Hl Build LLM App

Skill: Build LLM Chat Application

Build a complete LLM text generation app running on Hailo-10H.

When This Skill Is Loaded

• User wants text generation or chat without vision
• User mentions: LLM, chat, chatbot, text generation, Q&A
• User needs on-device language model inference (NOT vision — that's VLM)

Reference Implementation

Study hailo_apps/python/gen_ai_apps/simple_llm_chat/simple_llm_chat.py — the canonical LLM app.

Build Process

Step 1: Create App Directory

Create the app directory:

hailo_apps/python/<type>/<app_name>/
├── app.yaml              # App manifest (type: gen_ai)
├── run.sh                # Launch wrapper
├── __init__.py
├── <app_name>.py         # Main app
└── README.md             # Usage documentation (REQUIRED — never skip)

Create app.yaml with type: gen_ai and run.sh wrapper. Do NOT register in defines.py or resources_config.yaml.

Step 2: Build Main App

import signal
import sys

from hailo_platform import VDevice
from hailo_platform.genai import LLM

from hailo_apps.python.core.common.hailo_logger import get_logger
from hailo_apps.python.core.common.core import resolve_hef_path
from hailo_apps.python.core.common.parser import get_standalone_parser
from hailo_apps.python.core.common.defines import (
    SHARED_VDEVICE_GROUP_ID,
    HAILO10H_ARCH,
)

logger = get_logger(__name__)

APP_NAME = "my_llm_app"

SYSTEM_PROMPT = "You are a helpful assistant."


def format_prompt(system_prompt, user_text):
    return [
        {"role": "system", "content": [{"type": "text", "text": system_prompt}]},
        {"role": "user", "content": [{"type": "text", "text": user_text}]},
    ]


def main():
    parser = get_standalone_parser()
    parser.add_argument("--max-tokens", type=int, default=200, help="Max tokens to generate")
    parser.add_argument("--temperature", type=float, default=0.1, help="Sampling temperature")
    parser.add_argument("--system-prompt", type=str, default=SYSTEM_PROMPT, help="System prompt")
    args = parser.parse_args()

    # Signal handling
    running = True
    def signal_handler(sig, frame):
        nonlocal running
        running = False
        print("\nShutting down...")
    signal.signal(signal.SIGINT, signal_handler)

    # Device and model
    params = VDevice.create_params()
    params.group_id = SHARED_VDEVICE_GROUP_ID
    vdevice = VDevice(params)

    hef_path = resolve_hef_path(args.hef_path, APP_NAME, arch=HAILO10H_ARCH)
    llm = LLM(vdevice, str(hef_path))

    logger.info("LLM loaded: %s", hef_path)
    print(f"Chat started. Type 'quit' to exit.\n")

    try:
        while running:
            try:
                user_input = input("You: ").strip()
            except EOFError:
                break
            if not user_input or user_input.lower() in ("quit", "exit", "q"):
                break

            prompt = format_prompt(args.system_prompt, user_input)
            response = llm.generate_all(
                prompt=prompt,
                temperature=args.temperature,
                seed=42,
                max_generated_tokens=args.max_tokens,
            )
            print(f"Assistant: {response}\n")
            llm.clear_context()
    finally:
        llm.release()
        vdevice.release()
        logger.info("Cleanup complete")


if __name__ == "__main__":
    main()

Step 4: Validate

python3 .github/scripts/validate_app.py hailo_apps/python/gen_ai_apps/my_llm_app --smoke-test

Critical Conventions

1. Hailo-10H only: Use HAILO10H_ARCH — LLM is not available on Hailo-8/8L
2. VDevice sharing: params.group_id = SHARED_VDEVICE_GROUP_ID
3. Prompt format: [{"role": "...", "content": [{"type": "text", "text": "..."}]}]
4. Clear context: Always llm.clear_context() after each generation
5. Cleanup order: llm.clear_context() → llm.release() → vdevice.release() in finally
6. End token: Filter <|im_end|> from streaming output
7. Token streaming: with llm.generate(...) as gen: for chunk in gen: for real-time output
8. HEF resolution: resolve_hef_path(path, APP_NAME, arch=HAILO10H_ARCH)

Streaming Pattern (Alternative)

print("Assistant: ", end="", flush=True)
with llm.generate(prompt=prompt, temperature=0.1, max_generated_tokens=200) as gen:
    for chunk in gen:
        if chunk != "<|im_end|>":
            print(chunk, end="", flush=True)
print()
llm.clear_context()