nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[refactor] v2: use absolute imports (v2.*) everywhere instead of relative

Browse Source 
Mechanical conversion of every relative import under v2/ to an absolute v2.* path
(from .x / ..x / ...x -> from v2.<pkg>.x) so imports are unambiguous, grep-able, and
stable when code is copied/moved between entrypoints (VideoGenerator, CLI, server).

Surgical prefix-only rewrite: only the 'from <dots><module>' prefix changed — import
names, parentheses, multi-line formatting, comments, and ordering are byte-for-byte
preserved (no collapsing, no reorder, no unrelated reformatting).

- 471 imports across 125 files; v2/tests/ was already absolute (untouched).
- Validated: all 125 files compile, every 'from v2.* import' target resolves to a real
  module/package, zero relative imports remain (full sweep), CPU mini suite green
  (202 passed, 2 skipped).
This commit is contained in:
SolitaryThinker
2026-07-04 17:15:34 +00:00
committed by Peiyuan Zhang
parent c6d2976fc2
commit 4e331eb7ff
125 changed files with 471 additions and 471 deletions
Download Patch File Download Diff File Expand all files Collapse all files
v2/__init__.py
+6 -6
View File
@@ -16,14 +16,14 @@ adapters (see ``v2/models/*/components.py``) that are off the test path.
"""
from __future__ import annotations
from ._enums import (
from v2._enums import (
Capability,
ConsistencyLevel,
ExecutionProfile,
LoopKind,
WorkUnitKind,
)
from .card import (
from v2.card import (
CapabilityMatrix,
ComponentSpec,
CostModel,
@@ -34,8 +34,8 @@ from .card import (
RecipeSpec,
load_card,
)
from .program import ComponentNode, ModelLoopNode, Program, ProgramKind, when_opt, when_task
from .request import (
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind, when_opt, when_task
from v2.request import (
DiffusionParams,
Output,
Request,
@@ -44,7 +44,7 @@ from .request import (
TaskType,
make_request,
)
from .runtime import AsyncEngine, Engine
from v2.runtime import AsyncEngine, Engine
__version__ = "0.2.0"
@@ -63,6 +63,6 @@ def __getattr__(name: str):
# Lazy: the GPU entrypoint imports torch / fastvideo, so resolve it only on access — plain
# ``import v2`` (and the CPU-only mini) stay torch-free.
if name == "VideoGenerator":
from .video_generator import VideoGenerator
from v2.video_generator import VideoGenerator
return VideoGenerator
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
v2/cache/__init__.py
Vendored
+3 -3
View File
@@ -1,9 +1,9 @@
"""Cache plane — correct by key, per-class pools (design_v3 §7)."""
from __future__ import annotations
from .classes import FeatureCache, PagedKVCache, ResidualCache, Slab, SlabKVCache, make_pool
from .keys import CacheKey, CachePolicy, content_hash
from .manager import CacheManager
from v2.cache.classes import FeatureCache, PagedKVCache, ResidualCache, Slab, SlabKVCache, make_pool
from v2.cache.keys import CacheKey, CachePolicy, content_hash
from v2.cache.manager import CacheManager
__all__ = ["CacheKey", "CachePolicy", "content_hash", "CacheManager",
"FeatureCache", "ResidualCache", "SlabKVCache", "PagedKVCache", "Slab", "make_pool"]
v2/cache/classes.py
Vendored
+1 -1
View File
@@ -18,7 +18,7 @@ from collections import OrderedDict
from dataclasses import dataclass, field
from typing import Any
from .keys import CacheKey, CachePolicy
from v2.cache.keys import CacheKey, CachePolicy
def _nbytes(value: Any) -> int:
v2/cache/manager.py
Vendored
+2 -2
View File
@@ -8,8 +8,8 @@ from __future__ import annotations
from typing import Any
from .classes import _Pool, make_pool
from .keys import CachePolicy
from v2.cache.classes import _Pool, make_pool
from v2.cache.keys import CachePolicy
class CacheManager:
v2/card/__init__.py
+3 -3
View File
@@ -1,9 +1,9 @@
"""Model Plane — the center (design_v3 §4). The (recipe, runtime) pair as a typed card."""
from __future__ import annotations
from .._enums import Capability, ConsistencyLevel, ExecutionProfile, LoopKind, WorkUnitKind
from .instance import ModelInstance, load_card
from .specs import (
from v2._enums import Capability, ConsistencyLevel, ExecutionProfile, LoopKind, WorkUnitKind
from v2.card.instance import ModelInstance, load_card
from v2.card.specs import (
CacheContract,
CapabilityMatrix,
CardValidationError,
v2/card/instance.py
+2 -2
View File
@@ -13,8 +13,8 @@ from __future__ import annotations
from typing import Any
from ..platform import Platform
from .specs import ModelCard
from v2.platform import Platform
from v2.card.specs import ModelCard
class ModelInstance:
v2/card/specs.py
+2 -2
View File
@@ -15,10 +15,10 @@ from __future__ import annotations
from dataclasses import dataclass, field
from typing import TYPE_CHECKING, Any, Callable
from .._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
if TYPE_CHECKING: # avoid runtime card -> parallel coupling
from ..parallel.plan import ParallelPlan
from v2.parallel.plan import ParallelPlan
# --------------------------------------------------------------------------- #
v2/deploy/__init__.py
+3 -3
View File
@@ -6,9 +6,9 @@ consumers.
"""
from __future__ import annotations
from .card import DeploymentCard, HealthSchema, SLOSchema, build_deployment_card
from .dynamo import DynamoWorkerAdapter, FakeDynamoRuntime
from .fleet import LocalFleet, NoWorkerAvailable, Worker
from v2.deploy.card import DeploymentCard, HealthSchema, SLOSchema, build_deployment_card
from v2.deploy.dynamo import DynamoWorkerAdapter, FakeDynamoRuntime
from v2.deploy.fleet import LocalFleet, NoWorkerAvailable, Worker
__all__ = ["DeploymentCard", "HealthSchema", "SLOSchema", "build_deployment_card",
"LocalFleet", "Worker", "NoWorkerAvailable",
v2/deploy/card.py
+2 -2
View File
@@ -12,8 +12,8 @@ import dataclasses
from dataclasses import dataclass, field
from typing import Any
from .._enums import Capability
from ..card import CostModel
from v2._enums import Capability
from v2.card import CostModel
@dataclass
v2/deploy/dynamo.py
+2 -2
View File
@@ -12,8 +12,8 @@ from __future__ import annotations
from typing import Any
from ..request.artifacts import Output
from .card import DeploymentCard
from v2.request.artifacts import Output
from v2.deploy.card import DeploymentCard
class DynamoWorkerAdapter:
v2/deploy/fleet.py
+2 -2
View File
@@ -11,8 +11,8 @@ from __future__ import annotations
from typing import Any, AsyncIterator
from ..request.artifacts import Output
from .card import DeploymentCard, HealthSchema
from v2.request.artifacts import Output
from v2.deploy.card import DeploymentCard, HealthSchema
class NoWorkerAvailable(RuntimeError):
v2/examples.py
+9 -9
View File
@@ -8,11 +8,11 @@ from __future__ import annotations
import numpy as np
from .models import build_default_engine, build_omni_engine
from .models.wan21 import build_wan21_card
from .parity import assert_interleave_parity
from .request import DiffusionParams, OutputSpec, SamplingParams, TaskType, make_request
from .training import build_diffusion_nft
from v2.models import build_default_engine, build_omni_engine
from v2.models.wan21 import build_wan21_card
from v2.parity import assert_interleave_parity
from v2.request import DiffusionParams, OutputSpec, SamplingParams, TaskType, make_request
from v2.training import build_diffusion_nft
def _t2v(mid, prompt, seed, steps=4, **kw):
@@ -80,10 +80,10 @@ def example_g_omni_mot() -> None:
async def _serving_demo() -> None:
import asyncio
from .deploy import DynamoWorkerAdapter, FakeDynamoRuntime, LocalFleet, build_deployment_card
from .models.wan21 import build_wan21_card, build_wan_t2v_program
from .runtime import AsyncEngine, PoolSet, wan_t2v_disaggregated
from .serving import OmniOpenAIServer
from v2.deploy import DynamoWorkerAdapter, FakeDynamoRuntime, LocalFleet, build_deployment_card
from v2.models.wan21 import build_wan21_card, build_wan_t2v_program
from v2.runtime import AsyncEngine, PoolSet, wan_t2v_disaggregated
from v2.serving import OmniOpenAIServer
eng = build_default_engine()
build_omni_engine(eng)
v2/extend/__init__.py
+4 -4
View File
@@ -1,10 +1,10 @@
"""Extension plane — observers (read-only) and interceptors (compute-altering) (design_v3 §11)."""
from __future__ import annotations
from .base import Interceptor, InterceptorChain, InterceptorConflict, Observer, ObserverBus
from .interceptors import ResidualSkipInterceptor
from .observers import NaNWatch, Profiler
from .registry import PluginRegistry
from v2.extend.base import Interceptor, InterceptorChain, InterceptorConflict, Observer, ObserverBus
from v2.extend.interceptors import ResidualSkipInterceptor
from v2.extend.observers import NaNWatch, Profiler
from v2.extend.registry import PluginRegistry
__all__ = ["Observer", "Interceptor", "ObserverBus", "InterceptorChain", "InterceptorConflict",
"Profiler", "NaNWatch", "ResidualSkipInterceptor", "PluginRegistry"]
v2/extend/registry.py
+1 -1
View File
@@ -8,7 +8,7 @@ from __future__ import annotations
from typing import Any, Callable
from .base import InterceptorChain, Observer
from v2.extend.base import InterceptorChain, Observer
class PluginRegistry:
v2/loop/__init__.py
+4 -4
View File
@@ -1,8 +1,8 @@
"""Loop plane — driven loops + policies (design_v3 §5)."""
from __future__ import annotations
from .._enums import LoopKind, WorkUnitKind
from .contracts import (
from v2._enums import LoopKind, WorkUnitKind
from v2.loop.contracts import (
CacheOp,
CachePlan,
Done,
@@ -17,8 +17,8 @@ from .contracts import (
StepResult,
WorkPlan,
)
from .driver import LoopRunner
from .sampler import add_noise, build_flow_sigmas, flow_match_euler_step, x0_from_velocity
from v2.loop.driver import LoopRunner
from v2.loop.sampler import add_noise, build_flow_sigmas, flow_match_euler_step, x0_from_velocity
__all__ = [
"Loop", "LoopContext", "LoopState", "LoopResult", "WorkPlan", "Done", "StepResult",
v2/loop/contracts.py
+3 -3
View File
@@ -26,9 +26,9 @@ from __future__ import annotations
from dataclasses import dataclass, field
from typing import Any, Protocol, runtime_checkable
from .._enums import ExecutionProfile, WorkUnitKind
from .._types import TensorLike
from ..request.streams import StreamChunk
from v2._enums import ExecutionProfile, WorkUnitKind
from v2._types import TensorLike
from v2.request.streams import StreamChunk
# --------------------------------------------------------------------------- #
v2/loop/driver.py
+1 -1
View File
@@ -10,7 +10,7 @@ from __future__ import annotations
from typing import Any
from .contracts import Done, LoopContext, LoopResult, LoopState, WorkPlan
from v2.loop.contracts import Done, LoopContext, LoopResult, LoopState, WorkPlan
class LoopRunner:
v2/loop/policies/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""Policies — the default step decomposition (design_v3 §5.1, §6.2.3)."""
from __future__ import annotations
from .base import (
from v2.loop.policies.base import (
BoundaryTimestepRouting,
ConditioningInjector,
ExpertRouting,
@@ -10,7 +10,7 @@ from .base import (
PassthroughConditioning,
PrecisionPolicy,
)
from .cfg import (
from v2.loop.policies.cfg import (
AdaptiveGateCFG,
BatchedCFG,
CFGPolicy,
v2/loop/policies/base.py
+2 -2
View File
@@ -15,8 +15,8 @@ from typing import Any
import numpy as np
from ..contracts import StepContext
from ..sampler import build_flow_sigmas
from v2.loop.contracts import StepContext
from v2.loop.sampler import build_flow_sigmas
# --------------------------------------------------------------------------- #
v2/loop/policies/cfg.py
+1 -1
View File
@@ -17,7 +17,7 @@ from typing import Any
import numpy as np
from ..contracts import StepContext
from v2.loop.contracts import StepContext
class CFGPolicy(ABC):
v2/memory/__init__.py
+1 -1
View File
@@ -1,6 +1,6 @@
"""Memory plane — reservation before admission, sleep/wake by tag (design_v3 §7.3)."""
from __future__ import annotations
from .allocator import MemoryManager, OutOfMemory, Reservation
from v2.memory.allocator import MemoryManager, OutOfMemory, Reservation
__all__ = ["MemoryManager", "OutOfMemory", "Reservation"]
v2/models/__init__.py
+28 -28
View File
@@ -7,11 +7,11 @@ from __future__ import annotations
from typing import Any
from .adaptive import build_adaptive_card
from .adapters import build_adapter_card, build_adapter_program
from .bagel import build_bagel_card, build_bagel_program
from .cosmos3 import build_cosmos3_card, build_cosmos3_program
from .image_video import (
from v2.models.adaptive import build_adaptive_card
from v2.models.adapters import build_adapter_card, build_adapter_program
from v2.models.bagel import build_bagel_card, build_bagel_program
from v2.models.cosmos3 import build_cosmos3_card, build_cosmos3_program
from v2.models.image_video import (
build_flux_t2i_card,
build_flux_t2i_program,
build_t2i_i2v_extend_workflow,
@@ -19,14 +19,14 @@ from .image_video import (
build_wan_i2v_card,
build_wan_i2v_program,
)
from .ltx2 import build_ltx2_av_program, build_ltx2_card, build_ltx2_program
from .qwen_omni import build_qwen_omni_card, build_qwen_omni_program
from .reward import build_reward_card
from .speculative import build_speculative_card, build_speculative_program
from .tiled import build_tiled_card, build_tiled_program
from .unified import build_unified_card, build_unified_program
from .wan21 import build_wan21_card, build_wan_t2v_program
from .wan_causal import build_wan_causal_card, build_wan_causal_program
from v2.models.ltx2 import build_ltx2_av_program, build_ltx2_card, build_ltx2_program
from v2.models.qwen_omni import build_qwen_omni_card, build_qwen_omni_program
from v2.models.reward import build_reward_card
from v2.models.speculative import build_speculative_card, build_speculative_program
from v2.models.tiled import build_tiled_card, build_tiled_program
from v2.models.unified import build_unified_card, build_unified_program
from v2.models.wan21 import build_wan21_card, build_wan_t2v_program
from v2.models.wan_causal import build_wan_causal_card, build_wan_causal_program
__all__ = [
"build_wan21_card", "build_wan_t2v_program",
@@ -63,9 +63,9 @@ _OMNI_BUILDERS = [
def build_default_engine(engine: Any = None) -> Any:
"""Register Wan2.1, LTX2.3, and Wan-causal onto one engine (one resident instance each)."""
from ..cache import CacheManager
from ..card import load_card
from ..runtime import Engine
from v2.cache import CacheManager
from v2.card import load_card
from v2.runtime import Engine
eng = engine if engine is not None else Engine()
for build_card, build_program in _BUILDERS:
card = build_card()
@@ -80,9 +80,9 @@ def build_omni_engine(engine: Any = None) -> Any:
Each is ONE resident MoT instance whose ``transformer`` runs BOTH an ar_decode loop and a
diffusion_denoise loop (shared weights) — the §16 'true omni/MoT serving' claim made native.
"""
from ..cache import CacheManager
from ..card import load_card
from ..runtime import Engine
from v2.cache import CacheManager
from v2.card import load_card
from v2.runtime import Engine
eng = engine if engine is not None else Engine()
for build_card, build_program in _OMNI_BUILDERS:
card = build_card()
@@ -104,8 +104,8 @@ _WORKFLOWS: dict[str, tuple] = {
def register_workflows(engine: Any, *, only: list[str] | None = None) -> Any:
"""Register catalog workflows (and the cards they require) onto ``engine``. ``only`` selects a
subset by workflow_id; default registers all whose cards aren't yet present."""
from ..cache import CacheManager
from ..card import load_card
from v2.cache import CacheManager
from v2.card import load_card
names = only if only is not None else list(_WORKFLOWS)
for wf_id in names:
build_workflow, card_builders = _WORKFLOWS[wf_id]
@@ -122,7 +122,7 @@ def build_image_video_engine(engine: Any = None) -> Any:
"""Register the T2I (``flux-t2i``) and I2V (``wan-i2v``) cards plus the ``image_video.t2i_i2v``
workflow on one engine — two *separate* models chained by a cross-model workflow, addressable by
its workflow_id like any servable (design_v3 §13)."""
from ..runtime import Engine
from v2.runtime import Engine
eng = engine if engine is not None else Engine()
return register_workflows(eng, only=["image_video.t2i_i2v"])
@@ -131,9 +131,9 @@ def build_tiled_engine(engine: Any = None) -> Any:
"""Register the tiled-decode card (``wan-tiled``, whose VAE decode emits ``VAE_TILE`` work units)
alongside plain Wan2.1 — so an interleaved batch mixes ``VAE_TILE`` and ``DIFFUSION_STEP`` units
through one admission/scheduler budget (the §17 heterogeneous-co-scheduling probe)."""
from ..cache import CacheManager
from ..card import load_card
from ..runtime import Engine
from v2.cache import CacheManager
from v2.card import load_card
from v2.runtime import Engine
eng = engine if engine is not None else Engine()
for build_card, build_program in [(build_tiled_card, build_tiled_program),
(build_wan21_card, build_wan_t2v_program)]:
@@ -147,9 +147,9 @@ def build_unified_engine(engine: Any = None) -> Any:
"""Register the unified LM+generator card (UniRL/PromptRL): a prompt-refiner ``llm`` expert + a
flow ``transformer`` expert on the SAME engine, each driving its own loop. Two separate experts,
one request, both trainable under one RL reward (the §10 joint-RL stress test)."""
from ..cache import CacheManager
from ..card import load_card
from ..runtime import Engine
from v2.cache import CacheManager
from v2.card import load_card
from v2.runtime import Engine
eng = engine if engine is not None else Engine()
card = build_unified_card()
inst = load_card(card, cache_manager=CacheManager.from_card(card))
v2/models/adapters/__init__.py
+3 -3
View File
@@ -1,8 +1,8 @@
"""Adapter plane: one base + swappable LoRA/ControlNet adapters, selected per request (design_v3 §9.19)."""
from __future__ import annotations
from .card import ADAPTERS, build_adapter_card
from .loop import AdapterDenoiseLoop
from .program import build_adapter_program
from v2.models.adapters.card import ADAPTERS, build_adapter_card
from v2.models.adapters.loop import AdapterDenoiseLoop
from v2.models.adapters.program import build_adapter_program
__all__ = ["build_adapter_card", "build_adapter_program", "AdapterDenoiseLoop", "ADAPTERS"]
v2/models/adapters/card.py
+6 -6
View File
@@ -6,8 +6,8 @@ apply (``DiffusionParams.adapters``). Adapters are versioned independently (the
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -19,10 +19,10 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyControlNet, ToyDiT, ToyLoRA, ToyTextEncoder, ToyVAE, _seed_from
from .loop import AdapterDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyControlNet, ToyDiT, ToyLoRA, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.adapters.loop import AdapterDenoiseLoop
ADAPTERS = ("lora_anime", "lora_realistic", "control_pose") # the served adapter library
v2/models/adapters/loop.py
+3 -3
View File
@@ -10,9 +10,9 @@ from __future__ import annotations
import numpy as np
from ...loop.contracts import Done
from ...platform import FLOW_MATCH_STEP
from ..wan21.loop import WanDenoiseLoop
from v2.loop.contracts import Done
from v2.platform import FLOW_MATCH_STEP
from v2.models.wan21.loop import WanDenoiseLoop
class AdapterDenoiseLoop(WanDenoiseLoop):
v2/models/adapters/program.py
+2 -2
View File
@@ -3,8 +3,8 @@ from __future__ import annotations
import numpy as np
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn as _text_encode
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn as _text_encode
def _control_signal(instance, slots, request, ctx) -> None:
v2/models/adaptive/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""Adaptive-compute denoise: the loop owns cache-dit skip + early-exit (design_v3 §2.2, §9.12)."""
from __future__ import annotations
from .card import build_adaptive_card
from .loop import CacheDiTDenoiseLoop
from v2.models.adaptive.card import build_adaptive_card
from v2.models.adaptive.loop import CacheDiTDenoiseLoop
__all__ = ["build_adaptive_card", "CacheDiTDenoiseLoop"]
v2/models/adaptive/card.py
+6 -6
View File
@@ -3,8 +3,8 @@
``diffusion_denoise``); only the loop differs."""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -16,10 +16,10 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from .loop import CacheDiTDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.adaptive.loop import CacheDiTDenoiseLoop
def build_adaptive_card(model_id: str = "wan-adaptive", *, cache_threshold: float = 0.02,
v2/models/adaptive/loop.py
+3 -3
View File
@@ -16,9 +16,9 @@ from __future__ import annotations
import numpy as np
from ...loop.contracts import Done, StepResult
from ...platform import FLOW_MATCH_STEP
from ..wan21.loop import WanDenoiseLoop
from v2.loop.contracts import Done, StepResult
from v2.platform import FLOW_MATCH_STEP
from v2.models.wan21.loop import WanDenoiseLoop
class CacheDiTDenoiseLoop(WanDenoiseLoop):
v2/models/bagel/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""BAGEL/lance — canonical vllm-omni MoT (AR generate_text + diffusion generate_image). Phase 2."""
from __future__ import annotations
from .card import build_bagel_card
from .program import build_bagel_program
from v2.models.bagel.card import build_bagel_card
from v2.models.bagel.program import build_bagel_program
__all__ = ["build_bagel_card", "build_bagel_program"]
v2/models/bagel/card.py
+7 -7
View File
@@ -11,8 +11,8 @@ differentiation the §1 thesis names.
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -24,11 +24,11 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyMoTDiT, ToyTokenizer, ToyVAE, _seed_from
from ..omni import ARDecodeLoop
from ..wan21.loop import WanDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyMoTDiT, ToyTokenizer, ToyVAE, _seed_from
from v2.models.omni import ARDecodeLoop
from v2.models.wan21.loop import WanDenoiseLoop
def build_bagel_card(model_id: str = "bagel-mot") -> ModelCard:
v2/models/bagel/program.py
+2 -2
View File
@@ -7,8 +7,8 @@ scheduler interleaves — the shared-weight MoT made native, not a monolith bypa
"""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..omni import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.omni import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
def build_bagel_program() -> Program:
v2/models/common.py
+1 -1
View File
@@ -7,7 +7,7 @@ from __future__ import annotations
from typing import Any
from ..cache.keys import CacheKey, content_hash
from v2.cache.keys import CacheKey, content_hash
def cached_text_encode(instance: Any, text: str) -> Any:
v2/models/cosmos3/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""Cosmos3 — omni MoT (reasoner + joint denoise on one resident instance). Phase 2."""
from __future__ import annotations
from .card import build_cosmos3_card
from .program import build_cosmos3_program
from v2.models.cosmos3.card import build_cosmos3_card
from v2.models.cosmos3.program import build_cosmos3_program
__all__ = ["build_cosmos3_card", "build_cosmos3_program"]
v2/models/cosmos3/card.py
+7 -7
View File
@@ -10,8 +10,8 @@ MoT module. ``sound_vae`` is declared ``optional_for`` non-t2vs tasks (the lazy-
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -23,11 +23,11 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyMoTDiT, ToyTokenizer, ToyVAE, _seed_from
from ..omni import ARDecodeLoop
from ..wan21.loop import WanDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyMoTDiT, ToyTokenizer, ToyVAE, _seed_from
from v2.models.omni import ARDecodeLoop
from v2.models.wan21.loop import WanDenoiseLoop
def build_cosmos3_card(model_id: str = "cosmos3-vfm") -> ModelCard:
v2/models/cosmos3/program.py
+2 -2
View File
@@ -8,8 +8,8 @@ and severs the shared state).
"""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..omni import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.omni import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
def build_cosmos3_program() -> Program:
v2/models/image_video/__init__.py
+2 -2
View File
@@ -3,8 +3,8 @@
the design composes across model *instances*, not just loops within one (which LTX-2 already covers)."""
from __future__ import annotations
from .card import build_flux_t2i_card, build_wan_i2v_card
from .program import (
from v2.models.image_video.card import build_flux_t2i_card, build_wan_i2v_card
from v2.models.image_video.program import (
build_flux_t2i_program,
build_t2i_i2v_extend_workflow,
build_t2i_then_i2v_workflow,
v2/models/image_video/card.py
+6 -6
View File
@@ -11,8 +11,8 @@ is the point: a new pipeline is cards + a program + a workflow, never a new loop
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -24,10 +24,10 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from ..wan21.loop import WanDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.wan21.loop import WanDenoiseLoop
def _diffusion_card(model_id, family, loop_id, caps, tasks, *, shift):
v2/models/image_video/program.py
+4 -4
View File
@@ -15,10 +15,10 @@ from __future__ import annotations
import numpy as np
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind, Workflow, WorkflowStage
from ...request import DiffusionParams, TaskType, make_request
from ...request.modalpart import ImagePart, TextPart
from ..common import text_encode_node_fn as _text_encode
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind, Workflow, WorkflowStage
from v2.request import DiffusionParams, TaskType, make_request
from v2.request.modalpart import ImagePart, TextPart
from v2.models.common import text_encode_node_fn as _text_encode
# --- stage 1: text → image -------------------------------------------------------- #
v2/models/ltx2/__init__.py
+3 -3
View File
@@ -1,9 +1,9 @@
"""LTX2.3 — two-stage distilled video diffusion (base → upsample → refine)."""
from __future__ import annotations
from .card import build_ltx2_3_card, build_ltx2_base_card, build_ltx2_card
from .loop import BASE_SIGMAS, REFINE_SIGMAS, LTX2DenoiseLoop, LTX23DenoiseLoop
from .program import (
from v2.models.ltx2.card import build_ltx2_3_card, build_ltx2_base_card, build_ltx2_card
from v2.models.ltx2.loop import BASE_SIGMAS, REFINE_SIGMAS, LTX2DenoiseLoop, LTX23DenoiseLoop
from v2.models.ltx2.program import (
build_ltx2_3_program,
build_ltx2_av_program,
build_ltx2_base_program,
v2/models/ltx2/card.py
+5 -5
View File
@@ -14,8 +14,8 @@ Real wiring: DiT ``ltx2:LTX2Transformer3DModel``, causal VAE ``ltx2vae``, Gemma
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -27,9 +27,9 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...parallel import ParallelPlan
from ..backend import ToyAudioVAE, ToyDiT, ToyTextEncoder, ToyUpsampler, ToyVAE, ToyVocoder, _seed_from
from .loop import BASE_SIGMAS, REFINE_SIGMAS, LTX2DenoiseLoop, LTX23DenoiseLoop
from v2.parallel import ParallelPlan
from v2.models.backend import ToyAudioVAE, ToyDiT, ToyTextEncoder, ToyUpsampler, ToyVAE, ToyVocoder, _seed_from
from v2.models.ltx2.loop import BASE_SIGMAS, REFINE_SIGMAS, LTX2DenoiseLoop, LTX23DenoiseLoop
def build_ltx2_card(model_id: str = "ltx2-2stage-distilled") -> ModelCard:
v2/models/ltx2/loop.py
+4 -4
View File
@@ -18,8 +18,8 @@ from __future__ import annotations
import numpy as np
from ..._enums import ExecutionProfile, WorkUnitKind
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -28,8 +28,8 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...platform import FLOW_MATCH_STEP
from ..backend import LATENT_CHANNELS
from v2.platform import FLOW_MATCH_STEP
from v2.models.backend import LATENT_CHANNELS
BASE_SIGMAS = [1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0]
REFINE_SIGMAS = [0.909375, 0.725, 0.421875, 0.0]
v2/models/ltx2/program.py
+3 -3
View File
@@ -9,9 +9,9 @@ from __future__ import annotations
import numpy as np
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn as _text_encode
from .loop import REFINE_SIGMAS
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn as _text_encode
from v2.models.ltx2.loop import REFINE_SIGMAS
def _upsample(instance, slots, request, ctx) -> None:
v2/models/multi_expert/__init__.py
+1 -1
View File
@@ -2,6 +2,6 @@
substrate for N-way joint RL (design_v3 §4, §10). Generalizes the two-expert ``unified`` card."""
from __future__ import annotations
from .card import N_REFINE_ACTIONS, build_multi_expert_card, refiner_ids
from v2.models.multi_expert.card import N_REFINE_ACTIONS, build_multi_expert_card, refiner_ids
__all__ = ["build_multi_expert_card", "refiner_ids", "N_REFINE_ACTIONS"]
v2/models/multi_expert/card.py
+7 -7
View File
@@ -13,8 +13,8 @@ offset), so the generator's output depends on every refiner — and one reward t
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -26,11 +26,11 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyPromptRefiner, ToyTextEncoder, ToyVAE, _seed_from
from ..omni import ARDecodeLoop
from ..wan21.loop import WanDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyPromptRefiner, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.omni import ARDecodeLoop
from v2.models.wan21.loop import WanDenoiseLoop
N_REFINE_ACTIONS = 6
v2/models/omni/__init__.py
+3 -3
View File
@@ -1,9 +1,9 @@
"""Shared omni building blocks (design_v3 §4): the AR decode loop + pack/emit program nodes."""
from __future__ import annotations
from .ar_loop import ARDecodeLoop
from .pack import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
from .vocoder_loop import VocoderLoop
from v2.models.omni.ar_loop import ARDecodeLoop
from v2.models.omni.pack import emit_text_node, pack_cond_from_tokens, tokenize_node, vae_decode_node
from v2.models.omni.vocoder_loop import VocoderLoop
__all__ = ["ARDecodeLoop", "VocoderLoop", "tokenize_node", "pack_cond_from_tokens",
"emit_text_node", "vae_decode_node"]
v2/models/omni/ar_loop.py
+3 -3
View File
@@ -8,8 +8,8 @@ claim). All per-request state (the token list, paged-KV namespace) lives in ``Lo
"""
from __future__ import annotations
from ..._enums import ExecutionProfile, WorkUnitKind
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -18,7 +18,7 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...request.streams import StreamChunk
from v2.request.streams import StreamChunk
class ARDecodeLoop:
v2/models/omni/pack.py
+1 -1
View File
@@ -7,7 +7,7 @@ module, so this is weight-sharing, not a hand-off between two engines.
"""
from __future__ import annotations
from ..backend import ToyTokenizer
from v2.models.backend import ToyTokenizer
def tokenize_node(instance, slots, request, ctx) -> None:
v2/models/omni/vocoder_loop.py
+3 -3
View File
@@ -11,8 +11,8 @@ from __future__ import annotations
import numpy as np
from ..._enums import ExecutionProfile, WorkUnitKind
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -21,7 +21,7 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...request.streams import StreamChunk
from v2.request.streams import StreamChunk
class VocoderLoop:
v2/models/qwen_omni/__init__.py
+2 -2
View File
@@ -3,7 +3,7 @@ loops (ar_decode → ar_decode → audio_decode), cascaded conditioning + stream
third weight-sharing topology proving the Card/Loop/Program vocabulary holds (design_v3 §4)."""
from __future__ import annotations
from .card import build_qwen_omni_card
from .program import build_qwen_omni_program
from v2.models.qwen_omni.card import build_qwen_omni_card
from v2.models.qwen_omni.program import build_qwen_omni_program
__all__ = ["build_qwen_omni_card", "build_qwen_omni_program"]
v2/models/qwen_omni/card.py
+5 -5
View File
@@ -19,8 +19,8 @@ two language stages, ``LoopKind.AUDIO_DECODE`` for the vocoder — no new primit
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -32,9 +32,9 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...parallel import ParallelPlan
from ..backend import ToyMoTDiT, ToyTalker, ToyTokenizer, ToyVocoder, _seed_from
from ..omni import ARDecodeLoop, VocoderLoop
from v2.parallel import ParallelPlan
from v2.models.backend import ToyMoTDiT, ToyTalker, ToyTokenizer, ToyVocoder, _seed_from
from v2.models.omni import ARDecodeLoop, VocoderLoop
def build_qwen_omni_card(model_id: str = "qwen-omni-tts") -> ModelCard:
v2/models/qwen_omni/program.py
+2 -2
View File
@@ -17,8 +17,8 @@ from __future__ import annotations
import numpy as np
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..omni import emit_text_node, tokenize_node
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.omni import emit_text_node, tokenize_node
def thinker_to_talker_node(instance, slots, request, ctx) -> None:
v2/models/reward/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""Reward-model card: a served scorer/verifier emitting REWARD_BATCH work units (design_v3 §10)."""
from __future__ import annotations
from .card import build_reward_card
from .loop import RewardLoop
from v2.models.reward.card import build_reward_card
from v2.models.reward.loop import RewardLoop
__all__ = ["build_reward_card", "RewardLoop"]
v2/models/reward/card.py
+5 -5
View File
@@ -7,8 +7,8 @@ run on its own role-pool (reward disaggregation).
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CapabilityMatrix,
ComponentSpec,
CostModel,
@@ -19,9 +19,9 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...parallel import ParallelPlan
from ..backend import ToyRewardModel, _seed_from
from .loop import RewardLoop
from v2.parallel import ParallelPlan
from v2.models.backend import ToyRewardModel, _seed_from
from v2.models.reward.loop import RewardLoop
def build_reward_card(model_id: str = "pickscore-reward", *, batch: int = 4) -> ModelCard:
v2/models/reward/loop.py
+2 -2
View File
@@ -8,8 +8,8 @@ one (or a few) batched units — the homogeneous-batch win the scheduler likes.
"""
from __future__ import annotations
from ..._enums import WorkUnitKind
from ...loop.contracts import (
from v2._enums import WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
v2/models/speculative/__init__.py
+3 -3
View File
@@ -1,8 +1,8 @@
"""Speculative (draft-verify) decoding: a draft + target AR pair, exact + lower-latency (design_v3 §9.16)."""
from __future__ import annotations
from .card import build_speculative_card
from .loop import SpeculativeARLoop
from .program import build_speculative_program
from v2.models.speculative.card import build_speculative_card
from v2.models.speculative.loop import SpeculativeARLoop
from v2.models.speculative.program import build_speculative_program
__all__ = ["build_speculative_card", "build_speculative_program", "SpeculativeARLoop"]
v2/models/speculative/card.py
+5 -5
View File
@@ -1,8 +1,8 @@
"""Speculative-decoding card — a draft + target AR pair on one instance (design_v3 §2.2, §9.16)."""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -14,9 +14,9 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...parallel import ParallelPlan
from ..backend import ToyDraftModel, ToyTargetModel, ToyTokenizer, _seed_from
from .loop import SpeculativeARLoop
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDraftModel, ToyTargetModel, ToyTokenizer, _seed_from
from v2.models.speculative.loop import SpeculativeARLoop
def build_speculative_card(model_id: str = "spec-decode", *, spec_len: int = 4,
v2/models/speculative/loop.py
+3 -3
View File
@@ -16,8 +16,8 @@ one request — and every round is an ``AR_TOKEN`` WorkUnit.
"""
from __future__ import annotations
from ..._enums import ExecutionProfile, WorkUnitKind
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -26,7 +26,7 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...request.streams import StreamChunk
from v2.request.streams import StreamChunk
class SpeculativeARLoop:
v2/models/speculative/program.py
+2 -2
View File
@@ -1,8 +1,8 @@
"""Speculative-decoding program: tokenize → spec_decode (draft+verify) → emit_text (design_v3 §9.16)."""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..omni import emit_text_node, tokenize_node
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.omni import emit_text_node, tokenize_node
def build_speculative_program() -> Program:
v2/models/tiled/__init__.py
+3 -3
View File
@@ -1,8 +1,8 @@
"""Tiled VAE decode: the VAE_TILE WorkUnit kind co-scheduled with denoise steps (design_v3 §4, §17)."""
from __future__ import annotations
from .card import build_tiled_card
from .loop import VAETileLoop
from .program import build_tiled_program
from v2.models.tiled.card import build_tiled_card
from v2.models.tiled.loop import VAETileLoop
from v2.models.tiled.program import build_tiled_program
__all__ = ["build_tiled_card", "build_tiled_program", "VAETileLoop"]
v2/models/tiled/card.py
+7 -7
View File
@@ -2,8 +2,8 @@
``vae_tile`` loop, so its tiles co-schedule with denoise steps under the one WorkUnit budget."""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -15,11 +15,11 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from ..wan21.loop import WanDenoiseLoop
from .loop import VAETileLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.wan21.loop import WanDenoiseLoop
from v2.models.tiled.loop import VAETileLoop
def build_tiled_card(model_id: str = "wan-tiled", *, tile_rows: int = 1) -> ModelCard:
v2/models/tiled/loop.py
+2 -2
View File
@@ -14,8 +14,8 @@ from __future__ import annotations
import numpy as np
from ..._enums import WorkUnitKind
from ...loop.contracts import (
from v2._enums import WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
v2/models/tiled/program.py
+2 -2
View File
@@ -6,8 +6,8 @@ units. (Contrast the plain Wan program, where decode is a single in-tick node.)
"""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn as _text_encode
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn as _text_encode
def build_tiled_program() -> Program:
v2/models/unified/__init__.py
+2 -2
View File
@@ -3,7 +3,7 @@ both trainable under one RL reward. The stress test that the Card/Loop/Program s
joint multi-expert RL (design_v3 §4, §10)."""
from __future__ import annotations
from .card import N_REFINE_ACTIONS, build_unified_card
from .program import apply_refinement_node, build_unified_program
from v2.models.unified.card import N_REFINE_ACTIONS, build_unified_card
from v2.models.unified.program import apply_refinement_node, build_unified_program
__all__ = ["build_unified_card", "build_unified_program", "apply_refinement_node", "N_REFINE_ACTIONS"]
v2/models/unified/card.py
+7 -7
View File
@@ -16,8 +16,8 @@ capture, and serve path are unchanged.
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -29,11 +29,11 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyPromptRefiner, ToyTextEncoder, ToyVAE, _seed_from
from ..omni import ARDecodeLoop
from ..wan21.loop import WanDenoiseLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, NoRouting, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyPromptRefiner, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.omni import ARDecodeLoop
from v2.models.wan21.loop import WanDenoiseLoop
N_REFINE_ACTIONS = 8
v2/models/unified/program.py
+4 -4
View File
@@ -10,10 +10,10 @@ profiles — the §10 "train ≡ serve through one loop" invariant, now spanning
"""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn
from ..omni import vae_decode_node
from .card import N_REFINE_ACTIONS
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn
from v2.models.omni import vae_decode_node
from v2.models.unified.card import N_REFINE_ACTIONS
def apply_refinement_node(refine_slot: str, n_actions: int = N_REFINE_ACTIONS):
v2/models/wan21/__init__.py
+3 -3
View File
@@ -1,14 +1,14 @@
"""Wan2.1-1.3B — bidirectional video diffusion (T2V)."""
from __future__ import annotations
from .card import (
from v2.models.wan21.card import (
build_wan21_card,
build_wan22_a14b_card,
build_wan22_ti2v_card,
stamp_wan21_checkpoints,
)
from .loop import WanDenoiseLoop, latent_shape
from .program import build_wan_t2v_program
from v2.models.wan21.loop import WanDenoiseLoop, latent_shape
from v2.models.wan21.program import build_wan_t2v_program
__all__ = ["build_wan21_card", "build_wan22_ti2v_card", "build_wan22_a14b_card",
"stamp_wan21_checkpoints", "build_wan_t2v_program", "WanDenoiseLoop", "latent_shape"]
v2/models/wan21/card.py
+6 -6
View File
@@ -9,8 +9,8 @@ Mirrors the real Wan2.1-1.3B wiring discovered in the repo:
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -23,16 +23,16 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import (
from v2.loop.policies import (
BoundaryTimestepRouting,
ClassicCFG,
FlowShiftPolicy,
NoRouting,
PrecisionPolicy,
)
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from .loop import WanDenoiseLoop
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.wan21.loop import WanDenoiseLoop
def build_wan21_card(model_id: str = "wan2.1-1.3b", *, cfg_policy=None, flow_shift: float = 3.0,
v2/models/wan21/loop.py
+5 -5
View File
@@ -10,8 +10,8 @@ from __future__ import annotations
import numpy as np
from ..._enums import ExecutionProfile, WorkUnitKind
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -21,9 +21,9 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...platform import FLOW_MATCH_STEP, FLOW_SDE_STEP
from ...request.streams import StreamChunk
from ..backend import LATENT_CHANNELS
from v2.platform import FLOW_MATCH_STEP, FLOW_SDE_STEP
from v2.request.streams import StreamChunk
from v2.models.backend import LATENT_CHANNELS
# Real Wan2.1 VAE (AutoencoderKLWan) compression: z_dim=16, 4x temporal, 8x spatial.
v2/models/wan21/program.py
+2 -2
View File
@@ -6,8 +6,8 @@ reduction the vendored RL sampler cannot express.
"""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn as _text_encode
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn as _text_encode
def _vae_decode(instance, slots, request, ctx) -> None:
v2/models/wan_causal/__init__.py
+3 -3
View File
@@ -1,8 +1,8 @@
"""Wan-causal — causal/streaming video (chunk rollout + slab-KV), the self-forcing student."""
from __future__ import annotations
from .card import build_wan_causal_card
from .loop import ChunkRolloutLoop
from .program import build_wan_causal_program
from v2.models.wan_causal.card import build_wan_causal_card
from v2.models.wan_causal.loop import ChunkRolloutLoop
from v2.models.wan_causal.program import build_wan_causal_program
__all__ = ["build_wan_causal_card", "build_wan_causal_program", "ChunkRolloutLoop"]
v2/models/wan_causal/card.py
+6 -6
View File
@@ -7,8 +7,8 @@ this card by the validator (stale KV breaks causality).
"""
from __future__ import annotations
from ..._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from ...card import (
from v2._enums import Capability, ConsistencyLevel, LoopKind, WorkUnitKind
from v2.card import (
CacheContract,
CapabilityMatrix,
ComponentSpec,
@@ -20,10 +20,10 @@ from ...card import (
PrecisionContract,
RecipeSpec,
)
from ...loop.policies import ClassicCFG, FlowShiftPolicy, PrecisionPolicy
from ...parallel import ParallelPlan
from ..backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from .loop import ChunkRolloutLoop
from v2.loop.policies import ClassicCFG, FlowShiftPolicy, PrecisionPolicy
from v2.parallel import ParallelPlan
from v2.models.backend import ToyDiT, ToyTextEncoder, ToyVAE, _seed_from
from v2.models.wan_causal.loop import ChunkRolloutLoop
def build_wan_causal_card(model_id: str = "wan-causal-sf-1.3b", *,
v2/models/wan_causal/loop.py
+6 -6
View File
@@ -12,9 +12,9 @@ from __future__ import annotations
import numpy as np
from ..._enums import ExecutionProfile, WorkUnitKind
from ...cache.classes import Slab
from ...loop.contracts import (
from v2._enums import ExecutionProfile, WorkUnitKind
from v2.cache.classes import Slab
from v2.loop.contracts import (
Done,
LoopResult,
LoopState,
@@ -24,9 +24,9 @@ from ...loop.contracts import (
StepResult,
WorkPlan,
)
from ...platform import FLOW_MATCH_STEP
from ...request.streams import StreamChunk
from ..backend import LATENT_CHANNELS
from v2.platform import FLOW_MATCH_STEP
from v2.request.streams import StreamChunk
from v2.models.backend import LATENT_CHANNELS
class ChunkRolloutLoop:
v2/models/wan_causal/program.py
+2 -2
View File
@@ -1,8 +1,8 @@
"""Wan-causal streaming program: text_encode → chunk_rollout → vae_decode (design_v3 §4, §15e)."""
from __future__ import annotations
from ...program import ComponentNode, ModelLoopNode, Program, ProgramKind
from ..common import text_encode_node_fn as _text_encode
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
from v2.models.common import text_encode_node_fn as _text_encode
def _vae_decode(instance, slots, request, ctx) -> None:
v2/parallel/__init__.py
+3 -3
View File
@@ -1,9 +1,9 @@
"""Parallelism plane (design_v3 §8) — named axes → validated mesh, part of the cache key."""
from __future__ import annotations
from .mesh import FakeDeviceMesh, build_mesh
from .plan import AXIS_NAMES, ParallelPlan
from .validation import ParallelValidationError, validate_plan
from v2.parallel.mesh import FakeDeviceMesh, build_mesh
from v2.parallel.plan import AXIS_NAMES, ParallelPlan
from v2.parallel.validation import ParallelValidationError, validate_plan
__all__ = ["ParallelPlan", "AXIS_NAMES", "FakeDeviceMesh", "build_mesh",
"validate_plan", "ParallelValidationError"]
v2/parallel/mesh.py
+2 -2
View File
@@ -11,8 +11,8 @@ from __future__ import annotations
from itertools import product
from .plan import ParallelPlan
from .validation import validate_plan
from v2.parallel.plan import ParallelPlan
from v2.parallel.validation import validate_plan
class FakeDeviceMesh:
v2/parallel/validation.py
+2 -2
View File
@@ -7,8 +7,8 @@ All checks are CPU-testable on a fake mesh (design.md §6.3.4: CPU-testability).
"""
from __future__ import annotations
from .._enums import LoopKind
from .plan import AXIS_NAMES, ParallelPlan
from v2._enums import LoopKind
from v2.parallel.plan import AXIS_NAMES, ParallelPlan
class ParallelValidationError(ValueError):
v2/parity/__init__.py
+4 -4
View File
@@ -5,10 +5,10 @@ the consistency ladder is typed, and the interleave gate is non-negotiable.
"""
from __future__ import annotations
from .._enums import ConsistencyLevel, ExecutionProfile
from .aligner import ParityAligner
from .interleave_gate import assert_interleave_parity, compare_outputs
from .ladder import Divergence, array_diff, bit_identical, within
from v2._enums import ConsistencyLevel, ExecutionProfile
from v2.parity.aligner import ParityAligner
from v2.parity.interleave_gate import assert_interleave_parity, compare_outputs
from v2.parity.ladder import Divergence, array_diff, bit_identical, within
__all__ = ["ConsistencyLevel", "ExecutionProfile", "ParityAligner", "Divergence",
"array_diff", "bit_identical", "within", "assert_interleave_parity", "compare_outputs"]
v2/parity/aligner.py
+1 -1
View File
@@ -10,7 +10,7 @@ from __future__ import annotations
from typing import Any
from .ladder import ConsistencyLevel, Divergence, array_diff
from v2.parity.ladder import ConsistencyLevel, Divergence, array_diff
class ParityAligner:
v2/parity/interleave_gate.py
+2 -2
View File
@@ -12,7 +12,7 @@ from __future__ import annotations
from typing import Any
from ..request.artifacts import (
from v2.request.artifacts import (
AudioArtifact,
LatentArtifact,
Output,
@@ -20,7 +20,7 @@ from ..request.artifacts import (
TextArtifact,
VideoArtifact,
)
from .ladder import ConsistencyLevel, Divergence, bit_identical
from v2.parity.ladder import ConsistencyLevel, Divergence, bit_identical
def _artifact_payload(art: Any) -> Any:
v2/parity/ladder.py
+1 -1
View File
@@ -13,7 +13,7 @@ from typing import Any
import numpy as np
from .._enums import ConsistencyLevel # re-exported for the package
from v2._enums import ConsistencyLevel # re-exported for the package
@dataclass
v2/platform/__init__.py
+2 -2
View File
@@ -11,14 +11,14 @@ backend registrations in ``backends/`` are imported lazily on first ``Platform``
"""
from __future__ import annotations
from .platform import (
from v2.platform.platform import (
KernelTable,
Platform,
component_matrix,
ensure_backends_loaded,
kernel_matrix,
)
from .registry import (
from v2.platform.registry import (
COMPONENTS,
FLOW_MATCH_STEP,
FLOW_SDE_STEP,
v2/platform/backends/accel.py
+3 -3
View File
@@ -25,8 +25,8 @@ from __future__ import annotations
from typing import Any
from ...models.backend import ToyDiT
from ..registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_component, register_kernel
from v2.models.backend import ToyDiT
from v2.platform.registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_component, register_kernel
# --------------------------------------------------------------------------- #
@@ -100,7 +100,7 @@ register_kernel(FLOW_MATCH_STEP, _accel_flow_match_step,
# reuses the numpy reference, so the accel SDE path is the same function as cpu (the SDE parity test
# therefore checks dispatch-path equivalence, not an independent impl; the ODE op above is the real
# independent oracle). A device that DID specialize SDE would be checked the same way the ODE op is.
from ...loop.sampler import flow_sde_step_with_logprob # noqa: E402
from v2.loop.sampler import flow_sde_step_with_logprob # noqa: E402
register_kernel(FLOW_SDE_STEP, flow_sde_step_with_logprob,
device="accel", arch="sm90", source="accel(stand-in):sde_logprob")
v2/platform/backends/cpu.py
+2 -2
View File
@@ -10,8 +10,8 @@ matches). That keeps every existing card working with zero registration.
"""
from __future__ import annotations
from ...loop.sampler import flow_match_euler_step, flow_sde_step_with_logprob
from ..registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_kernel
from v2.loop.sampler import flow_match_euler_step, flow_sde_step_with_logprob
from v2.platform.registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_kernel
# The numpy solver primitives, registered as the terminal (cpu, numpy) kernels. These are the exact
# same functions the loops used to call directly — so dispatching through the registry on a CPU box
v2/platform/backends/torch_adapters.py
+6 -6
View File
@@ -308,7 +308,7 @@ def build_torch_dit(spec, instance, platform):
module = _load_component("TransformerLoader", ckpt, _fastvideo_args(spec))
device, dtype = _device(platform), _native_dtype(module)
if "LTX2" in type(module).__name__: # LTX2Transformer3DModel
from .torch_ltx2 import TorchLTX2DiT
from v2.platform.backends.torch_ltx2 import TorchLTX2DiT
return TorchLTX2DiT(module, device=device, dtype=dtype)
# Wan2.2 MoE (A14B): >1 DiT expert -> CPU-offload all but the active one (swapped at the boundary)
# so the two 14B experts fit one 80GB GPU. A single-expert Wan stays GPU-resident (grp=None).
@@ -328,7 +328,7 @@ def build_torch_vae(spec, instance, platform):
module = _load_component("VAELoader", ckpt, _fastvideo_args(spec))
device, dtype = _device(platform), _native_dtype(module)
if "LTX2" in type(module).__name__: # LTX2CausalVideoAutoencoder
from .torch_ltx2 import TorchLTX2VAE
from v2.platform.backends.torch_ltx2 import TorchLTX2VAE
return TorchLTX2VAE(module, device=device, dtype=dtype)
return TorchWanVAE(module, device=device, dtype=dtype)
@@ -342,7 +342,7 @@ def build_torch_text_encoder(spec, instance, platform):
tokenizer = _load_component("TokenizerLoader", os.path.join(_model_root(spec), "tokenizer"), args)
device, dtype = _device(platform), _native_dtype(module)
if "Gemma" in type(module).__name__: # LTX2GemmaTextEncoderModel
from .torch_ltx2 import TorchGemma
from v2.platform.backends.torch_ltx2 import TorchGemma
return TorchGemma(module, tokenizer, device=device, dtype=dtype)
return TorchT5Encoder(module, tokenizer, device=device, dtype=dtype)
@@ -361,7 +361,7 @@ def build_torch_upsampler(spec, instance, platform):
# ``per_channel_statistics`` lives on the AE's decoder/encoder sub-modules, NOT the top-level
# LTX2CausalVideoAutoencoder — upsample_video needs an object exposing it (same stats either side).
stats_owner = getattr(vae_module, "decoder", None) or getattr(vae_module, "encoder", None) or vae_module
from .torch_ltx2 import TorchLTX2Upsampler
from v2.platform.backends.torch_ltx2 import TorchLTX2Upsampler
return TorchLTX2Upsampler(module, stats_owner, device=device, dtype=dtype)
@@ -371,7 +371,7 @@ def build_torch_audio_vae(spec, instance, platform):
module = _load_component("AudioDecoderLoader", ckpt, _fastvideo_args(spec)) # LTX2AudioDecoder
device, dtype = _device(platform), _native_dtype(module)
voc = instance.component("vocoder") # build/fetch the vocoder; TorchLTX2AudioVAE chains decoder->vocoder
from .torch_ltx2 import TorchLTX2AudioVAE
from v2.platform.backends.torch_ltx2 import TorchLTX2AudioVAE
return TorchLTX2AudioVAE(module, voc, device=device, dtype=dtype)
@@ -380,5 +380,5 @@ def build_torch_vocoder(spec, instance, platform):
_ensure_fastvideo_runtime()
module = _load_component("VocoderLoader", ckpt, _fastvideo_args(spec)) # LTX2Vocoder
device, dtype = _device(platform), _native_dtype(module)
from .torch_ltx2 import TorchLTX2Vocoder
from v2.platform.backends.torch_ltx2 import TorchLTX2Vocoder
return TorchLTX2Vocoder(module, device=device, dtype=dtype)
v2/platform/backends/torch_cuda.py
+9 -9
View File
@@ -22,7 +22,7 @@ from __future__ import annotations
import importlib.util
from ..registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_component, register_kernel
from v2.platform.registry import FLOW_MATCH_STEP, FLOW_SDE_STEP, register_component, register_kernel
def _torch_present() -> bool:
@@ -41,42 +41,42 @@ def _cuda_available() -> bool:
# --- lazy trampolines: import torch only when actually called (on a GPU box) ------------------- #
def _build_dit(spec, instance, platform):
from .torch_adapters import build_torch_dit
from v2.platform.backends.torch_adapters import build_torch_dit
return build_torch_dit(spec, instance, platform)
def _build_vae(spec, instance, platform):
from .torch_adapters import build_torch_vae
from v2.platform.backends.torch_adapters import build_torch_vae
return build_torch_vae(spec, instance, platform)
def _build_text_encoder(spec, instance, platform):
from .torch_adapters import build_torch_text_encoder
from v2.platform.backends.torch_adapters import build_torch_text_encoder
return build_torch_text_encoder(spec, instance, platform)
def _build_upsampler(spec, instance, platform):
from .torch_adapters import build_torch_upsampler
from v2.platform.backends.torch_adapters import build_torch_upsampler
return build_torch_upsampler(spec, instance, platform)
def _build_audio_vae(spec, instance, platform):
from .torch_adapters import build_torch_audio_vae
from v2.platform.backends.torch_adapters import build_torch_audio_vae
return build_torch_audio_vae(spec, instance, platform)
def _build_vocoder(spec, instance, platform):
from .torch_adapters import build_torch_vocoder
from v2.platform.backends.torch_adapters import build_torch_vocoder
return build_torch_vocoder(spec, instance, platform)
def _flow_match_cuda(*args, **kwargs):
from .torch_kernels import flow_match_step
from v2.platform.backends.torch_kernels import flow_match_step
return flow_match_step(*args, **kwargs)
def _flow_sde_cuda(*args, **kwargs):
from .torch_kernels import flow_sde_step
from v2.platform.backends.torch_kernels import flow_sde_step
return flow_sde_step(*args, **kwargs)
v2/platform/backends/torch_ltx2.py
+1 -1
View File
@@ -15,7 +15,7 @@ from __future__ import annotations
import torch
from .torch_adapters import _to_numpy, _to_torch
from v2.platform.backends.torch_adapters import _to_numpy, _to_torch
class TorchLTX2DiT:
v2/platform/platform.py
+2 -2
View File
@@ -18,7 +18,7 @@ from __future__ import annotations
from typing import Any, Iterator
from .registry import COMPONENTS, KERNELS
from v2.platform.registry import COMPONENTS, KERNELS
_BACKENDS_LOADED = False
@@ -30,7 +30,7 @@ def ensure_backends_loaded() -> None:
return
_BACKENDS_LOADED = True
# imported for side effects (each module calls register_component/register_kernel at import)
from .backends import accel, cpu, torch_cuda # noqa: F401
from v2.platform.backends import accel, cpu, torch_cuda # noqa: F401
# Arch fallback chains (newest → oldest → portable terminal). Resolution walks these in order.
v2/program/__init__.py
+2 -2
View File
@@ -1,7 +1,7 @@
"""Program plane — compose a card's loops into a task (design_v3 §13)."""
from __future__ import annotations
from .specs import (
from v2.program.specs import (
ComponentNode,
Edge,
EdgeKind,
@@ -13,7 +13,7 @@ from .specs import (
when_opt,
when_task,
)
from .workflow import (
from v2.program.workflow import (
BestOfNWorkflow,
ParallelWorkflow,
Workflow,
v2/program/specs.py
+1 -1
View File
@@ -13,7 +13,7 @@ from dataclasses import dataclass, field
from enum import Enum
from typing import Any, Callable
from ..request.tasks import TaskType
from v2.request.tasks import TaskType
class ProgramKind(str, Enum):
v2/program/workflow.py
+1 -1
View File
@@ -111,7 +111,7 @@ class ParallelWorkflow(Workflow):
def run(self, engine: Any, **initial: Any) -> Any:
if not self.stages:
raise ValueError(f"workflow {self.workflow_id!r} has no stages")
from ..request.artifacts import Output
from v2.request.artifacts import Output
merged: dict[str, Any] = {}
metrics: dict[str, float] = {}
rid = self.workflow_id
v2/registry.py
+6 -6
View File
@@ -35,7 +35,7 @@ class ModelEntry:
def _entries() -> list[ModelEntry]:
"""The explicit registry (PRIMARY). Builders imported lazily to avoid import cycles with the model
packages and keep importing this module cheap."""
from .models.ltx2 import (
from v2.models.ltx2 import (
build_ltx2_3_card,
build_ltx2_3_program,
build_ltx2_base_card,
@@ -43,8 +43,8 @@ def _entries() -> list[ModelEntry]:
build_ltx2_card,
build_ltx2_program,
)
from .models.wan_causal import build_wan_causal_card, build_wan_causal_program
from .models.wan21 import (
from v2.models.wan_causal import build_wan_causal_card, build_wan_causal_program
from v2.models.wan21 import (
build_wan21_card,
build_wan22_a14b_card,
build_wan22_ti2v_card,
@@ -99,7 +99,7 @@ def select_by_architecture(sig: dict):
path / renamed repo / new distilled variant of a known arch still resolves with no registry entry."""
tr, pipe = sig.get("transformer_cls"), sig.get("pipeline")
if tr == "LTX2Transformer3DModel":
from .models.ltx2 import (
from v2.models.ltx2 import (
build_ltx2_base_card,
build_ltx2_base_program,
build_ltx2_card,
@@ -109,10 +109,10 @@ def select_by_architecture(sig: dict):
return build_ltx2_card, build_ltx2_program # two-stage (base -> upsample -> refine)
return build_ltx2_base_card, build_ltx2_base_program # single-stage
if tr == "CausalWanTransformer3DModel":
from .models.wan_causal import build_wan_causal_card, build_wan_causal_program
from v2.models.wan_causal import build_wan_causal_card, build_wan_causal_program
return build_wan_causal_card, build_wan_causal_program
if tr == "WanTransformer3DModel":
from .models.wan21 import (
from v2.models.wan21 import (
build_wan21_card,
build_wan22_a14b_card,
build_wan22_ti2v_card,
v2/request/__init__.py
+7 -7
View File
@@ -1,7 +1,7 @@
"""Request plane — typed runtime objects crossing the product boundary (design_v3 §12)."""
from __future__ import annotations
from .artifacts import (
from v2.request.artifacts import (
Artifact,
AudioArtifact,
LatentArtifact,
@@ -10,8 +10,8 @@ from .artifacts import (
TextArtifact,
VideoArtifact,
)
from .cancel import CancelKind, Cancelled, CancelScope
from .modalpart import (
from v2.request.cancel import CancelKind, Cancelled, CancelScope
from v2.request.modalpart import (
ActionPart,
AudioPart,
ImagePart,
@@ -21,10 +21,10 @@ from .modalpart import (
TextPart,
VideoPart,
)
from .params import CaptureMode, DiffusionParams, OutputSpec, SamplingParams
from .requests import Request, Session, make_request, new_request_id
from .streams import OmniEvent, Stream, StreamChunk
from .tasks import TaskType
from v2.request.params import CaptureMode, DiffusionParams, OutputSpec, SamplingParams
from v2.request.requests import Request, Session, make_request, new_request_id
from v2.request.streams import OmniEvent, Stream, StreamChunk
from v2.request.tasks import TaskType
__all__ = [
"TaskType", "Modality", "ModalPart", "TextPart", "ImagePart", "VideoPart",
v2/request/artifacts.py
+2 -2
View File
@@ -8,8 +8,8 @@ from __future__ import annotations
from dataclasses import dataclass, field
from .._types import TensorLike
from .modalpart import Modality
from v2._types import TensorLike
from v2.request.modalpart import Modality
@dataclass
v2/request/modalpart.py
+1 -1
View File
@@ -10,7 +10,7 @@ from __future__ import annotations
from dataclasses import dataclass
from enum import Enum
from .._types import TensorLike
from v2._types import TensorLike
class Modality(str, Enum):
v2/request/requests.py
+5 -5
View File
@@ -9,11 +9,11 @@ from __future__ import annotations
import itertools
from dataclasses import dataclass, field
from .cancel import CancelKind, CancelScope
from .modalpart import ImagePart, ModalPart, TextPart
from .params import DiffusionParams, OutputSpec, SamplingParams
from .streams import Stream
from .tasks import TaskType
from v2.request.cancel import CancelKind, CancelScope
from v2.request.modalpart import ImagePart, ModalPart, TextPart
from v2.request.params import DiffusionParams, OutputSpec, SamplingParams
from v2.request.streams import Stream
from v2.request.tasks import TaskType
_counter = itertools.count(1)
v2/request/streams.py
+1 -1
View File
@@ -14,7 +14,7 @@ from __future__ import annotations
from dataclasses import dataclass, field
from .._types import TensorLike
from v2._types import TensorLike
@dataclass(frozen=True)
v2/runtime/__init__.py
+7 -7
View File
@@ -1,13 +1,13 @@
"""Execution plane — engine + scheduler (design_v3 §6)."""
from __future__ import annotations
from .async_engine import AsyncEngine, RequestState
from .context import RuntimeLoopContext
from .disaggregated import DisaggregatedRunner
from .engine import Engine, ProgramRunner
from .pools import DeployConfig, PoolSet, RolePool, RolePoolSpec, wan_t2v_disaggregated
from .session import WorldModelSession
from .scheduler import (
from v2.runtime.async_engine import AsyncEngine, RequestState
from v2.runtime.context import RuntimeLoopContext
from v2.runtime.disaggregated import DisaggregatedRunner
from v2.runtime.engine import Engine, ProgramRunner
from v2.runtime.pools import DeployConfig, PoolSet, RolePool, RolePoolSpec, wan_t2v_disaggregated
from v2.runtime.session import WorldModelSession
from v2.runtime.scheduler import (
AdmissionController,
AdmissionInfeasible,
BatchScheduler,
v2/runtime/async_engine.py
+5 -5
View File
@@ -14,11 +14,11 @@ from __future__ import annotations
import asyncio
from typing import Any, AsyncIterator
from ..request.artifacts import Output
from ..request.cancel import Cancelled
from ..request.streams import OmniEvent
from .disaggregated import DisaggregatedRunner
from .engine import Engine
from v2.request.artifacts import Output
from v2.request.cancel import Cancelled
from v2.request.streams import OmniEvent
from v2.runtime.disaggregated import DisaggregatedRunner
from v2.runtime.engine import Engine
_SENTINEL = object()
v2/runtime/context.py
+4 -4
View File
@@ -10,9 +10,9 @@ from __future__ import annotations
from time import perf_counter
from typing import Any
from .._enums import ExecutionProfile
from ..loop.contracts import StepResult, WorkPlan
from ..request.streams import OmniEvent, Stream, StreamChunk
from v2._enums import ExecutionProfile
from v2.loop.contracts import StepResult, WorkPlan
from v2.request.streams import OmniEvent, Stream, StreamChunk
class RuntimeLoopContext:
@@ -77,7 +77,7 @@ class RuntimeLoopContext:
if getattr(loop, "graph_capture", "eager") != "breakable_cudagraph":
return eager(override)
if self.instance.graphs is None:
from .cudagraph import GraphCapturer
from v2.runtime.cudagraph import GraphCapturer
self.instance.graphs = GraphCapturer()
return self.instance.graphs.dispatch(plan, self.instance, override, eager)

Some files were not shown because too many files have changed in this diff Show More