Text Generation
Transformers
Safetensors
English
Chinese
Russian
yue2
music-generation
orbitquant
quantization
4-bit precision
custom-code
8-bit precision
Instructions to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="WaveCut/YuE2-3B-OrbitQuant-W4A4")# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("WaveCut/YuE2-3B-OrbitQuant-W4A4", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "WaveCut/YuE2-3B-OrbitQuant-W4A4" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WaveCut/YuE2-3B-OrbitQuant-W4A4", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/WaveCut/YuE2-3B-OrbitQuant-W4A4
- SGLang
How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 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 "WaveCut/YuE2-3B-OrbitQuant-W4A4" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WaveCut/YuE2-3B-OrbitQuant-W4A4", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "WaveCut/YuE2-3B-OrbitQuant-W4A4" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WaveCut/YuE2-3B-OrbitQuant-W4A4", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with Docker Model Runner:
docker model run hf.co/WaveCut/YuE2-3B-OrbitQuant-W4A4
File size: 12,497 Bytes
3ea3da7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 | """Dependency-free progress output; never inspects tensors or changes RNG state.
Use ``with Progress().stage("Generating audio tokens", unit="tokens") as s``
and pass ``s.token`` as the output-token callback. Counts belong to the caller:
prefill, external ABC, and extra CFG branches must not be counted as outputs.
Rates are observed counts divided by elapsed stage wall time, including waits.
An unknown total stays unknown; a generation limit is not a progress target.
"""
from __future__ import annotations
import math
import operator
import os
import sys
import threading
import time
_clock = time.monotonic
_STATUSES = {
"completed": "Completed",
"failed": "Failed",
"cancelled": "Cancelled",
"truncated": "Finished (generation limit reached)",
}
def _count(value, name):
if isinstance(value, bool):
raise ValueError(f"{name} must be a nonnegative integer")
try:
value = operator.index(value)
except TypeError as exc:
raise ValueError(f"{name} must be a nonnegative integer") from exc
if value < 0:
raise ValueError(f"{name} must be a nonnegative integer")
return value
def _ascii(text):
return "".join(c if " " <= c <= "~" else " " for c in str(text)).strip()
def _terminal_columns(stream):
# Unconfigured PTYs can report (0, 0), including those created by `script`.
# Honor a positive COLUMNS override, then the device, then a usable fallback.
try:
columns = int(os.environ.get("COLUMNS", ""))
if columns > 0:
return columns
except ValueError:
pass
try:
columns = os.get_terminal_size(stream.fileno()).columns
if columns > 0:
return columns
except (AttributeError, OSError, ValueError):
pass
return 80
def _exit_status(exc_type):
if exc_type is None:
return "completed"
if issubclass(exc_type, (InterruptedError, KeyboardInterrupt)) or any(
base.__name__ in {"CancelledError", "CanceledError"} for base in exc_type.__mro__
):
return "cancelled"
return "failed"
class Progress:
"""Write ASCII progress to stderr, with a heartbeat during blocking work.
TTYs at least 60 columns wide refresh one line. Narrow terminals and other
streams receive start/end lines and at most one intermediate line per five
seconds, without clipping labels or counts. ``enabled=False`` starts no thread
and performs no stream operations. A stage owns its heartbeat lifetime, so
an outer ``with Progress()`` is optional. ``close()`` is also available.
"""
def __init__(self, enabled=True, stream=None, refresh_interval=0.25):
if not math.isfinite(refresh_interval) or refresh_interval <= 0:
raise ValueError("refresh_interval must be positive and finite")
self.enabled = bool(enabled)
self.stream = sys.stderr if stream is None else stream
self.refresh_interval = float(refresh_interval)
self._lock = threading.RLock()
self._active = []
self._thread = None
self._stop = None
self._closed = False
self._summary_written = False
self._line_width = 0
self._tty = False
self._columns = 80
if self.enabled:
try:
self._tty = bool(self.stream.isatty())
except (AttributeError, OSError, ValueError):
pass
if self._tty:
self._columns = _terminal_columns(self.stream)
self._tty = self._columns >= 60
self._interval = self.refresh_interval if self._tty else max(5.0, self.refresh_interval)
def __enter__(self):
with self._lock:
if self._closed:
raise RuntimeError("Progress is closed")
return self
def __exit__(self, exc_type, exc, traceback):
self.close(status=_exit_status(exc_type))
return False
def stage(self, label, total=None, unit=None):
"""Return a context manager; entering it starts this stage at zero."""
return _Stage(self, label, total=total, unit=unit)
def _write(self, text, final=False):
if not self.enabled:
return
try:
if self._tty:
text = text[:self._columns - 1]
self.stream.write("\r" + text + " " * max(0, self._line_width - len(text)))
self._line_width = len(text)
if final:
self.stream.write("\n")
self._line_width = 0
else:
self.stream.write(text + "\n")
self.stream.flush()
except (OSError, ValueError):
# A closed terminal must not turn successful inference into failure.
self.enabled = False
if self._stop is not None:
self._stop.set()
def _newline(self):
if self.enabled and self._tty and self._line_width:
self._write("", final=True)
def _render(self, stage, now, status=None, force=False):
if not self.enabled or (not force and now - stage._last_render < self._interval):
return
stage._last_render = now
elapsed = max(0.0, now - stage._started)
parts = []
if stage.total is not None:
# Keep an honest count even if a caller exceeds its advertised total.
amount = f"{stage.completed}/{stage.total} {stage.unit or 'items'}"
if stage.total > 0:
amount += f" ({100 * stage.completed / stage.total:.0f}%)"
if self._tty:
filled = min(8, int(8 * stage.completed / stage.total))
amount = "[" + "#" * filled + "-" * (8 - filled) + "] " + amount
parts.append(amount)
elif stage.unit is not None or stage.completed:
parts.append(f"{stage.completed} {stage.unit or 'items'}")
if stage.unit == "tokens":
rate = stage.completed / elapsed if elapsed > 0 else 0.0
parts.append(f"{rate:.1f} tokens/s")
parts.append(f"{elapsed:.1f}s" if self._tty else f"elapsed {elapsed:.1f}s")
suffix = " | ".join(parts)
if status is not None:
prefix = "Limit reached" if self._tty and status == "truncated" else _STATUSES[status]
elif self._tty:
prefix = "|/-\\"[int(elapsed / self.refresh_interval) % 4]
else:
prefix = "Starting" if force and stage.completed == 0 else "Running"
label = stage.label
if self._tty:
available = self._columns - 1 - len(f"[YuE2] {prefix} : {suffix}")
if len(label) > max(0, available):
label = label[:max(0, available - 3)] + ("..." if available >= 3 else "")
self._write(f"[YuE2] {prefix} {label}: {suffix}", final=status is not None)
def _heartbeat(self, stop):
while not stop.wait(self._interval):
with self._lock:
if stop.is_set() or not self.enabled:
return
if self._active:
self._render(self._active[-1], _clock())
def _start(self, stage):
with self._lock:
if self._closed:
raise RuntimeError("Progress is closed")
if stage._started is not None:
raise RuntimeError("A progress stage can only be entered once")
stage._started = _clock()
self._newline()
self._active.append(stage)
self._render(stage, stage._started, force=True)
if self.enabled and self._thread is None:
self._stop = threading.Event()
self._thread = threading.Thread(target=self._heartbeat, args=(self._stop,),
name="yue2-progress", daemon=True)
self._thread.start()
def _finish(self, stage, status):
if status not in _STATUSES:
raise ValueError("Unknown progress status")
thread = None
with self._lock:
if stage._finished:
return
if stage._started is None:
raise RuntimeError("Enter a progress stage before finishing it")
stage._finished = True
stage.status = status
if self._active[-1] is not stage:
self._newline()
self._active.remove(stage)
self._render(stage, _clock(), status=status, force=True)
if self._active:
self._render(self._active[-1], _clock(), force=True)
elif self._thread is not None:
self._stop.set()
thread, self._thread = self._thread, None
self._stop = None
if thread is not None and thread is not threading.current_thread():
thread.join()
def close(self, status="completed"):
"""Finish active stages and stop the heartbeat; safe to call repeatedly."""
if status not in _STATUSES:
raise ValueError("Unknown progress status")
with self._lock:
self._closed = True
active = list(reversed(self._active))
for stage in active:
self._finish(stage, status)
def complete(self, audio_seconds, elapsed, *, truncated=False):
"""Write one final summary without starting a heartbeat thread."""
if not all(math.isfinite(x) and x >= 0 for x in (audio_seconds, elapsed)):
raise ValueError("audio_seconds and elapsed must be nonnegative and finite")
with self._lock:
if self._active:
raise RuntimeError("Finish active stages before the completion summary")
if self._summary_written:
return
self._summary_written = True
status = _STATUSES["truncated" if truncated else "completed"]
self._write(f"[YuE2] {status}: {audio_seconds:.1f}s audio in {elapsed:.1f}s", final=True)
class _Stage:
"""Stage handle returned by :meth:`Progress.stage`; updates are thread-safe."""
def __init__(self, owner, label, total=None, unit=None):
self._owner = owner
self.label = _ascii(label)
self.unit = _ascii(unit) if unit is not None else None
self.total = None if total is None else _count(total, "total")
self.completed = 0
self.status = None
self._started = None
self._last_render = -math.inf
self._finished = False
def __enter__(self):
self._owner._start(self)
return self
def __exit__(self, exc_type, exc, traceback):
self.finish(status=_exit_status(exc_type))
return False
def _check_active(self):
if self._started is None:
raise RuntimeError("Enter a progress stage before updating it")
def update(self, completed, total=None):
"""Set an absolute count and optionally the total discovered at runtime."""
completed = _count(completed, "completed")
total = None if total is None else _count(total, "total")
with self._owner._lock:
self._check_active()
if self._finished:
return
if completed < self.completed:
raise ValueError("completed must not decrease")
self.completed = completed
if total is not None:
self.total = total
if self._owner._active[-1] is self:
self._owner._render(self, _clock())
def set_total(self, total):
"""Set the real work total; ``None`` restores an unknown total."""
total = None if total is None else _count(total, "total")
with self._owner._lock:
self._check_active()
if self._finished:
return
self.total = total
if self._owner._active[-1] is self:
self._owner._render(self, _clock())
def advance(self, count=1):
"""Add completed work without a read/update race between callbacks."""
count = _count(count, "count")
with self._owner._lock:
self.update(self.completed + count)
def token(self, phase, token):
"""Count one actual emitted token; phase/token contents are not inspected."""
self.advance()
def finish(self, status="completed"):
"""Stop this stage. Supports completed, failed, cancelled, or truncated."""
self._owner._finish(self, status)
|