unsloth/studio/backend
Daniel Han 40e3747d43 Absorb the first-generation compile hitch with a post-load background prewarm
vLLM and SGLang finish every compilation at server startup (dummy batches
through each compiled shape) so no request ever pays a compile mid-serving.
The video backend's compiled tier instead paid a first-generation extra after
every restart: ~54 s cold and ~11.3 s even with a warm Mega-cache bundle (the
residual is dynamo tracing plus cudnn.benchmark autotune, which the bundle
cannot carry).

After a compiled DEFAULT-tier resident load commits, a daemon thread now runs
one tiny throwaway generation (192x128 snapped, 4k+1-lattice 9 frames, 2
steps) under the generate lock. The default tier compiles dynamic=True, so
the small trace serves every later resolution. Measured through the real
backend (HunyuanVideo-1.5-480p, B200, 480x288/17f/30 steps):

  warm bundle: first-generation extra 11.3 s -> 2.1 s (9.6 s background warmup)
  cold start:  the full compile moves off the user's first request entirely
               (14.5 s background; first generation extra 2.1 s), and the
               warmup persists the Mega-cache bundle itself
  steady state: unchanged (2.4-2.5 s per 30-step clip in every phase)

Exactly lossless by construction: the warmup only changes when compilation
work happens. It resets its step-cache residuals, the real generation seeds
its own generator, and no process-wide flag is touched.

The warmup registers itself as the active cancellable job, so unload, a new
load, or cancel_generate abort it at a step boundary (verified: unload 2 s
into a running prewarm returns in 6.7 s with the warmup cancelled). It yields
untouched when a real request arrived first and is token-scoped against
superseded loads. Gated per family (supports_compile_prewarm), skipped for
speed=max (static per-shape graphs a warmup shape cannot serve), offload
(every warmup forward would stream the DiT over PCIe), and CFG parallel (its
planner owns compile-sensitive runs); the UNSLOTH_DIFFUSION_COMPILE_PREWARM
kill switch disables it. The decision and reason land in the resolved record.

Tests: +4 hermetic (decision gates, engaged-load spawn with snapped tiny
shape, skip-without-compile, yield to generations / stale tokens); related
backend set 551 passed; ruff clean.
2026-07-11 06:49:07 +00:00
..
assets Add DeepSeek-V4-Flash-GGUF to Studio with none/high/max reasoning (#6908) 2026-07-07 06:13:43 -07:00
auth Studio: opt-in OpenAI /v1 model auto-switch and idle keep-warm (#6392) 2026-07-01 06:42:23 -07:00
core Absorb the first-generation compile hitch with a post-load background prewarm 2026-07-11 06:49:07 +00:00
hub Studio: resolve the repo-root MTP drafter after the MTP/ GGUF rename (#7031) 2026-07-09 06:46:00 -07:00
loggers Studio: trim serving-log noise and surface llama-server engine stats (#6377) 2026-06-17 05:37:57 -07:00
models Merge branch 'image-generation' into video-diffusion-improvements 2026-07-10 17:44:25 +00:00
plugins Reduce and tighten code comments and docstrings repo-wide (#6095) 2026-06-08 23:09:51 -07:00
requirements Merge remote-tracking branch 'origin/main' into fold-integration 2026-07-07 05:52:16 +00:00
routes Merge branch 'image-generation' into video-diffusion-improvements 2026-07-10 17:44:25 +00:00
state Studio: shareable per-checkpoint preview links (#6486) 2026-06-24 06:31:53 -07:00
storage Sort chat recents by last activity (#6844) 2026-07-07 17:54:32 +01:00
tests Absorb the first-generation compile hitch with a post-load background prewarm 2026-07-11 06:49:07 +00:00
utils Studio: add Vulkan llama.cpp support (#5819) 2026-07-09 03:39:48 -07:00
__init__.py Final cleanup 2026-03-12 18:28:04 +00:00
_platform_compat.py Reduce and tighten code comments and docstrings repo-wide (#6095) 2026-06-08 23:09:51 -07:00
cloudflare_tunnel.py Reap Studio child processes when the parent dies abnormally (#6425) 2026-06-18 05:51:22 -07:00
colab.py Studio Colab: opt-in shareable Cloudflare tunnel link (#6684) 2026-06-26 00:56:23 -07:00
main.py Fix picker dead-end for single-file repos, stale LoRA state after deferred compile, slash upload cap 2026-07-10 08:07:01 +00:00
run.py Studio: fix torch CUDA undefined-symbol errors from a conflicting LD_LIBRARY_PATH (#6905) 2026-07-06 23:48:23 -07:00
startup_banner.py Fix Windows Studio UTF-8 startup handling (#6614) 2026-07-01 13:47:33 +01:00