[feat]: Cosmos3Config + registry wire-up for nvidia/Cosmos3-Nano

Cosmos3Config(PipelineConfig): reuses Cosmos25VAEConfig (Cosmos3 uses DistributedAutoencoderKLWan); text_encoder_configs=() since Cosmos3LanguageModel lives inside the DiT; flow_shift=1.0 default (T2V/I2V engine init; T2I uses 3.0 per-request via _set_flow_shift). Registry entry registers BEFORE the generic cosmos detector to win path precedence (same pattern as GEN3C). Resolves nvidia/Cosmos3-Nano -> Cosmos3Config without regressing Cosmos25/GEN3C/Cosmos.

fastvideo/configs/pipelines/cosmos3.py

@@ -0,0 +1,65 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Cosmos3 pipeline configuration.
+
+Port reference: vllm-omni PR #3454 (https://github.com/vllm-project/vllm-omni/pull/3454)
+HEAD ``8536f5b1421f``. Cosmos3 is structurally different from Cosmos 2.5:
+
+- Dual-stack DiT (UND + GEN) lives entirely inside ``Cosmos3VFMTransformer``.
+- No separate text encoder — ``Cosmos3LanguageModel`` is inside the DiT, so
+  ``text_encoder_configs`` is the empty tuple. The tokenizer is loaded by the
+  pipeline's ``__init__`` from the checkpoint's ``text_tokenizer/`` subfolder;
+  that is an instance-level concern, not a config-level concern.
+- VAE is ``DistributedAutoencoderKLWan`` (same as Cosmos 2.5), so the
+  ``Cosmos25VAEConfig`` is reused as-is.
+- T2I default ``flow_shift`` is 3.0 (set per-request by ``_set_flow_shift``);
+  T2V/I2V use the engine-init default of 1.0 baked into this config.
+"""
+from collections.abc import Callable
+from dataclasses import dataclass, field
+
+import torch
+
+from fastvideo.configs.models import DiTConfig, EncoderConfig, VAEConfig
+from fastvideo.configs.models.dits.cosmos3 import (Cosmos3ArchConfig, Cosmos3VideoConfig)
+from fastvideo.configs.models.encoders import BaseEncoderOutput
+from fastvideo.configs.models.vaes import Cosmos25VAEConfig  # Cosmos3 reuses the Wan-based VAE
+from fastvideo.configs.pipelines.base import PipelineConfig
+
+
+def _identity_preprocess_text(prompt: str) -> str:
+    return prompt
+
+
+@dataclass
+class Cosmos3Config(PipelineConfig):
+    """Configuration for the Cosmos3 video generation pipeline (T2V/I2V/T2I).
+
+    Tier A scope: architectural skeleton. Real weights from ``nvidia/Cosmos3-Nano``
+    are not yet published; this config is shaped to be activatable when they
+    land without further config changes.
+    """
+
+    dit_config: DiTConfig = field(default_factory=lambda: Cosmos3VideoConfig(arch_config=Cosmos3ArchConfig()))
+
+    vae_config: VAEConfig = field(default_factory=Cosmos25VAEConfig)
+
+    # No separate text encoder: Cosmos3LanguageModel lives inside the DiT.
+    text_encoder_configs: tuple[EncoderConfig, ...] = field(default_factory=tuple)
+
+    preprocess_text_funcs: tuple[Callable[[str], str],
+                                 ...] = field(default_factory=lambda: (_identity_preprocess_text, ))
+    postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], torch.Tensor], ...] = field(default_factory=tuple)
+
+    dit_precision: str = "bf16"
+    vae_precision: str = "bf16"
+    text_encoder_precisions: tuple[str, ...] = field(default_factory=tuple)
+
+    embedded_cfg_scale: float = 0.0
+    flow_shift: float = 1.0  # T2V/I2V engine-init; T2I uses 3.0 set per-request
+
+    vae_tiling: bool = False
+    vae_sp: bool = False
+
+    def __post_init__(self):
+        self.vae_config.load_encoder = True
+        self.vae_config.load_decoder = True

fastvideo/registry.py

@@ -20,6 +20,7 @@ from fastvideo.configs.pipelines.cosmos2_5 import (
     Cosmos25Config,
     Cosmos25_14BConfig,
 )
+from fastvideo.configs.pipelines.cosmos3 import Cosmos3Config
 from fastvideo.configs.pipelines.hunyuan import FastHunyuanConfig, HunyuanConfig
 from fastvideo.configs.pipelines.hunyuangamecraft import HunyuanGameCraftPipelineConfig
 from fastvideo.configs.pipelines.gen3c import Gen3CConfig
@@ -555,6 +556,22 @@ def _register_configs() -> None:
         default_preset="gen3c_cosmos_7b",
     )
 
+    # Cosmos 3 (must register before Cosmos 2.5 and generic Cosmos detectors
+    # so the cosmos3 path-detection takes precedence)
+    register_configs(
+        sampling_param_cls=None,
+        pipeline_config_cls=Cosmos3Config,
+        workload_types=(WorkloadType.T2V, ),
+        hf_model_paths=[
+            "nvidia/Cosmos3-Nano",
+        ],
+        model_detectors=[
+            lambda path: "cosmos3" in path.lower() or "cosmos-3" in path.lower(),
+        ],
+        model_family="cosmos3",
+        default_preset="cosmos3_nano",
+    )
+
     # Cosmos 2.5 (2B)
     register_configs(
         sampling_param_cls=None,