Text Generation
Transformers
Safetensors
qwen2
godot
gdscript
qlora
game-development
conversational
text-generation-inference
Instructions to use khursheed/VibeGame-8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use khursheed/VibeGame-8B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="khursheed/VibeGame-8B") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("khursheed/VibeGame-8B") model = AutoModelForCausalLM.from_pretrained("khursheed/VibeGame-8B", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use khursheed/VibeGame-8B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "khursheed/VibeGame-8B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "khursheed/VibeGame-8B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/khursheed/VibeGame-8B
- SGLang
How to use khursheed/VibeGame-8B with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "khursheed/VibeGame-8B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "khursheed/VibeGame-8B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "khursheed/VibeGame-8B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "khursheed/VibeGame-8B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use khursheed/VibeGame-8B with Docker Model Runner:
docker model run hf.co/khursheed/VibeGame-8B
Download code/src/vibegame/validate.py from khursheed/VibeGame-8B: direct link, hf CLI and curl.
- Browser
- Download file 7.79 kB
-
https://huggingface.co/khursheed/VibeGame-8B/resolve/main/code/src/vibegame/validate.py
- Command line
-
hf download hf://khursheed/VibeGame-8B/code/src/vibegame/validate.py
-
curl -L -o validate.py https://huggingface.co/khursheed/VibeGame-8B/resolve/main/code/src/vibegame/validate.py
7.79 kB
| """Evidence-producing import and wall-clock runtime validation in an OS sandbox.""" | |
| from __future__ import annotations | |
| import hashlib | |
| import json | |
| import math | |
| import os | |
| from pathlib import Path | |
| import platform | |
| import re | |
| import shutil | |
| import signal | |
| import subprocess | |
| import time | |
| from .project import ProjectError, project_sha256, write_project | |
| ERROR_PATTERN = re.compile(r"(?im)(?:\b(?:SCRIPT ERROR|Parse Error|Parser Error|Compile Error|ERROR):|Failed to load script|Failed loading resource|Failed to instantiate|Couldn't load|Segmentation fault|CRASH HANDLER)") | |
| def sandbox_command(command: list[str], project: Path) -> tuple[list[str], str]: | |
| """Do not read the user's home/Drive or permit network access in the child.""" | |
| project = project.resolve() | |
| binary = Path(command[0]).resolve() | |
| home = project / ".sandbox-home" | |
| home.mkdir(exist_ok=True) | |
| (home / "tmp").mkdir(exist_ok=True) | |
| if platform.system() == "Darwin": | |
| raise RuntimeError("macOS validation is disabled: restrictive sandbox crashes the Godot dynamic loader on tested macOS 26.3.1. Use Linux bubblewrap; no local Godot process was launched.") | |
| if platform.system() == "Linux": | |
| bwrap = shutil.which("bwrap") | |
| if not bwrap: | |
| raise RuntimeError("OS sandbox unavailable: install bubblewrap; evaluation fails closed") | |
| wrapped = [bwrap, "--unshare-all", "--die-with-parent", "--new-session", "--cap-drop", "ALL", "--tmpfs", "/proc", "--dev", "/dev", "--tmpfs", "/tmp"] | |
| for item in ("/usr", "/lib", "/lib64", "/bin", "/sbin"): | |
| if Path(item).exists(): | |
| wrapped += ["--ro-bind", item, item] | |
| for item in ("/etc/ld.so.cache", "/etc/fonts"): | |
| if Path(item).exists(): | |
| wrapped += ["--ro-bind", item, item] | |
| wrapped += ["--ro-bind", str(binary.parent), str(binary.parent), "--bind", str(project), str(project), "--chdir", str(project), "--", *command] | |
| return wrapped, "Linux bubblewrap: unshared namespaces, no network, restricted mounts" | |
| raise RuntimeError("Unsupported OS sandbox; no untrusted code was executed") | |
| def with_resource_limits(command: list[str], *, file_bytes: int = 32 * 1024 * 1024) -> list[str]: | |
| """External prlimit is safe when validation is called from Python threads.""" | |
| limiter = shutil.which("prlimit") | |
| if platform.system() != "Linux" or not limiter: | |
| raise RuntimeError("Linux prlimit is required for bounded generated-code execution") | |
| return [limiter, f"--fsize={file_bytes}:{file_bytes}", "--nofile=256:256", "--core=0:0", | |
| "--as=4294967296:4294967296", "--", *command] | |
| def run_sandboxed(command: list[str], project: Path, log_path: Path, seconds: float, *, expected_running: bool = False) -> dict: | |
| wrapped, sandbox = sandbox_command(command, project) | |
| home = project / ".sandbox-home" | |
| env = {"PATH": "/usr/bin:/bin:/usr/sbin:/sbin", "HOME": str(home), "TMPDIR": str(home / "tmp"), | |
| "XDG_CONFIG_HOME": str(home / "config"), "XDG_DATA_HOME": str(home / "data"), | |
| "XDG_CACHE_HOME": str(home / "cache"), "LANG": "C.UTF-8", "GODOT_SILENCE_ROOT_WARNING": "1"} | |
| start = time.monotonic() | |
| terminated_by_harness = False | |
| with log_path.open("wb") as log: | |
| process = subprocess.Popen(with_resource_limits(wrapped), cwd=project, stdout=log, stderr=subprocess.STDOUT, | |
| env=env, start_new_session=True) | |
| startup_seconds = None | |
| deadline = start + seconds | |
| while process.poll() is None and time.monotonic() < deadline: | |
| if startup_seconds is None: | |
| with log_path.open("rb") as probe: | |
| if b"Godot Engine v4." in probe.read(4096): | |
| startup_seconds = time.monotonic() - start | |
| if expected_running: | |
| deadline = start + startup_seconds + seconds | |
| time.sleep(min(0.1, max(0, deadline - time.monotonic()))) | |
| if process.poll() is None: | |
| terminated_by_harness = True | |
| os.killpg(process.pid, signal.SIGTERM) | |
| try: | |
| returncode = process.wait(timeout=5) | |
| except subprocess.TimeoutExpired: | |
| os.killpg(process.pid, signal.SIGKILL) | |
| returncode = process.wait(timeout=5) | |
| else: | |
| returncode = process.returncode | |
| elapsed = time.monotonic() - start | |
| text = log_path.read_text(errors="replace") | |
| errors = [line for line in text.splitlines() if ERROR_PATTERN.search(line)] | |
| sandbox_failed = bool(re.search(r"(?i)sandbox-exec:|bwrap:|operation not permitted|deny\(.*\)", text)) | |
| runtime_started = "Godot Engine v4." in text and not sandbox_failed | |
| return {"command": command, "sandbox": sandbox, "returncode": returncode, | |
| "wall_seconds": round(elapsed, 4), "required_seconds": seconds, | |
| "terminated_by_harness": terminated_by_harness, "sandbox_failed": sandbox_failed, | |
| "runtime_started": runtime_started, "startup_seconds": startup_seconds, "errors": errors[:100], "log": str(log_path), | |
| "log_sha256": hashlib.sha256(log_path.read_bytes()).hexdigest(), | |
| "success": bool(runtime_started and not errors and ((terminated_by_harness and elapsed >= seconds) if expected_running else (not terminated_by_harness and returncode == 0)))} | |
| def validate_project(files: dict, work_dir: Path, godot_binary: str, run_seconds: float = 60) -> dict: | |
| """Import, then require a complete wall-clock window; preserve all diagnostics. | |
| run_seconds < 60 is allowed only for smoke tests and never qualifies as the | |
| published clean_run_60s metric. Existing evidence directories are not reused. | |
| """ | |
| if not math.isfinite(run_seconds) or run_seconds <= 0: | |
| raise ValueError("run_seconds must be positive and finite") | |
| validation_start = time.monotonic() | |
| work_dir = Path(work_dir).absolute() | |
| for ancestor in (work_dir, *work_dir.parents): | |
| if ancestor.is_symlink(): | |
| raise ProjectError(f"Symlink evidence directory rejected: {ancestor}") | |
| work_dir.mkdir(parents=True, exist_ok=True) | |
| evidence = work_dir / "validation.json" | |
| if evidence.exists(): | |
| raise FileExistsError(f"Refusing to overwrite evidence: {evidence}") | |
| result = {"compile_success": False, "clean_run": False, "clean_run_60s": False, | |
| "run_seconds": run_seconds, "project_sha256": None, "error": None} | |
| try: | |
| project = write_project(files, work_dir / "project") | |
| result["project_sha256"] = project_sha256(files) | |
| binary = str(Path(godot_binary).expanduser().resolve(strict=True)) | |
| result["godot_binary_sha256"] = hashlib.sha256(Path(binary).read_bytes()).hexdigest() | |
| result["preimport_seconds"] = time.monotonic() - validation_start | |
| imported = run_sandboxed([binary, "--headless", "--path", str(project), "--editor", "--import"], project, work_dir / "import.log", 90) | |
| result["import"] = imported | |
| result["compile_success"] = imported["success"] | |
| if imported["success"]: | |
| runtime = run_sandboxed([binary, "--headless", "--path", str(project), "--audio-driver", "Dummy", "--max-fps", "60"], project, work_dir / "runtime.log", run_seconds, expected_running=True) | |
| result["runtime"] = runtime | |
| result["clean_run"] = runtime["success"] | |
| result["clean_run_60s"] = bool(runtime["success"] and run_seconds >= 60) | |
| except (ProjectError, OSError, RuntimeError, subprocess.SubprocessError) as exc: | |
| result["error"] = f"{type(exc).__name__}: {exc}" | |
| with evidence.open("x", encoding="utf-8") as handle: | |
| json.dump(result, handle, indent=2) | |
| handle.write("\n") | |
| return result | |