llama-swap: add artemis-31b-v1i (BeaverAI Gemma 4 fine-tune, Q6_K)

stacks/llama-swap/conf/config.yaml

@@ -37,6 +37,11 @@
 #               imatrix Q6_K, ~6.5GB). New JUDGE / EVAL section. Pre-pulled
 #               via playbooks/pull-hf-repo.yaml (canonical replacement
 #               for ad-hoc huggingface_hub.snapshot_download calls).
+#   2026-05-25: Added artemis-31b-v1i (BeaverAI Artemis-31B-v1i Gemma 4
+#               dense fine-tune, Q6_K ~28.6GB) via -hf syntax. Pre-pulled
+#               via playbooks/pull-hf-repo.yaml. Sampler defaults mirrored
+#               from gemma4-31b-dense (temp 1.0, top-p 0.95, top-k 64,
+#               repeat_penalty 1.0); ctx 32K starting per Gemma 4 KB.
 # ============================================================================
 
 # Default 1200 seconds (20 min) to wait for model to be available to load.
@@ -313,6 +318,24 @@ models:
       --repeat-penalty 1.0
       --chat-template-kwargs '{"enable_thinking":true}'
 
+  "artemis-31b-v1i":
+    name: "Artemis 31B v1i Q6_K"
+    description: "BeaverAI Artemis-31B-v1i — Gemma 4 31B dense fine-tune, Q6_K (~28.6GB). 256K context."
+    ttl: 600
+    cmd: |
+      /app/llama-server
+      --context-shift
+      -hf BeaverAI/Artemis-31B-v1i-GGUF:Q6_K
+      --port ${PORT}
+      --n-gpu-layers 999
+      --ctx-size 32768
+      --flash-attn on
+      --temp 1.0
+      --top-p 0.95
+      --top-k 64
+      --repeat-penalty 1.0
+      --chat-template-kwargs '{"enable_thinking":true}'
+
   # ==========================================================================
   # GLM MODELS
   # ==========================================================================