feat: initial ray-serve-apps PyPI package

Implements ADR-0024: Ray Repository Structure

- Ray Serve deployments for GPU-shared AI inference
- Published as PyPI package for dynamic code loading
- Deployments: LLM, embeddings, reranker, whisper, TTS
- CI/CD workflow publishes to Gitea PyPI on push to main

Extracted from kuberay-images repo per ADR-0024

ray_serve/serve_llm.py

@@ -0,0 +1,108 @@
+"""
+Ray Serve deployment for vLLM with OpenAI-compatible API.
+Runs on: khelben (Strix Halo 64GB, ROCm)
+"""
+
+import os
+import time
+import uuid
+from typing import Any
+
+from ray import serve
+
+
+@serve.deployment(name="LLMDeployment", num_replicas=1)
+class LLMDeployment:
+    def __init__(self):
+        from vllm import LLM, SamplingParams
+
+        self.model_id = os.environ.get("MODEL_ID", "meta-llama/Llama-3.1-70B-Instruct")
+        self.max_model_len = int(os.environ.get("MAX_MODEL_LEN", "8192"))
+        self.gpu_memory_utilization = float(os.environ.get("GPU_MEMORY_UTILIZATION", "0.9"))
+
+        print(f"Loading vLLM model: {self.model_id}")
+        print(f"Max model length: {self.max_model_len}")
+        print(f"GPU memory utilization: {self.gpu_memory_utilization}")
+
+        self.llm = LLM(
+            model=self.model_id,
+            max_model_len=self.max_model_len,
+            gpu_memory_utilization=self.gpu_memory_utilization,
+            trust_remote_code=True,
+        )
+        self.SamplingParams = SamplingParams
+        print(f"Model {self.model_id} loaded successfully")
+
+    async def __call__(self, request: dict[str, Any]) -> dict[str, Any]:
+        """
+        Handle OpenAI-compatible chat completion requests.
+
+        Expected request format:
+        {
+            "model": "model-name",
+            "messages": [{"role": "user", "content": "Hello"}],
+            "temperature": 0.7,
+            "max_tokens": 256,
+            "top_p": 1.0,
+            "stream": false
+        }
+        """
+        messages = request.get("messages", [])
+        temperature = request.get("temperature", 0.7)
+        max_tokens = request.get("max_tokens", 256)
+        top_p = request.get("top_p", 1.0)
+        stop = request.get("stop")
+
+        # Convert messages to prompt
+        prompt = self._format_messages(messages)
+
+        sampling_params = self.SamplingParams(
+            temperature=temperature,
+            max_tokens=max_tokens,
+            top_p=top_p,
+            stop=stop,
+        )
+
+        outputs = self.llm.generate([prompt], sampling_params)
+        generated_text = outputs[0].outputs[0].text
+
+        # Return OpenAI-compatible response
+        return {
+            "id": f"chatcmpl-{uuid.uuid4().hex[:8]}",
+            "object": "chat.completion",
+            "created": int(time.time()),
+            "model": self.model_id,
+            "choices": [
+                {
+                    "index": 0,
+                    "message": {
+                        "role": "assistant",
+                        "content": generated_text,
+                    },
+                    "finish_reason": "stop",
+                }
+            ],
+            "usage": {
+                "prompt_tokens": len(prompt.split()),
+                "completion_tokens": len(generated_text.split()),
+                "total_tokens": len(prompt.split()) + len(generated_text.split()),
+            },
+        }
+
+    def _format_messages(self, messages: list[dict[str, str]]) -> str:
+        """Format chat messages into a prompt string."""
+        formatted = ""
+        for msg in messages:
+            role = msg.get("role", "user")
+            content = msg.get("content", "")
+            if role == "system":
+                formatted += f"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{content}<|eot_id|>"
+            elif role == "user":
+                formatted += f"<|start_header_id|>user<|end_header_id|>\n\n{content}<|eot_id|>"
+            elif role == "assistant":
+                formatted += f"<|start_header_id|>assistant<|end_header_id|>\n\n{content}<|eot_id|>"
+        formatted += "<|start_header_id|>assistant<|end_header_id|>\n\n"
+        return formatted
+
+
+app = LLMDeployment.bind()