frank-secondary: 2×114688 ctx (112k per-slot), ubatch 1024

Lower per-slot context from 131,072 (128k) to 114,688 (112k) to free
~1 GiB of KV, which funds the ubatch 512 -> 1024 increase (prefill
throughput). The per-slot window is the product contract, so models.yml
contextWindow follows to 114,688.

- CTX_SIZE: 262144 -> 229376 (2 × 114,688)
- UBATCH_SIZE: 512 -> 1024
- Test 194: per-slot 131072 -> 114688
- KV budget test: 131072 -> 114688, ubatch 512 -> 1024

scripts/frank-secondary.env

@@ -57,23 +57,22 @@ ALIAS=frank-secondary
 BIND_ADDR=0.0.0.0
 LISTEN_PORT=7771
 
-# TWO conversations at Frank's 128k cap (2026-10-06): --ctx-size is the
+# TWO conversations at Frank's 112k cap (2026-10-06 evening): --ctx-size is
-# POOL llama.cpp divides across slots; 262144 / 2 = 131,072 per
+# the POOL llama.cpp divides across slots; 229376 / 2 = 114,688 per
-# conversation. Frank accepted the tight fit explicitly (arithmetic above);
+# conversation. Lowering from 128k (131,072) frees ~1 GiB of KV, which
-# the per-slot window is the product contract, so models.yml stays at
+# funds the ubatch 512 -> 1024 increase (prefill throughput). The
-# 131,072.
+# per-slot window is the product contract, so models.yml follows to
-CTX_SIZE=262144
+# 114,688.
+CTX_SIZE=229376
 PARALLEL=2
 
 # Compute-buffer geometry. Ubatch is a prefill knob (decode never moves);
-# on the 9B the 1024->2048 rung measured +4.6 % prefill for 616 MiB, and
+# the 1024 rung was the first A/B back once the per-slot context dropped
-# Frank's 2026-10-06 order spends the other direction: ubatch 512 claws
+# to 112k (2026-10-06 evening) and freed ~1 GiB of KV. The pair is a
-# back ~0.3 GiB toward the second slot at some prefill cost. The pair is a
+# measurement, not a law (loud failure at graph capture if it does not
-# measurement, not a law: ubatch 1024 is the first A/B back if a future
-# boot lands with headroom (loud failure at graph capture if it does not
 # fit).
 BATCH_SIZE=4096
-UBATCH_SIZE=512
+UBATCH_SIZE=1024
 
 # The KV contract for this swap (Frank, 2026-10-05): q8_0 K + q8_0 V —
 # llama.cpp's FP8 class. The flags stay explicit so the choice is visible in

tests/frank_secondary_unit.bats

@@ -103,7 +103,7 @@ argv_value() {
     done
 }
 
-@test "the pool is exactly the native 262144 window, so each of the two conversations is 131072" {
+@test "the pool is 229376, so each of the two conversations is 114688" {
     serve_argv
     [ "$status" -eq 0 ]
     [ "$(argv_value --alias)" = "frank-secondary" ]
@@ -113,13 +113,14 @@ argv_value() {
     [ "$(argv_value --n-gpu-layers)" -gt 0 ]
     # llama.cpp divides --ctx-size across slots and a mamba hybrid cannot
     # share one cache, so the observable contract is the per-slot window.
-    # 2026-10-06 (Frank): drop MTP, drop ubatch, and take TWO conversations
+    # 2026-10-06 evening (Frank): lower per-slot to 114,688 (112k) to fund
-    # at 131,072 — pool 262144, parallel 2, per-slot 131072 (the cap).
+    # the ubatch 512 -> 1024 increase — pool 229376, parallel 2, per-slot
+    # 114688 (the cap).
     [ "$(argv_value --parallel)" -eq 2 ]
     ctx="$(argv_value --ctx-size)"
     par="$(argv_value --parallel)"
     [ $((ctx % par)) -eq 0 ]
-    [ $((ctx / par)) -eq 131072 ]
+    [ $((ctx / par)) -eq 114688 ]
 }
 
 @test "the KV contract is q8_0 both halves (FP8 class), flash-attn on" {
@@ -136,7 +137,7 @@ argv_value() {
     [ "$(argv_value --flash-attn)" = "on" ]
 }
 
-@test "the KV budget fits two 131072 slots at q8_0 and forbids a third" {
+@test "the KV budget fits two 114688 slots at q8_0 and forbids a third" {
     # Rate, from the RCO-IQ3_XXS GGUF header (same qwen35 architecture):
     # SIXTEEN full-attention layers (block_count 65 = 64 real layers + the
     # blk.64 MTP head, full_attention_interval 4) x 4 kv heads x 256 dim x
@@ -145,15 +146,15 @@ argv_value() {
     # RCO-IQ3_XXS weights 10,148 MiB (10,442,827,392 B) — 1.56 GiB LIGHTER
     # than the IQ3_S-mtp predecessor — + vision tower ~1.4 GiB (931,145,888
     # B BF16 mmproj at the measured ~1.45x cost law) + compute buffers at
-    # ubatch 512 + CUDA graphs + TWO DeltaNet states (~150 MiB each) + the
+    # ubatch 1024 + CUDA graphs + TWO DeltaNet states (~150 MiB each) + the
-    # MTP draft ctx => 14 GiB budgeted. Two 131,072 slots are the
+    # MTP draft ctx => 14 GiB budgeted. Two 114,688 slots are the
-    # 2026-10-06 shape (the lighter quant funds the draft ctx plus the
+    # 2026-10-06 evening shape (the lighter per-slot funds the ubatch
-    # second slot); a THIRD slot must not fit. Arithmetic, not a boot log:
+    # increase); a THIRD slot must not fit. Arithmetic, not a boot log:
     # change CTX_SIZE and this bites before the deploy does; the boot's
     # nvidia-smi free is the measured truth.
     ctx="$(sed -nE 's/^CTX_SIZE=([0-9]+)$/\1/p' "$ENV_FILE")"
     kv_gib=$((ctx * 37 / 1048576))
-    third_slot_kv=$(( (ctx + 131072) * 37 / 1048576 ))
+    third_slot_kv=$(( (ctx + 114688) * 37 / 1048576 ))
     [ $((kv_gib + 14)) -le 24 ]
     [ $((third_slot_kv + 14)) -gt 24 ]
 }