Switch frank-default-small to Ornith Q6_K

scripts/frank-default-small.env

@@ -1,34 +1,37 @@
-# Tunables for frank-default-small (Ornith-1.5-35B-A3B CRACK Q8_0, MTP MoE,
+# Tunables for frank-default-small (Ornith-1.5-35B-A3B CRACK Q6_K, MTP MoE,
 # Dankserver, RTX 3090).
 #
 # Runtime: upstream llama.cpp llama-server (b10121). RTX 3090 24 GB; Ryzen
 # 7 7700X; 62 GB RAM. MoE expert weights pinned to CPU with an explicit
 # --override-tensor regex (not --n-cpu-moe). Replaced the HauhauCS
-# Qwen3.6-35B-A3B model on 2026-08-24.
+# Qwen3.6-35B-A3B model on 2026-08-24. Switched from CRACK Q8_0 to Q6_K on
+# 2026-08-25.
 #
 # WARNING: do not name any variable TEMP, HOST, or GPU_MAX_HEAP_SIZE. The
 # ROCm HIP runtime reads those bare env vars (see the 2026-08-13 TEMP
 # incident). This box is NVIDIA/CUDA, but the rule keeps the env portable.
 
-# Model files (Q8_0 CRACK GGUF + f16 mmproj vision encoder + MTP head bundled
+# Model files (Q6_K CRACK GGUF + f16 mmproj vision encoder + MTP head bundled
 # in the GGUF). Dealign2026-08-24.
 MODEL_DIR=/home/dank/models/Ornith-1.5-35B-A3B-CRACK
-MODEL_FILE=Ornith-1.5-35B-A3B-CRACK-Q8_0.gguf
+MODEL_FILE=Ornith-1.5-35B-A3B-CRACK-Q6_K.gguf
 MMPROJ_FILE=mmproj-Ornith-1.5-35B-A3B-f16.gguf
 # Upstream llama.cpp binary (b10121).
 LLAMA_SERVER=/home/dank/llama.cpp/build/bin/llama-server
 
-# Qwen-Sharp chat template (tracked in the repo, staged on the host).
+# Qwen-Sharp chat template from peculiar-ragdoll (tracked in the repo,
+# staged on the host).
+# Source: https://huggingface.co/peculiar-ragdoll/Qwen-Sharp-Chat-Templates
 CHAT_TEMPLATE_FILE=/home/dank/serverAppConfigs/scripts/chat-template.jinja
 
 ALIAS=frank-default-small
 BIND_ADDR=0.0.0.0
 LISTEN_PORT=7770
 
-# 3 parallel slots x 229376 ctx each (262144 lowered by 32k per stream for
-# cache headroom) = 688128 total context pool. The per-slot 229376 matches
-# the per-slot context advertised in models.json (229376).
-CTX_SIZE=688128
+# 3 parallel slots x 196608 ctx each (262144 lowered by 64k per stream for
+# cache and checkpoint headroom) = 589824 total context pool. The per-slot
+# 196608 matches the per-slot context advertised in models.json (196608).
+CTX_SIZE=589824
 PARALLEL=3
 
 # Spec: MoE model with a bundled MTP head -> native draft-mtp, n_max 4
@@ -38,11 +41,12 @@ SPEC_DRAFT_N_MAX=4
 SPEC_DRAFT_N_MIN=0
 SPEC_DRAFT_P_MIN=0.0
 
-# Expert offload via explicit tensor override (replaces --n-cpu-moe). Only
-# the MoE FFN exps tensors move to CPU; everything else up to --gpu-layers
-# stays on the 24 GB card. The regex is the same one the built-in --cpu-moe
-# expands. Buffer type "CPU" is the CPU buft name on this host.
-OVERRIDE_EXPERT='\.ffn_(up|down|gate|gate_up)_(ch|)exps'
+# Expert offload via explicit tensor override (replaces --n-cpu-moe). The Q6_K
+# layer audit found 630 MiB of expert tensors per block. Keep the four
+# output-adjacent blocks 37-40 on CUDA, including the bundled MTP block 40.
+# Move expert tensors from blocks 0-36 to CPU. All non-expert tensors stay on
+# the 24 GB card. Buffer type "CPU" is the CPU buft name on this host.
+OVERRIDE_EXPERT='blk\.(0|1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36)\.ffn_(up|down|gate|gate_up)_(ch|)exps'
 OVERRIDE_BUFT=CPU
 N_GPU_LAYERS=all