refactor(llama-swap): reorganize models by GPU hardware section

modules/nixos/services/llama-swap/config.nix

@@ -8,122 +8,9 @@ in
 {
   healthCheckTimeout = 500;
   models = {
-    # https://huggingface.co/unsloth/GLM-4.7-Flash-GGUF/tree/main
-    "glm-4.7-flash" = {
-      name = "GLM 4.7 Flash (30B) - Thinking";
-      macros.ctx = "131072";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/GLM/GLM-4.7-Flash-UD-Q6_K_XL.gguf \
-          -c ''${ctx} \
-          --temp 0.7 \
-          --top-p 1.0 \
-          --min-p 0.01 \
-          --repeat-penalty 1.0 \
-          -fit off \
-          -ts 70,30
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF/tree/main
-    "devstral-small-2-instruct" = {
-      name = "Devstral Small 2 (24B) - Instruct";
-      macros.ctx = "131072";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Devstral/Devstral-Small-2-24B-Instruct-2512-UD-Q6_K_XL.gguf \
-          --temp 0.15 \
-          -c ''${ctx} \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -fit off \
-          -ts 75,25
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
-    # https://huggingface.co/mradermacher/gpt-oss-20b-heretic-v2-i1-GGUF/tree/main
-    "gpt-oss-20b-thinking" = {
-      name = "GPT OSS (20B) - Thinking";
-      macros.ctx = "131072";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/GPT-OSS/gpt-oss-20b-heretic-v2.i1-MXFP4_MOE.gguf \
-          -c ''${ctx} \
-          --temp 1.0 \
-          --top-p 1.0 \
-          --top-k 40 \
-          -dev CUDA0
-      '';
-      metadata = {
-        type = [ "text-generation" ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/tree/main
-    "qwen3-coder-next-80b-instruct" = {
-      name = "Qwen3 Coder Next (80B) - Instruct";
-      macros.ctx = "131072";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-Coder-Next-UD-Q4_K_XL.gguf \
-          -c ''${ctx} \
-          --temp 1.0 \
-          --top-p 0.95 \
-          --min-p 0.01 \
-          --top-k 40 \
-          -fit off \
-          -ncmoe 19 \
-          -ts 78,22
-      '';
-
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
-    # https://huggingface.co/AesSedai/Qwen3.5-35B-A3B-GGUF/tree/main
-    "qwen3.5-35b-thinking" = {
-      name = "Qwen3.5 (35B) - Thinking";
-      macros.ctx = "262144";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3.5/Qwen3.5-35B-A3B-IQ4_XS-00001-of-00002.gguf \
-          -c ''${ctx} \
-          --temp 0.6 \
-          --top-p 0.95 \
-          --top-k 20 \
-          --min-p 0.00 \
-          -dev CUDA0 \
-          -fit off
-      '';
-      # --chat-template-kwargs "{\"enable_thinking\": false}"
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
+    # ---------------------------------------
+    # -------------- RTX 3090 ---------------
+    # ---------------------------------------
 
     # https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/tree/main
     "qwen3.6-35b-thinking" = {
@@ -152,32 +39,6 @@ in
       };
     };
 
-    # https://huggingface.co/bartowski/Qwen_Qwen3.5-27B-GGUF/tree/main
-    "qwen3.5-27b-thinking" = {
-      name = "Qwen3.5 (27B) - Thinking";
-      macros.ctx = "196608";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3.5/Qwen_Qwen3.5-27B-IQ4_XS.gguf \
-          -c ''${ctx} \
-          --temp 0.6 \
-          --top-p 0.95 \
-          --top-k 20 \
-          --min-p 0.00 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -dev CUDA0 \
-          -fit off
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
     # https://huggingface.co/unsloth/Qwen3.6-27B-GGUF/tree/main
     "qwen3.6-27b-thinking" = {
       name = "Qwen3.6 (27B) - Thinking";
@@ -217,150 +78,6 @@ in
       };
     };
 
-    # https://huggingface.co/unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF/tree/main
-    "qwen3-30b-2507-instruct" = {
-      name = "Qwen3 2507 (30B) - Instruct";
-      macros.ctx = "262144";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-30B-A3B-Instruct-2507-Q4_K_M.gguf \
-          -c ''${ctx} \
-          --temp 0.7 \
-          --min-p 0.0 \
-          --top-p 0.8 \
-          --top-k 20 \
-          --presence-penalty 1.0 \
-          --repeat-penalty 1.0 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -ts 70,30 \
-          -fit off
-      '';
-      metadata = {
-        type = [ "text-generation" ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF/tree/main
-    "qwen3-coder-30b-instruct" = {
-      name = "Qwen3 Coder (30B) - Instruct";
-      macros.ctx = "131072";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-Coder-30B-A3B-Instruct-UD-Q6_K_XL.gguf \
-          -c ''${ctx} \
-          --temp 0.7 \
-          --min-p 0.0 \
-          --top-p 0.8 \
-          --top-k 20 \
-          --repeat-penalty 1.05 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -ts 70,30 \
-          -fit off
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Qwen3-30B-A3B-Thinking-2507-GGUF/tree/main
-    "qwen3-30b-2507-thinking" = {
-      name = "Qwen3 2507 (30B) - Thinking";
-      macros.ctx = "262144";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-30B-A3B-Thinking-2507-UD-Q4_K_XL.gguf \
-          -c ''${ctx} \
-          --temp 0.6 \
-          --min-p 0.0 \
-          --top-p 0.95 \
-          --top-k 20 \
-          --presence-penalty 1.0 \
-          --repeat-penalty 1.0 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -ts 70,30 \
-          -fit off
-      '';
-      metadata = {
-        type = [ "text-generation" ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Nemotron-3-Nano-30B-A3B-GGUF/tree/main
-    "nemotron-3-nano-30b-thinking" = {
-      name = "Nemotron 3 Nano (30B) - Thinking";
-      macros.ctx = "1048576";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Nemotron/Nemotron-3-Nano-30B-A3B-UD-Q4_K_XL.gguf \
-          -c ''${ctx} \
-          --temp 1.1 \
-          --top-p 0.95 \
-          -fit off
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Qwen3-VL-8B-Instruct-GGUF/tree/main
-    "qwen3-8b-vision" = {
-      name = "Qwen3 Vision (8B) - Thinking";
-      macros.ctx = "65536";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf \
-          --mmproj /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL_mmproj-F16.gguf \
-          -c ''${ctx} \
-          --temp 0.7 \
-          --min-p 0.0 \
-          --top-p 0.8 \
-          --top-k 20 \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -fit off \
-          -dev CUDA1
-      '';
-      metadata = {
-        type = [
-          "text-generation"
-          "vision"
-        ];
-      };
-    };
-
-    # https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF/tree/main
-    "qwen3-4b-2507-instruct" = {
-      name = "Qwen3 2507 (4B) - Instruct";
-      macros.ctx = "98304";
-      cmd = ''
-        ${llama-cpp}/bin/llama-server \
-          --port ''${PORT} \
-          -m /mnt/ssd/Models/Qwen3/Qwen3-4B-Instruct-2507-Q4_K_M.gguf \
-          -c ''${ctx} \
-          -fit off \
-          -ctk q8_0 \
-          -ctv q8_0 \
-          -dev CUDA1
-      '';
-      metadata = {
-        type = [ "text-generation" ];
-      };
-    };
-
     # https://github.com/noonghunna/club-3090/tree/v0.20-experimental/models/qwen3.6-27b/vllm
     # Long-text variant - v0.20 experimental single-3090 profile, text-only (no vision)
     # TurboQuant 3-bit KV + MTP n=3 + workspace-lock sidecar.
@@ -451,84 +168,6 @@ in
       };
     };
 
-    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
-    # Tools-text variant - 75K context, text-only (no vision)
-    # fp8_e5m2 KV + MTP n=3. This is the repo's validated long-context
-    # tool-calling profile and should be more stable than TurboQuant 128K.
-    "vllm-qwen3.6-27b-tools-text" = {
-      name = "vLLM Qwen3.6 (27B) - Tools Text";
-      macros.ctx = "75000";
-      proxy = "http://127.0.0.1:\${PORT}";
-      cmd =
-        let
-          vllmCmd = ''
-            set -e; pip install xxhash pandas scipy -q;
-            python3 -m vllm._genesis.patches.apply_all;
-            python3 /patches/patch_tolist_cudagraph.py;
-            python3 /patches/patch_timings_07351e088.py;
-            exec vllm serve
-            --served-model-name ''${MODEL_ID}
-            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
-            --quantization auto_round
-            --dtype float16
-            --tensor-parallel-size 1
-            --max-model-len ''${ctx}
-            --gpu-memory-utilization 0.97
-            --max-num-seqs 1
-            --max-num-batched-tokens 2048
-            --kv-cache-dtype fp8_e5m2
-            --language-model-only
-            --trust-remote-code
-            --reasoning-parser qwen3
-            --enable-auto-tool-choice
-            --tool-call-parser qwen3_coder
-            --enable-prefix-caching
-            --enable-chunked-prefill
-            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
-            --host 0.0.0.0
-            --port 8000
-          '';
-          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
-        in
-        ''
-          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
-            --name ''${MODEL_ID} \
-            --ipc=host \
-            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
-            -e NCCL_CUMEM_ENABLE=0 \
-            -e NCCL_P2P_DISABLE=1 \
-            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1 \
-            -e VLLM_NO_USAGE_STATS=1 \
-            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
-            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
-            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
-            -e OMP_NUM_THREADS=1 \
-            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
-            -e CUDA_VISIBLE_DEVICES=0 \
-            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
-            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
-            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
-            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
-            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
-            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
-            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
-            -v /mnt/ssd/vLLM/Patches/patch_tolist_cudagraph.py:/patches/patch_tolist_cudagraph.py:ro \
-            -v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
-            -p ''${PORT}:8000 \
-            --entrypoint /bin/bash \
-            vllm/vllm-openai:nightly-07351e0883470724dd5a7e9730ed10e01fc99d08 \
-            -c "${vllmCmdFlat}"
-        '';
-      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
-
-      metadata = {
-        type = [
-          "text-generation"
-          "coding"
-        ];
-      };
-    };
-
     # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
     # Long-vision variant - 140K context with vision tower active
     # TurboQuant 3-bit KV + MTP n=3 + PN12/P104 cliff-closure stack
@@ -620,6 +259,185 @@ in
       };
     };
 
+    # https://github.com/noonghunna/club-3090/tree/master/models/qwen3.6-27b/vllm
+    # Tools-text variant - 75K context, text-only (no vision)
+    # fp8_e5m2 KV + MTP n=3. This is the repo's validated long-context
+    # tool-calling profile and should be more stable than TurboQuant 128K.
+    "vllm-qwen3.6-27b-tools-text" = {
+      name = "vLLM Qwen3.6 (27B) - Tools Text";
+      macros.ctx = "75000";
+      proxy = "http://127.0.0.1:\${PORT}";
+      cmd =
+        let
+          vllmCmd = ''
+            set -e; pip install xxhash pandas scipy -q;
+            python3 -m vllm._genesis.patches.apply_all;
+            python3 /patches/patch_tolist_cudagraph.py;
+            python3 /patches/patch_timings_07351e088.py;
+            exec vllm serve
+            --served-model-name ''${MODEL_ID}
+            --model /root/.cache/huggingface/qwen3.6-27b-autoround-int4
+            --quantization auto_round
+            --dtype float16
+            --tensor-parallel-size 1
+            --max-model-len ''${ctx}
+            --gpu-memory-utilization 0.97
+            --max-num-seqs 1
+            --max-num-batched-tokens 2048
+            --kv-cache-dtype fp8_e5m2
+            --language-model-only
+            --trust-remote-code
+            --reasoning-parser qwen3
+            --enable-auto-tool-choice
+            --tool-call-parser qwen3_coder
+            --enable-prefix-caching
+            --enable-chunked-prefill
+            --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":3}'
+            --host 0.0.0.0
+            --port 8000
+          '';
+          vllmCmdFlat = builtins.replaceStrings [ "\n" ] [ " " ] vllmCmd;
+        in
+        ''
+          ${pkgs.docker}/bin/docker run --rm --device=nvidia.com/gpu=all \
+            --name ''${MODEL_ID} \
+            --ipc=host \
+            -e VLLM_WORKER_MULTIPROC_METHOD=spawn \
+            -e NCCL_CUMEM_ENABLE=0 \
+            -e NCCL_P2P_DISABLE=1 \
+            -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1 \
+            -e VLLM_NO_USAGE_STATS=1 \
+            -e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512 \
+            -e VLLM_FLOAT32_MATMUL_PRECISION=high \
+            -e VLLM_USE_FLASHINFER_SAMPLER=1 \
+            -e OMP_NUM_THREADS=1 \
+            -e CUDA_DEVICE_MAX_CONNECTIONS=8 \
+            -e CUDA_VISIBLE_DEVICES=0 \
+            -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
+            -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
+            -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
+            -e GENESIS_ENABLE_P64_QWEN3CODER_MTP_STREAMING=1 \
+            -e GENESIS_ENABLE_PN8_MTP_DRAFT_ONLINE_QUANT=1 \
+            -v /mnt/ssd/vLLM/Models:/root/.cache/huggingface \
+            -v /mnt/ssd/vLLM/Patches/genesis/vllm/_genesis:/usr/local/lib/python3.12/dist-packages/vllm/_genesis:ro \
+            -v /mnt/ssd/vLLM/Patches/patch_tolist_cudagraph.py:/patches/patch_tolist_cudagraph.py:ro \
+            -v /mnt/ssd/vLLM/Patches/patch_timings_07351e088.py:/patches/patch_timings_07351e088.py:ro \
+            -p ''${PORT}:8000 \
+            --entrypoint /bin/bash \
+            vllm/vllm-openai:nightly-07351e0883470724dd5a7e9730ed10e01fc99d08 \
+            -c "${vllmCmdFlat}"
+        '';
+      cmdStop = "${pkgs.docker}/bin/docker stop \${MODEL_ID}";
+
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
+    # ---------------------------------------
+    # ------------- GTX 1080 Ti -------------
+    # ---------------------------------------
+
+    # https://huggingface.co/unsloth/Qwen3-VL-8B-Instruct-GGUF/tree/main
+    "qwen3-8b-vision" = {
+      name = "Qwen3 Vision (8B) - Thinking";
+      macros.ctx = "60000";
+      env = [ "CUDA_VISIBLE_DEVICES=1" ];
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf \
+          --mmproj /mnt/ssd/Models/Qwen3/Qwen3-VL-8B-Instruct-UD-Q4_K_XL_mmproj-F16.gguf \
+          -c ''${ctx} \
+          --temp 0.7 \
+          --min-p 0.0 \
+          --top-p 0.8 \
+          --top-k 20 \
+          -ctk q8_0 \
+          -ctv q8_0 \
+          -fit off \
+          -dev CUDA0
+      '';
+      metadata = {
+        type = [
+          "text-generation"
+          "vision"
+        ];
+      };
+    };
+
+    # https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF/tree/main
+    "qwen3-4b-2507-instruct" = {
+      name = "Qwen3 2507 (4B) - Instruct";
+      macros.ctx = "98304";
+      env = [ "CUDA_VISIBLE_DEVICES=1" ];
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3/Qwen3-4B-Instruct-2507-Q4_K_M.gguf \
+          -c ''${ctx} \
+          -fit off \
+          -ctk q8_0 \
+          -ctv q8_0 \
+          -dev CUDA0
+      '';
+      metadata = {
+        type = [ "text-generation" ];
+      };
+    };
+
+    # https://huggingface.co/mradermacher/gpt-oss-20b-heretic-v2-i1-GGUF/tree/main
+    "gpt-oss-20b-thinking" = {
+      name = "GPT OSS (20B) - Thinking";
+      macros.ctx = "131072";
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/GPT-OSS/gpt-oss-20b-heretic-v2.i1-MXFP4_MOE.gguf \
+          -c ''${ctx} \
+          --temp 1.0 \
+          --top-p 1.0 \
+          --top-k 40 \
+          -dev CUDA0
+      '';
+      metadata = {
+        type = [ "text-generation" ];
+      };
+    };
+
+    # ---------------------------------------
+    # -------- RTX 3090 + GTX 1080 Ti -------
+    # ---------------------------------------
+
+    # https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/tree/main
+    "qwen3-coder-next-80b-instruct" = {
+      name = "Qwen3 Coder Next (80B) - Instruct";
+      macros.ctx = "131072";
+      cmd = ''
+        ${llama-cpp}/bin/llama-server \
+          --port ''${PORT} \
+          -m /mnt/ssd/Models/Qwen3/Qwen3-Coder-Next-UD-Q4_K_XL.gguf \
+          -c ''${ctx} \
+          --temp 1.0 \
+          --top-p 0.95 \
+          --min-p 0.01 \
+          --top-k 40 \
+          -fit off \
+          -ncmoe 19 \
+          -ts 78,22
+      '';
+
+      metadata = {
+        type = [
+          "text-generation"
+          "coding"
+        ];
+      };
+    };
+
     # ---------------------------------------
     # ---------- Stable Diffussion ----------
     # ---------------------------------------
@@ -717,21 +535,19 @@ in
   # everything.
   matrix = {
     vars = {
-      # CUDA0 Models
-      go = "gpt-oss-20b-thinking";
-      q35a = "qwen3.5-35b-thinking";
-      q36a = "qwen3.6-35b-thinking";
-      q35b = "qwen3.5-27b-thinking";
-      q36b = "qwen3.6-27b-thinking";
+      # --- RTX 3090 Models ---
       vlt = "vllm-qwen3.6-27b-long-text";
       vtt = "vllm-qwen3.6-27b-tools-text";
       vlv = "vllm-qwen3.6-27b-long-vision";
+      go = "gpt-oss-20b-thinking";
+      q36a = "qwen3.6-35b-thinking";
+      q36b = "qwen3.6-27b-thinking";
       zi = "z-image-turbo";
       qie = "qwen-image-edit-2511";
       qi = "qwen-image-2512";
       cr = "chroma-radiance";
 
-      # CUDA1 Models
+      # --- GTX 1080 Ti Models ---
       qv = "qwen3-8b-vision";
       q4 = "qwen3-4b-2507-instruct";
     };
@@ -743,7 +559,7 @@ in
     };
 
     sets = {
-      concurrent = "(go | q35a | q36a | q35b | q36b | vlt | vtt | vlv | zi | qie | qi | cr) & (qv | q4)";
+      concurrent = "(go | q36a | q36b | vlt | vtt | vlv | zi | qie | qi | cr) & (qv | q4)";
     };
   };
 }