added minimax 2.1 w/ sglang and minimax/qwen3 reasoning parser (#535)

Co-authored-by: dsingal <dsingal@dsingals-MacBook-Pro.local>

Author: dsingal0

minimax/minimax_m2_1/config.yaml

@@ -0,0 +1,51 @@
+base_image:
+  image: lmsysorg/sglang:nightly-dev-20260126-48f4340b
+docker_server:
+  liveness_endpoint: /health_generate
+  predict_endpoint: /v1/chat/completions
+  readiness_endpoint: /health_generate
+  server_port: 8000
+  start_command: sh -c "truss-transfer-cli && find /app/model_cache/checkpoint -type f -print0 | xargs -0 -P 0 -I {} dd if={} of=/dev/null bs=4M && python3 -m sglang.launch_server --model-path /app/model_cache/checkpoint --tp-size 8 --ep-size 8 --tool-call-parser minimax-m2 --trust-remote-code --host 0.0.0.0 --reasoning-parser minimax --port 8000 --mem-fraction-static 0.85"
+#environment_variables:
+#  SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN: 1
+model_cache:
+- allow_patterns:
+  - '*.json'
+  - '*.safetensors'
+  - '*.txt'
+  - '*.model'
+  - '*.py'
+  - '*.jinja'
+  repo_id: MiniMaxAI/MiniMax-M2.1
+  revision: 927ea2b64008fe4a1e31a4e107a6b75916b3b44a
+  use_volume: true
+  volume_folder: checkpoint
+model_metadata:
+  example_model_input:
+    max_tokens: 4096
+    messages:
+    - content: You are a helpful assistant.
+      role: system
+    - content: Who won the world series in 2020?
+      role: user
+    model: MiniMaxAI/MiniMax-M2
+    stream: true
+    temperature: 0.6
+  model_name: MiniMax-M2
+  tags:
+  - openai-compatible
+model_name: minimax
+resources:
+  accelerator: H100:8
+  cpu: '1'
+  memory: 2Gi
+  use_gpu: true
+runtime:
+  health_checks:
+    restart_check_delay_seconds: 1200
+    restart_threshold_seconds: 600
+    stop_traffic_threshold_seconds: 1800
+  is_websocket_endpoint: false
+  predict_concurrency: 32
+  transport:
+    kind: http