubatch 1024`
 (generation about 48.7 t/s, prefill about 766 t/s, about 1.0 GiB headroom).
-If stability headroom matters most, use `ncmoe 34 - ubatch 1536` (about 2.0
-GiB headroom).
 
 ## Limitations