[Model]: Support Helios from ByteDance

[princepride]

Purpose

Add full support for the Helios video generation model family in vLLM-Omni, covering 3 model variants (Helios-Base, Helios-Mid, Helios-Distilled) across 3 generation tasks (T2V, I2V, V2V) — a complete 3×3 support matrix.

Variant	Pipeline Class	Key Features
Helios-Base	HeliosPipeline	Stage 1 single-stage denoising, guidance_scale=5.0
Helios-Mid	HeliosPyramidPipeline	Stage 2 pyramid denoising, CFG-Zero*
Helios-Distilled	HeliosPyramidPipeline	Stage 2 pyramid + DMD few-step inference, guidance_scale=1.0

Test Plan

Manual end-to-end testing across all supported model×task combinations, NPU is supported as well. Run from the repository root:

T2V (Text-to-Video):

# Helios-Base (Stage 1)
python3 examples/offline_inference/helios/end2end.py \
--sample-type t2v \
--model BestWishYsh/Helios-Base \
--prompt "A vibrant tropical fish swimming gracefully among colorful coral reefs in a clear, turquoise ocean. The fish has bright blue and yellow scales with a small, distinctive orange spot on its side, its fins moving fluidly. The coral reefs are alive with a variety of marine life, including small schools of colorful fish and sea turtles gliding by. The water is crystal clear, allowing for a view of the sandy ocean floor below. The reef itself is adorned with a mix of hard and soft corals in shades of red, orange, and green. The photo captures the fish from a slightly elevated angle, emphasizing its lively movements and the vivid colors of its surroundings. A close-up shot with dynamic movement." \
--num-frames 600 \
--seed 42 \
--output helios_t2v_base.mp4

# Helios-Mid (Stage 2 + CFG-Zero*)
python examples/offline_inference/helios/end2end.py \
    --model BestWishYsh/Helios-Base --sample-type t2v \
    --prompt "A dynamic time-lapse video showing the rapidly moving scenery from the window of a speeding train." \
    --guidance-scale 5.0 --is-enable-stage2 \
    --pyramid-num-inference-steps-list 20 20 20 \
    --use-cfg-zero-star --use-zero-init --zero-steps 1 \
    --output helios_t2v_mid.mp4

# Helios-Distilled (Stage 2 + DMD)
python examples/offline_inference/helios/end2end.py \
    --model BestWishYsh/Helios-Base --sample-type t2v \
    --prompt "A dynamic time-lapse video showing the rapidly moving scenery from the window of a speeding train." \
    --num-frames 240 --guidance-scale 1.0 --is-enable-stage2 \
    --pyramid-num-inference-steps-list 2 2 2 \
    --is-amplify-first-chunk --output helios_t2v_distilled.mp4

I2V (Image-to-Video):

# Helios-Base
python examples/offline_inference/helios/end2end.py \
    --model ./Helios-Base --sample-type i2v \
    --image-path wave.jpg \
    --prompt "A towering emerald wave surges forward, its crest curling with raw power and energy." \
    --guidance-scale 5.0 --output helios_i2v_base.mp4

V2V (Video-to-Video):

# Helios-Base
python examples/offline_inference/helios/end2end.py \
    --model ./Helios-Base --sample-type v2v \
    --video-path car.mp4 \
    --prompt "A bright yellow Lamborghini speeds along a curving mountain road." \
    --guidance-scale 5.0 --output helios_v2v_base.mp4

Test Result

Helios-Base, T2V

Details

helios_output_1.mp4

Helios-Mid, T2V

Details

helios_t2v_mid.mp4

Helios-Distilled, T2V

Details

helios_t2v_distilled.mp4

Helios-Base, I2V

Details

helios_i2v_base.mp4

Helios-Base, V2V

Details

helios_v2v_base.mp4

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 69497c6179

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[hsliuustc0106]

vllm-omni-reviewer Code Review

Found 3 critical issues:

1. Documentation formatting bug breaks table rendering

Line 10 has an orphaned HeliosPyramidPipeline outside the markdown table that breaks rendering. This appears to be a copy-paste error - the pipeline name is already in the table row above.

vllm-omni/docs/models/supported_models.md

Lines 9 to 11 in 83ce087

	## List of Supported Models for Nvidia GPU / AMD GPU
	<style>

Fix: Delete the orphaned line.

2. Missing test coverage for new model

This PR adds 3362 lines of new model code (transformer, pipeline, scheduler) but includes no test files. Per conventions, new model implementations require correctness validation tests.

https://github.com/vllm-project/vllm-omni/blob/83ce0872685aa373966665f8658b91f23d4a50c5/vllm_omni/diffusion/models/helios/

Fix: Add tests validating model loading and basic inference for the Helios variants.

3. Memory/hardware requirements not documented

The PR adds 3 model variants (Base, Mid, Distilled) with different memory profiles but doesn't document minimum VRAM, recommended GPUs, or memory configuration settings. Users have no guidance on hardware requirements.

vllm-omni/docs/models/supported_models.md

Line 9 in 83ce087

## List of Supported Models for Nvidia GPU / AMD GPU

Fix: Add memory requirements documentation for each variant (e.g., "Helios-Base: X GB VRAM minimum").

---

🤖 Generated with vllm-omni-reviewer

[hsliuustc0106]

vllm-omni-reviewer Code Review

Found 2 issues requiring fixes:

1. Markdown table broken (line 10) - orphaned HeliosPyramidPipeline text needs deletion
2. Missing memory requirements (line 9) - document VRAM needs for 3 model variants

See inline comments below.

[Gaohan123]

The link https://github.com/BestWishYsh/Helios cannot open

[princepride]

The model still didn't open source, please wait

[gcanlin]

python3 examples/offline_inference/helios/end2end.py \
--sample-type t2v \
--model ./Helios-Base \
--prompt "A vibrant tropical fish swimming gracefully among colorful coral reefs in a clear, turquoise ocean. The fish has bright blue and yellow scales with a small, distinctive orange spot on its side, its fins moving fluidly. The coral reefs are alive with a variety of marine life, including small schools of colorful fish and sea turtles gliding by. The water is crystal clear, allowing for a view of the sandy ocean floor below. The reef itself is adorned with a mix of hard and soft corals in shades of red, orange, and green. The photo captures the fish from a slightly elevated angle, emphasizing its lively movements and the vivid colors of its surroundings. A close-up shot with dynamic movement." \
--num-frames 600 \
--seed 42 \
--output helios_t2v_base.mp4 \
--tensor-parallel-size 2

NPU needs to enable TP=2. And end-to-end can run successfully.

[Sy0307]

If Helios has any plans for optimization on vllm-omni in the future, I would be happy to participate.