SGLang RadixAttention: Structured Output + JSON-Mode + Multi-Branch Caching

# === SGLang Structured Output Lab ===
import sglang as sgl
 
# Backend: Llama 3.1 8B
sgl.set_default_backend(sgl.RuntimeEndpoint("http://localhost:30000"))
 
# JSON schema-constrained generation
@sgl.function
def extract_person(s, text):
    s += sgl.user(f"Aşağıdaki cümleden kişi bilgilerini JSON olarak çıkar: {text}")
    s += sgl.assistant(sgl.gen("output",
        regex=r'\\{\\"name\\": \\"[a-zA-ZşıçöüğŞIÇÖÜĞ ]+\\", \\"age\\": \\d+\\}',
        max_tokens=100))
 
# Run
result = extract_person.run(text="Ahmet 35 yaşında bir mühendistir.")
import json
data = json.loads(result["output"])
print(data)  # {"name": "Ahmet", "age": 35}
 
# Server başlatma
# python -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct \
#     --port 30000 --enable-radix-cache