chore: cleanup recipes

recipes/deepseek-v4-flash/recipe.toml

@@ -14,7 +14,7 @@ shm_size = "64gb"
 args = [
     "--kv-cache-dtype", "fp8",
     "--block-size", "256",
-    "--max-model-len", "1000000",
+    "--max-model-len", "1048576",
     "--max-num-seqs", "6",
     "--max-num-batched-tokens", "8192",
     "--gpu-memory-utilization", "0.82",
@@ -29,7 +29,7 @@ args = [
     "--default-chat-template-kwargs.reasoning_effort", "high",
     "--enable-flashinfer-autotune",
     "--trust-remote-code",
-    "--served-model-name", "deepseek-v4-flash",
+    "--served-model-name", "deepseek-v4-flash", "default",
 ]
 
 [env]

recipes/glm51/recipe.toml

@@ -2,7 +2,6 @@ engine = "vllm"
 name = "GLM-5.1 NVFP8"
 docker_image = "vllm/vllm-openai:glm51-cu130"
 model = "lukealonso/GLM-5.1-NVFP4"
-port = 30000
 
 required_nodes = 4                  # 0 = любое, иначе строгое совпадение
 pipeline_parallel_size = 0          # 0 = авто (равно nnodes), иначе кастомное значение
@@ -25,7 +24,7 @@ args = [
     "--max-num-seqs", "48",
     "--gpu-memory-utilization", "0.88",
     "--trust-remote-code",
-    "--host", "0.0.0.0"
+    "--served-model-name", "glm51", "default",
 ]
 
 [env]

recipes/minimax-m3/recipe.toml

@@ -2,7 +2,6 @@ engine = "sglang"
 name = "Minimax-M3-v0 NVFP4"
 docker_image = "scitrera/dgx-spark-sglang-mm:v0"
 model = "sparkarena/Minimax-M3-v0-NVFP4"
-port = 30000
 required_nodes = 4
 tensor_parallel_size = 4
 distributed_backend = "mp"
@@ -11,7 +10,7 @@ entrypoint = ["sglang", "serve"]
 
 args = [
     "--mem-fraction-static", "0.8",
-    "--context-length", "200000",
+    "--context-length", "524288",
     "--chunked-prefill-size", "8192",
     "--fp4-gemm-backend", "cutlass",
     "--load-format", "instanttensor",
@@ -22,7 +21,7 @@ args = [
     "--trust-remote-code",
     "--cuda-graph-max-bs", "16",
     "--cuda-graph-bs", "1", "2", "4", "8", "12", "16",
-    "--host", "0.0.0.0",
+    "--served-model-name", "minimax-m3", "default",
 ]
 
 [env]