SolitaryThinker
a16ce1a07f
Five self-contained, runnable scripts (CPU/numpy) for the Wan2.1-1.3B card on the v2 runtime, each bootstrapping the repo onto sys.path so they run from anywhere: - 01_basic_t2v.py minimal path: build engine → T2V request → run → video - 02_params_and_reproducibility.py DiffusionParams knobs + seeded bit-identical reproducibility - 03_streaming.py per-denoise-step preview chunks (OutputSpec stream) - 04_concurrent_interleaved.py step-interleaved batching + interleave parity gate + cache reuse - 05_async_serving.py AsyncEngine: concurrent generate, event stream, step-boundary cancel + README.md indexing them. All five run green; use only the public API.
53 lines
1.9 KiB
Python
53 lines
1.9 KiB
Python
#!/usr/bin/env python3
|
|
"""03 — Streaming previews (Wan2.1-1.3B).
|
|
|
|
Pass `OutputSpec(stream={"video": True})` and the denoise loop emits one preview chunk per step (a
|
|
`StreamChunk` carrying the in-progress latent), surfaced as `media.chunk` events and counted in
|
|
`metrics["stream_chunks"]`. Streaming is **off by default** — the serve path is byte-for-byte the same
|
|
whether or not anyone subscribes, so previews never change the result.
|
|
|
|
Run: python3 v2_examples/inference/03_streaming.py
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import pathlib
|
|
import sys
|
|
|
|
sys.path.insert(0, str(pathlib.Path(__file__).resolve().parents[2]))
|
|
|
|
import numpy as np
|
|
|
|
from v2.models import build_default_engine
|
|
from v2.request import DiffusionParams, OutputSpec, TaskType, make_request
|
|
|
|
MODEL = "wan2.1-1.3b"
|
|
STEPS = 10
|
|
|
|
|
|
def _run(eng, *, stream: bool):
|
|
outputs = OutputSpec(stream={"video": True}) if stream else OutputSpec()
|
|
req = make_request(TaskType.T2V, MODEL, "a hot air balloon over a valley",
|
|
diffusion=DiffusionParams(num_steps=STEPS, seed=5), outputs=outputs)
|
|
return eng.run(req)
|
|
|
|
|
|
def main() -> None:
|
|
eng = build_default_engine()
|
|
|
|
off = _run(eng, stream=False)
|
|
on = _run(eng, stream=True)
|
|
|
|
print(f"streaming OFF : stream_chunks={off.metrics.get('stream_chunks', 0):.0f}")
|
|
print(f"streaming ON : stream_chunks={on.metrics.get('stream_chunks', 0):.0f} "
|
|
f"(one preview per denoise step ⇒ == num_steps={STEPS})")
|
|
|
|
# streaming must NOT change the result — it's a read-only preview channel
|
|
same = np.array_equal(np.asarray(off.artifacts["video"].frames),
|
|
np.asarray(on.artifacts["video"].frames))
|
|
print(f"result identical with/without streaming : {same}")
|
|
print("\n(For LIVE async event streaming — media.chunk as it happens — see 05_async_serving.py.)")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|