- Guardrail install points to ifm-ai/pan2-guardrail, as in the vLLM-Omni and Diffusers ports.
- The basic example runs the release's own text-to-video structured caption (10 s, 241 frames) instead of a plain prompt.
- Docs and the flow-shift comment name the release checkpoint IFM/PAN2; the docs and cookbook no longer describe its release status or the parity tests' checkpoint.
- FASTVIDEO_FA4 is set only while the PAN2 pipeline loads, instead of for the rest of the process.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- mypy types for the guardrail broadcast and the negative-prompt fill; yapf kept off the negative prompt.
- Classify PAN2PipelineConfig's kernel and guardrail fields in the schema parity inventory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Attention: on sm_100+ with flash_attn.cute installed and FASTVIDEO_FA4 unset, the PAN2 pipeline turns FA4 on
before its modules load. One B300, 241 frames: 288 s instead of 670 s with the SDPA fallback. FASTVIDEO_FA4=0
keeps the previous backend.
- First frame: an image_path is read as its stored pixels, without the EXIF orientation rotation, as the reference
sampler reads it (load_image gains exif_transpose=True). All 11 sign-off inputs, one EXIF-rotated, now reach the
VAE bit-identical to the reference.
- Wan2.2 VAE decode kernel: each process builds in a temporary directory of its own instead of a shared
rank<LOCAL_RANK> directory, and finished builds go through the loader's keyed cache, added by atomic rename.
Two processes building at once both succeed; a later one fetches the build in about 2 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first sequence-parallel rank loads the guard models and runs each check. When either step raised there, the
other ranks waited at their next collective until the 10-minute store timeout. The first rank now broadcasts its
error as well as its result, and the other ranks raise with it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An empty negative prompt now selects the release's structured negative caption with its time ranges set to the
clip's duration, as the reference sampler does; a single space still gives the empty unconditional prompt.
The serving config and cookbook now say guidance 2.5, matching the presets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PAN 2's vLLM-Omni and Diffusers integrations check every prompt and every
generated video with the `pan2-guardrail` package's PAN2SafetyChecker, on by
default. FastVideo now does the same:
- `PAN2PromptGuardrailStage` checks the prompt right after input validation,
before any compute. It also refuses `output_type="latent"`, which the video
check cannot see.
- `PAN2VideoGuardrailStage` checks each decoded video after the decoding stage,
as the uint8 `[T, H, W, 3]` frames the generator delivers.
- A blocked prompt or video raises `PAN2GuardrailViolationError`.
- `PAN2PipelineConfig.enable_guardrails` (default True) turns both checks off;
`offload_guardrail_models` keeps the guard models on CPU between checks.
Loading fails with install instructions when the package is missing and the
guardrails are on.
- With sequence parallelism, only the first rank of the SP group loads the
checker and runs the checks; it broadcasts the verdict, so every rank raises
together and the guard models take memory on one GPU.
The parity test turns the guardrails off (random-weight checkpoint). Docs and
PORT_STATUS updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- fastvideo-kernel/wan22_vae_decode: the Wan2.2 (TI2V-5B) VAE decoder on sm_100a / sm_103a at 704x1248, built on
first use against fastvideo-kernel's CUTLASS submodule. It frees intermediates early and carries 4 latent frames
per pass: a 5 s video peaks at 20.3 GiB, down from 94.6 GiB in the kernel's first version, with the same output.
- WanVAEConfig.decode_kernel routes AutoencoderKLWan.decode through it where it applies (bf16 with fp32
accumulation, about 7x faster than the fp32 decode on a B300, PSNR 61-63 dB, SSIM 0.998 against it);
decode_kernel_parallel splits the decode over the sequence-parallel group (2, 4 or 8 GPUs) and broadcasts the
video, with the ranks agreeing on whether every one of them could build the kernel.
- PAN2's DiT runs fused Triton kernels for Q/K RMSNorm + RoPE and LayerNorm + modulation (pan2_fusions,
set_fused_kernels).
- PAN2PipelineConfig turns all three on (vae_decode_kernel, vae_decode_kernel_parallel, dit_fused_kernels).
- VideoGenerator copies a decoded RGB video's frames to the host once instead of once per frame.
- The PAN2 presets sample at guidance 2.5, the reference launcher's default since 2026-10-23.
These speed-ups for the MiniMax-H3 pipelines keep the output bit-identical. Each
one has a switch that is off by default:
- FASTVIDEO_MINIMAX_H3_EXACT_KERNELS: Triton RoPE, AdaLN modulation, gated
residual and SwiGLU kernels that round to bf16 at the same points as the
eager ops.
- FASTVIDEO_H3_VSA_HEADS_FIRST_TILE: VSA-H3 tiles are written directly in the
heads-first layout of the kernel.
- FASTVIDEO_H3_VAE_TILE_PARALLEL: VAE spatial tiles are split across the
sequence-parallel ranks.
- FASTVIDEO_H3_REF2VA_MEMO_ENTRIES: repeat reference encodes are reused.
The example basic_fasth3_omniref_pdd.py --lossless-accel turns them all on.
On GB200 with 4 GPUs, MiniMax-H3 text to video with audio (124 frames,
1344x768) gives bit-identical frames and audio with the first three switches
on and off.
Adds the Cosmos Predict 2.5 distilled text-to-world sampler (sCM, per-frame timesteps) and the DFD video-to-world pipeline, with their schedulers, registry entries, presets and examples.
Maintainer review fixes (Swarm orders #718 and #748): the DiT receives one shared fps value (its RoPE table is batch-independent, so per-sample vectors crashed batch > 1; differing values now raise), non-finite rollouts raise FloatingPointError instead of being zeroed by nan_to_num, the DFD detector matches "dfd" only in the last path component, the distilled example defaults to 24 fps, and the distilled rollout keeps its state in FP64 through the scheduler step as the official driver does.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
profiler_region_inference_denoising was wrapped in exactly one place (minimax_h3_denoising.py), so every model using the generic DenoisingStage (Wan among them) exported a trace with no denoising region. Wrap the denoising loops in DenoisingStage and DmdDenoisingStage with profiler_region("inference_denoising"); the loop bodies are untouched.
These are the two classes STAGE_METRIC_MAP names directly. The perf harness also reports dit_time_s for every stage that inherits performance_component_metric, and those forward overrides (Cosmos*, LongCat*, Gen3C, GameCraft, HYWorld, MatrixGame2/3, Causal*, DreamXWorldAR, LingBot, GlmImage) remain outside the region for now. With no profiler configured the wrap is a no-op; SSIM is exactly 1.0 on FLASH_ATTN and TORCH_SDPA.
Caveat: the region span now appears in exported traces, but a pre-existing profiler behaviour (region exit toggles CUDA collection off, fastvideo/profiler.py) may leave GPU kernel rows out of the trace.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With the device-to-host copy gone, the remaining host-to-device copy is the
stage's cost, and from pageable memory it is staged through a bounce buffer at
roughly 2.6 GB/s instead of full PCIe speed. The host copies are made once, so
pinning them costs nothing per request.
It does cost the module's size in non-pageable host memory (10.4 GB for the
H3 video VAE, 0.6 GB for the audio VAE, per rank) for as long as the pipeline
is alive. It follows --pin-cpu-memory, which is on by default; --no-pin-cpu-
memory keeps the copies pageable and only loses the bandwidth. Note that this
flag did not already reach the VAE: its other use is FSDP's own CPUOffloadPolicy
and the VAE does not go through FSDP.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ATw7g6cxq9N2LtnMratsru
The decode stages move the 10.4 GB fp32 video VAE (and the 0.6 GB audio VAE)
with module.to(device) / module.to("cpu") on every request. The weights are
frozen -- the whole stage runs under no_grad -- so the copy back to the host
returns bytes the host already has.
Keep one host copy per tensor, made the first time the module is loaded, and
on unload point .data back at it instead of copying. One H2D per request, no
D2H at all; the module still lives on the host between requests. Where the old
path allocated a fresh host tensor per cycle and freed the previous one, this
reuses one buffer, so peak host memory can only go down.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ATw7g6cxq9N2LtnMratsru
Accept the causal-VAE-aligned 362-frame bucket (15.083 s) for 15-second H3 requests.
- Hoist MINIMAX_H3_MIN/MAX_ALIGNED_FRAMES next to align_num_frames and reuse them in the CUDA stage, the MLX runtime, and the LoRA example so the bound is consistent across entry points.
- State the accepted frame range in the error messages.
- Cover seconds="15" -> 362 at OpenAI admission and the MLX resolve_geometry bound.
- Update the Spark/openai cookbook frame caps from 345 to 362.
Known gaps: no GPU-lane coverage for the 362 bucket (SSIM runs 124 frames; the golden gate is a fixed-geometry fingerprint), and explicit num_frames=360 still hits the pre-existing grid gate.
