TeleFuser¶
An open-source streaming inference and serving framework for real-time world models and multimodal generation, built for continuous pipelines, distributed GPU execution, and production service interfaces.
PyTorch 2.6+ CUDA 12.8+ Triton kernels FastAPI service Ray distributed
Runtime Capabilities¶
Continuous execution, stateful sessions, and bidirectional control loops.
Ulysses, Ring Attention, tensor parallelism, pipeline parallelism, and FSDP.
Compile-aware ops with eager CUDA Triton kernels and PyTorch native fallbacks.
FastAPI batch serving and LiveKit-backed rooms for server-push and resilient interactive WebRTC.
AdaTaylorCache and runtime cache controls for repeated generation workloads.
Reusable stages, model configs, schedulers, and pipeline orchestration.
Supported Models¶
World Model and Real-Time¶
| Model | Tasks | Description |
|---|---|---|
| LingBot-World v2 | Bidirectional streaming | Camera-controlled interactive world model via LiveKit |
| LingBot-World-Fast | Bidirectional streaming | Legacy/causal-fast model via LiveKit reliable data messages |
| ABot-World-0-5B-LF | Single-GPU interactive generation | Direct HTTP or LiveKit browser control with persistent causal KV state |
Video Generation¶
| Model | Tasks | Description |
|---|---|---|
| WanVideo (Wan2.1 / Wan2.2) | T2V, I2V, FL2V | Video generation and editing |
| LTX Video | I2V + Audio | Video generation with audio |
| LTX-2.5 Distilled | T2V, I2V + Audio | ModuleManager-backed six-stage pipeline with ½/4-H100 Ulysses SP |
| MiniMax H3 | T2VA, FL2VA, Ref2VA + Audio | Local 768p joint audio-video generation |
| FlashVSR | VSR | Video super-resolution |
| SwiftVR | Causal video restoration | Stateful restoration with BF16, compile, FP8Linear, Ulysses SP, and stage-parallel options |
| LiveAct | S2V | Speech-to-video |
| LongCat-Video | T2V, I2V | Long video generation |
| LingBot-Video | T2I, T2V, TI2V, MoE refiner | Precision-first Dense and MoE video generation |
Image Generation¶
| Model | Tasks | Description |
|---|---|---|
| Qwen-Image | T2I, Edit | Image generation and editing |
| Z-Image | T2I | Image generation |
| Flux2 Klein | T2I | Image generation |
Vision-Language-Action¶
| Model | Tasks | Description |
|---|---|---|
| LingBot-VLA v2 | Robot manipulation | Vision-language-action inference for supported robot profiles |
Start Here¶
Core WebRTC ExperienceControl LingBot-World v2 and receive generated video in the browser. InstallationInstall the package and verify CUDA availability. Basic InferenceRun Wan2.1 1.3B locally and submit an HTTP task. Supported ModelsSelect a model, checkpoint source, and validated profile.
Documentation Sections¶
Runtime and OptimizationConfiguration, parallelism, attention, caching, quantization, and offload. Operations and ReferenceMetrics, logging, profiling, benchmarks, and troubleshooting. Developer GuideIntegrate models, stages, examples, and public operations. Technical BlogOptimization design, profiling evidence, results, and related work.