vllm_omni.diffusion.models.pi05.pipeline_pi05 ¶

π0.5 VLA pipeline for vllm-omni.

Entry point for DiffusionEngine.step() → pipeline.forward(req). Mirrors the DreamZero contract and the π0 pipeline: the pipeline owns ALL preprocessing. It reads the raw robot observation from req.sampling_params.extra_args["robot_obs"] (delivered by the OpenPI realtime serving layer), builds model inputs, runs flow-matching denoising, and returns DiffusionOutput(output={"actions": ndarray}).

π0.5 is stateless across calls (no KV reuse, first-order Markov), so session_id / reset from the OpenPI protocol are accepted but ignored.

The post-processing order is load-bearing and matches LeRobot::

unnormalize → absolute actions → to_cpu

AbsoluteActionsProcessorStep must run after unnormalization, because a relative-action checkpoint's norm_stats are computed in relative space.

DEFAULT_PI05_TOKENIZER = 'google/paligemma-3b-pt-224'

SUPPORTED_DTYPES = (torch.float32, torch.bfloat16)

logger = init_logger(__name__)

config = self._build_config(od_config)

model = self._initialize_model()

model_dir = self._resolve_model_dir(od_config.model)

od_config = od_config

prefix = prefix

processor = Pi05Processor(
    self.config, self.tokenizer, self._device
)

tokenizer = self._load_tokenizer()

tokenizer_source = str(
    custom_args.get(
        "tokenizer", self._resolve_tokenizer_source()
    )
)

forward(
    req: OmniDiffusionRequest, **kwargs
) -> DiffusionOutput

has_real_checkpoint() -> bool

load_weights(weights=())

get_pi05_post_process_func(od_config: OmniDiffusionConfig)