synth-containers 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- synth_containers/__init__.py +235 -0
- synth_containers/adapters.py +221 -0
- synth_containers/capabilities.py +270 -0
- synth_containers/compat/__init__.py +51 -0
- synth_containers/compat/archipelago.py +332 -0
- synth_containers/compat/base.py +447 -0
- synth_containers/compat/harbor.py +101 -0
- synth_containers/compat/openenv.py +141 -0
- synth_containers/compat/service_http.py +179 -0
- synth_containers/compatibility.py +422 -0
- synth_containers/contracts.py +234 -0
- synth_containers/formats.py +241 -0
- synth_containers/http_adapter.py +298 -0
- synth_containers/http_client.py +191 -0
- synth_containers/http_models.py +85 -0
- synth_containers/nouns.py +408 -0
- synth_containers/ontology.py +152 -0
- synth_containers/profiles.py +172 -0
- synth_containers/protocols.py +127 -0
- synth_containers/proxying.py +177 -0
- synth_containers/recovery.py +183 -0
- synth_containers/reference_runtime.py +849 -0
- synth_containers/resources.py +75 -0
- synth_containers/rewards.py +186 -0
- synth_containers/runtime_requests.py +128 -0
- synth_containers/serde.py +54 -0
- synth_containers/tasks.py +154 -0
- synth_containers/tool_runtime.py +154 -0
- synth_containers/wire.py +125 -0
- synth_containers-0.1.0.dist-info/METADATA +291 -0
- synth_containers-0.1.0.dist-info/RECORD +33 -0
- synth_containers-0.1.0.dist-info/WHEEL +5 -0
- synth_containers-0.1.0.dist-info/top_level.txt +1 -0
|
@@ -0,0 +1,235 @@
|
|
|
1
|
+
"""Shared Synth container and environment runtime substrate."""
|
|
2
|
+
|
|
3
|
+
from .adapters import (
|
|
4
|
+
FrameworkAdapterDescriptor,
|
|
5
|
+
all_framework_descriptors,
|
|
6
|
+
archipelago_descriptor,
|
|
7
|
+
environments_old_descriptor,
|
|
8
|
+
harbor_descriptor,
|
|
9
|
+
openenv_descriptor,
|
|
10
|
+
)
|
|
11
|
+
from .capabilities import (
|
|
12
|
+
DatasetDescriptor,
|
|
13
|
+
RouteHints,
|
|
14
|
+
RuntimeCapabilitySurface,
|
|
15
|
+
RuntimeMetadata,
|
|
16
|
+
TaskCatalog,
|
|
17
|
+
TaskInfo,
|
|
18
|
+
TokenEmissionCapabilities,
|
|
19
|
+
)
|
|
20
|
+
from .compatibility import (
|
|
21
|
+
CONSUMER_REQUIREMENTS,
|
|
22
|
+
CompatibilityIssue,
|
|
23
|
+
CompatibilityReport,
|
|
24
|
+
ConsumerRequirement,
|
|
25
|
+
ConsumerTarget,
|
|
26
|
+
assert_consumer_support,
|
|
27
|
+
compatibility_matrix,
|
|
28
|
+
consumer_requirement,
|
|
29
|
+
evaluate_consumer_support,
|
|
30
|
+
evaluate_runtime_requirement,
|
|
31
|
+
)
|
|
32
|
+
from .contracts import (
|
|
33
|
+
ArtifactContract,
|
|
34
|
+
ArtifactPaths,
|
|
35
|
+
CheckpointResumeContract,
|
|
36
|
+
ContainerExecutionContract,
|
|
37
|
+
GoExContractDump,
|
|
38
|
+
LeverBundle,
|
|
39
|
+
RunPolicyContract,
|
|
40
|
+
TaskContract,
|
|
41
|
+
TransformPreferencePolicy,
|
|
42
|
+
)
|
|
43
|
+
from .formats import execution_to_rollout_payload, execution_to_state_payload, metadata_to_http_payload, task_info_to_http_payload
|
|
44
|
+
from .http_adapter import ManagedRuntime, create_reference_app
|
|
45
|
+
from .http_client import HTTPContainerClient
|
|
46
|
+
from .nouns import (
|
|
47
|
+
Action,
|
|
48
|
+
Actor,
|
|
49
|
+
ArtifactDescriptor,
|
|
50
|
+
CheckpointDescriptor,
|
|
51
|
+
ExecutionRecord,
|
|
52
|
+
Observation,
|
|
53
|
+
Outcome,
|
|
54
|
+
RewardSignal,
|
|
55
|
+
StateSnapshot,
|
|
56
|
+
TaskDefinition,
|
|
57
|
+
TaskInstance,
|
|
58
|
+
TokenTrace,
|
|
59
|
+
ToolCallRecord,
|
|
60
|
+
ToolSpecification,
|
|
61
|
+
TraceEvent,
|
|
62
|
+
Trajectory,
|
|
63
|
+
TurnRecord,
|
|
64
|
+
VerifierResult,
|
|
65
|
+
)
|
|
66
|
+
from .ontology import (
|
|
67
|
+
CONTRACT_VERSION,
|
|
68
|
+
CapabilityLevel,
|
|
69
|
+
CheckpointSemantics,
|
|
70
|
+
CoreNoun,
|
|
71
|
+
ExecutionKind,
|
|
72
|
+
ExecutionProfile,
|
|
73
|
+
OutcomeKind,
|
|
74
|
+
PrimitiveProtocol,
|
|
75
|
+
ResumeSemantics,
|
|
76
|
+
RewardSource,
|
|
77
|
+
RolloutMode,
|
|
78
|
+
RuntimeFamily,
|
|
79
|
+
RuntimeKind,
|
|
80
|
+
StatefulnessTier,
|
|
81
|
+
)
|
|
82
|
+
from .profiles import all_profile_specs, infer_profiles, missing_protocols, spec_for_profile
|
|
83
|
+
from .proxying import AgentRuntimeTarget, InferenceApiFamily, InferenceTarget, ProxyMode, ProxyResolution, ToolCallStyle, TraceIdentity
|
|
84
|
+
from .recovery import (
|
|
85
|
+
ArtifactTrustLevel,
|
|
86
|
+
RecoveryActionKind,
|
|
87
|
+
RecoveryPoint,
|
|
88
|
+
RunRecoveryProjection,
|
|
89
|
+
TransformFailure,
|
|
90
|
+
derive_run_recovery_projection,
|
|
91
|
+
)
|
|
92
|
+
from .reference_runtime import CounterRuntime, InMemoryAsyncRolloutExecutor, ReferenceManagedRuntime
|
|
93
|
+
from .resources import ResourceKind, ResourceRef
|
|
94
|
+
from .rewards import NLEScoutRewardTracker, ScoutRewardUpdate, TimePenaltyMode
|
|
95
|
+
from .runtime_requests import (
|
|
96
|
+
CheckpointLabelsRequest,
|
|
97
|
+
PauseRequest,
|
|
98
|
+
RuntimeCheckpointRequest,
|
|
99
|
+
RuntimeResumeRequest,
|
|
100
|
+
TerminateRequest,
|
|
101
|
+
)
|
|
102
|
+
from .tasks import DatasetSplit, InMemoryTaskCatalog, TaskFilter, task_definition_from_task_info
|
|
103
|
+
from .tool_runtime import ToolCallSchemaKind, ToolOutputMode, ToolRuntimeCapabilities, ToolRuntimeKind
|
|
104
|
+
from .wire import (
|
|
105
|
+
RequestValidationIssue,
|
|
106
|
+
RolloutState,
|
|
107
|
+
SubmissionMode,
|
|
108
|
+
assert_valid_transition,
|
|
109
|
+
lifecycle_projection,
|
|
110
|
+
project_lifecycle_fields,
|
|
111
|
+
resolve_submission_mode,
|
|
112
|
+
state_from_status,
|
|
113
|
+
validate_submission_mode_request,
|
|
114
|
+
)
|
|
115
|
+
|
|
116
|
+
__all__ = [
|
|
117
|
+
"Action",
|
|
118
|
+
"Actor",
|
|
119
|
+
"AgentRuntimeTarget",
|
|
120
|
+
"ArtifactContract",
|
|
121
|
+
"ArtifactDescriptor",
|
|
122
|
+
"ArtifactPaths",
|
|
123
|
+
"ArtifactTrustLevel",
|
|
124
|
+
"CONSUMER_REQUIREMENTS",
|
|
125
|
+
"CapabilityLevel",
|
|
126
|
+
"CheckpointDescriptor",
|
|
127
|
+
"CheckpointLabelsRequest",
|
|
128
|
+
"CheckpointResumeContract",
|
|
129
|
+
"CheckpointSemantics",
|
|
130
|
+
"CompatibilityIssue",
|
|
131
|
+
"CompatibilityReport",
|
|
132
|
+
"CONTRACT_VERSION",
|
|
133
|
+
"ContainerExecutionContract",
|
|
134
|
+
"ConsumerRequirement",
|
|
135
|
+
"ConsumerTarget",
|
|
136
|
+
"CoreNoun",
|
|
137
|
+
"CounterRuntime",
|
|
138
|
+
"DatasetDescriptor",
|
|
139
|
+
"DatasetSplit",
|
|
140
|
+
"ExecutionKind",
|
|
141
|
+
"ExecutionProfile",
|
|
142
|
+
"ExecutionRecord",
|
|
143
|
+
"FrameworkAdapterDescriptor",
|
|
144
|
+
"GoExContractDump",
|
|
145
|
+
"HTTPContainerClient",
|
|
146
|
+
"InMemoryAsyncRolloutExecutor",
|
|
147
|
+
"InMemoryTaskCatalog",
|
|
148
|
+
"InferenceApiFamily",
|
|
149
|
+
"InferenceTarget",
|
|
150
|
+
"LeverBundle",
|
|
151
|
+
"ManagedRuntime",
|
|
152
|
+
"NLEScoutRewardTracker",
|
|
153
|
+
"Observation",
|
|
154
|
+
"Outcome",
|
|
155
|
+
"OutcomeKind",
|
|
156
|
+
"PauseRequest",
|
|
157
|
+
"PrimitiveProtocol",
|
|
158
|
+
"ProxyMode",
|
|
159
|
+
"ProxyResolution",
|
|
160
|
+
"RecoveryActionKind",
|
|
161
|
+
"RecoveryPoint",
|
|
162
|
+
"ReferenceManagedRuntime",
|
|
163
|
+
"RequestValidationIssue",
|
|
164
|
+
"ResumeSemantics",
|
|
165
|
+
"RewardSignal",
|
|
166
|
+
"RewardSource",
|
|
167
|
+
"ResourceKind",
|
|
168
|
+
"ResourceRef",
|
|
169
|
+
"RolloutMode",
|
|
170
|
+
"RolloutState",
|
|
171
|
+
"RouteHints",
|
|
172
|
+
"RunPolicyContract",
|
|
173
|
+
"RunRecoveryProjection",
|
|
174
|
+
"RuntimeCapabilitySurface",
|
|
175
|
+
"RuntimeCheckpointRequest",
|
|
176
|
+
"RuntimeFamily",
|
|
177
|
+
"RuntimeKind",
|
|
178
|
+
"RuntimeMetadata",
|
|
179
|
+
"RuntimeResumeRequest",
|
|
180
|
+
"ScoutRewardUpdate",
|
|
181
|
+
"StateSnapshot",
|
|
182
|
+
"StatefulnessTier",
|
|
183
|
+
"SubmissionMode",
|
|
184
|
+
"TaskCatalog",
|
|
185
|
+
"TaskContract",
|
|
186
|
+
"TaskDefinition",
|
|
187
|
+
"TaskFilter",
|
|
188
|
+
"TaskInfo",
|
|
189
|
+
"TaskInstance",
|
|
190
|
+
"TerminateRequest",
|
|
191
|
+
"TimePenaltyMode",
|
|
192
|
+
"TokenEmissionCapabilities",
|
|
193
|
+
"TokenTrace",
|
|
194
|
+
"ToolCallRecord",
|
|
195
|
+
"ToolCallSchemaKind",
|
|
196
|
+
"ToolCallStyle",
|
|
197
|
+
"ToolOutputMode",
|
|
198
|
+
"ToolRuntimeCapabilities",
|
|
199
|
+
"ToolRuntimeKind",
|
|
200
|
+
"ToolSpecification",
|
|
201
|
+
"TraceEvent",
|
|
202
|
+
"TraceIdentity",
|
|
203
|
+
"Trajectory",
|
|
204
|
+
"TransformFailure",
|
|
205
|
+
"TransformPreferencePolicy",
|
|
206
|
+
"TurnRecord",
|
|
207
|
+
"VerifierResult",
|
|
208
|
+
"all_framework_descriptors",
|
|
209
|
+
"all_profile_specs",
|
|
210
|
+
"archipelago_descriptor",
|
|
211
|
+
"assert_consumer_support",
|
|
212
|
+
"assert_valid_transition",
|
|
213
|
+
"compatibility_matrix",
|
|
214
|
+
"consumer_requirement",
|
|
215
|
+
"create_reference_app",
|
|
216
|
+
"derive_run_recovery_projection",
|
|
217
|
+
"environments_old_descriptor",
|
|
218
|
+
"evaluate_consumer_support",
|
|
219
|
+
"evaluate_runtime_requirement",
|
|
220
|
+
"execution_to_rollout_payload",
|
|
221
|
+
"execution_to_state_payload",
|
|
222
|
+
"harbor_descriptor",
|
|
223
|
+
"infer_profiles",
|
|
224
|
+
"lifecycle_projection",
|
|
225
|
+
"metadata_to_http_payload",
|
|
226
|
+
"missing_protocols",
|
|
227
|
+
"openenv_descriptor",
|
|
228
|
+
"project_lifecycle_fields",
|
|
229
|
+
"resolve_submission_mode",
|
|
230
|
+
"spec_for_profile",
|
|
231
|
+
"state_from_status",
|
|
232
|
+
"task_definition_from_task_info",
|
|
233
|
+
"task_info_to_http_payload",
|
|
234
|
+
"validate_submission_mode_request",
|
|
235
|
+
]
|
|
@@ -0,0 +1,221 @@
|
|
|
1
|
+
from __future__ import annotations
|
|
2
|
+
|
|
3
|
+
from dataclasses import dataclass, field
|
|
4
|
+
|
|
5
|
+
from .ontology import CapabilityLevel, CoreNoun, ExecutionProfile, PrimitiveProtocol, RuntimeKind
|
|
6
|
+
from .serde import JsonDataclassMixin
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
@dataclass(frozen=True, slots=True)
|
|
10
|
+
class FrameworkAdapterDescriptor(JsonDataclassMixin):
|
|
11
|
+
framework: str
|
|
12
|
+
runtime_kind: RuntimeKind
|
|
13
|
+
noun_fidelity: dict[CoreNoun, CapabilityLevel]
|
|
14
|
+
protocol_fidelity: dict[PrimitiveProtocol, CapabilityLevel]
|
|
15
|
+
profile_fidelity: dict[ExecutionProfile, CapabilityLevel]
|
|
16
|
+
notes: list[str] = field(default_factory=list)
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
def environments_old_descriptor() -> FrameworkAdapterDescriptor:
|
|
21
|
+
return FrameworkAdapterDescriptor(
|
|
22
|
+
framework="environments_old",
|
|
23
|
+
runtime_kind=RuntimeKind.ENVIRONMENT,
|
|
24
|
+
noun_fidelity={
|
|
25
|
+
CoreNoun.RUNTIME: CapabilityLevel.NATIVE,
|
|
26
|
+
CoreNoun.ACTOR: CapabilityLevel.NATIVE,
|
|
27
|
+
CoreNoun.ACTION: CapabilityLevel.NATIVE,
|
|
28
|
+
CoreNoun.OBSERVATION: CapabilityLevel.NATIVE,
|
|
29
|
+
CoreNoun.STATE: CapabilityLevel.NATIVE,
|
|
30
|
+
CoreNoun.EXECUTION: CapabilityLevel.NATIVE,
|
|
31
|
+
CoreNoun.OUTCOME: CapabilityLevel.DERIVED,
|
|
32
|
+
CoreNoun.TASK_INSTANCE: CapabilityLevel.NATIVE,
|
|
33
|
+
CoreNoun.CHECKPOINT: CapabilityLevel.NATIVE,
|
|
34
|
+
CoreNoun.TASK_CATALOG: CapabilityLevel.NATIVE,
|
|
35
|
+
},
|
|
36
|
+
protocol_fidelity={
|
|
37
|
+
PrimitiveProtocol.CATALOG_BACKED: CapabilityLevel.NATIVE,
|
|
38
|
+
PrimitiveProtocol.RESETTABLE: CapabilityLevel.NATIVE,
|
|
39
|
+
PrimitiveProtocol.STEPPABLE: CapabilityLevel.NATIVE,
|
|
40
|
+
PrimitiveProtocol.OBSERVABLE: CapabilityLevel.NATIVE,
|
|
41
|
+
PrimitiveProtocol.STATE_READABLE: CapabilityLevel.NATIVE,
|
|
42
|
+
PrimitiveProtocol.CHECKPOINTABLE: CapabilityLevel.NATIVE,
|
|
43
|
+
PrimitiveProtocol.RESTORABLE: CapabilityLevel.NATIVE,
|
|
44
|
+
PrimitiveProtocol.FORKABLE: CapabilityLevel.DERIVED,
|
|
45
|
+
PrimitiveProtocol.ROLLOUT_RUNNABLE: CapabilityLevel.APPROXIMATE,
|
|
46
|
+
PrimitiveProtocol.ASYNC_ROLLOUT_RUNNABLE: CapabilityLevel.DERIVED,
|
|
47
|
+
PrimitiveProtocol.TRACE_EMITTING: CapabilityLevel.DERIVED,
|
|
48
|
+
PrimitiveProtocol.REWARD_EMITTING: CapabilityLevel.APPROXIMATE,
|
|
49
|
+
PrimitiveProtocol.VERIFIER_BACKED: CapabilityLevel.APPROXIMATE,
|
|
50
|
+
PrimitiveProtocol.TOOL_CALLABLE: CapabilityLevel.APPROXIMATE,
|
|
51
|
+
PrimitiveProtocol.TOKEN_TRACE_EMITTING: CapabilityLevel.UNSUPPORTED,
|
|
52
|
+
PrimitiveProtocol.MULTI_ACTOR: CapabilityLevel.UNSUPPORTED,
|
|
53
|
+
PrimitiveProtocol.PROXIED_INFERENCE_BACKED: CapabilityLevel.UNSUPPORTED,
|
|
54
|
+
},
|
|
55
|
+
profile_fidelity={
|
|
56
|
+
ExecutionProfile.GYM_STYLE_ENVIRONMENT: CapabilityLevel.NATIVE,
|
|
57
|
+
ExecutionProfile.CHECKPOINTABLE_STATEFUL_ENVIRONMENT: CapabilityLevel.NATIVE,
|
|
58
|
+
ExecutionProfile.CHECKPOINTABLE_LONG_HORIZON_ENVIRONMENT: CapabilityLevel.DERIVED,
|
|
59
|
+
ExecutionProfile.RL_TRAJECTORY_EMITTER: CapabilityLevel.APPROXIMATE,
|
|
60
|
+
},
|
|
61
|
+
notes=[
|
|
62
|
+
"Preserves engine-vs-environment split and checkpointing semantics.",
|
|
63
|
+
"Async rollout control usually arrives through a wrapper service rather than the core framework.",
|
|
64
|
+
],
|
|
65
|
+
)
|
|
66
|
+
|
|
67
|
+
|
|
68
|
+
|
|
69
|
+
def openenv_descriptor() -> FrameworkAdapterDescriptor:
|
|
70
|
+
return FrameworkAdapterDescriptor(
|
|
71
|
+
framework="openenv",
|
|
72
|
+
runtime_kind=RuntimeKind.ENVIRONMENT,
|
|
73
|
+
noun_fidelity={
|
|
74
|
+
CoreNoun.RUNTIME: CapabilityLevel.NATIVE,
|
|
75
|
+
CoreNoun.ACTOR: CapabilityLevel.NATIVE,
|
|
76
|
+
CoreNoun.ACTION: CapabilityLevel.NATIVE,
|
|
77
|
+
CoreNoun.OBSERVATION: CapabilityLevel.NATIVE,
|
|
78
|
+
CoreNoun.STATE: CapabilityLevel.NATIVE,
|
|
79
|
+
CoreNoun.EXECUTION: CapabilityLevel.NATIVE,
|
|
80
|
+
CoreNoun.OUTCOME: CapabilityLevel.NATIVE,
|
|
81
|
+
CoreNoun.TASK_INSTANCE: CapabilityLevel.APPROXIMATE,
|
|
82
|
+
CoreNoun.TASK_CATALOG: CapabilityLevel.UNSUPPORTED,
|
|
83
|
+
},
|
|
84
|
+
protocol_fidelity={
|
|
85
|
+
PrimitiveProtocol.RESETTABLE: CapabilityLevel.NATIVE,
|
|
86
|
+
PrimitiveProtocol.STEPPABLE: CapabilityLevel.NATIVE,
|
|
87
|
+
PrimitiveProtocol.OBSERVABLE: CapabilityLevel.NATIVE,
|
|
88
|
+
PrimitiveProtocol.STATE_READABLE: CapabilityLevel.NATIVE,
|
|
89
|
+
PrimitiveProtocol.CHECKPOINTABLE: CapabilityLevel.APPROXIMATE,
|
|
90
|
+
PrimitiveProtocol.RESTORABLE: CapabilityLevel.APPROXIMATE,
|
|
91
|
+
PrimitiveProtocol.ROLLOUT_RUNNABLE: CapabilityLevel.APPROXIMATE,
|
|
92
|
+
PrimitiveProtocol.ASYNC_ROLLOUT_RUNNABLE: CapabilityLevel.APPROXIMATE,
|
|
93
|
+
PrimitiveProtocol.TRACE_EMITTING: CapabilityLevel.DERIVED,
|
|
94
|
+
PrimitiveProtocol.REWARD_EMITTING: CapabilityLevel.NATIVE,
|
|
95
|
+
PrimitiveProtocol.VERIFIER_BACKED: CapabilityLevel.UNSUPPORTED,
|
|
96
|
+
PrimitiveProtocol.TOOL_CALLABLE: CapabilityLevel.APPROXIMATE,
|
|
97
|
+
PrimitiveProtocol.TOKEN_TRACE_EMITTING: CapabilityLevel.APPROXIMATE,
|
|
98
|
+
PrimitiveProtocol.MULTI_ACTOR: CapabilityLevel.APPROXIMATE,
|
|
99
|
+
PrimitiveProtocol.PROXIED_INFERENCE_BACKED: CapabilityLevel.APPROXIMATE,
|
|
100
|
+
PrimitiveProtocol.CATALOG_BACKED: CapabilityLevel.UNSUPPORTED,
|
|
101
|
+
},
|
|
102
|
+
profile_fidelity={
|
|
103
|
+
ExecutionProfile.GYM_STYLE_ENVIRONMENT: CapabilityLevel.NATIVE,
|
|
104
|
+
ExecutionProfile.RL_TRAJECTORY_EMITTER: CapabilityLevel.NATIVE,
|
|
105
|
+
ExecutionProfile.CHECKPOINTABLE_STATEFUL_ENVIRONMENT: CapabilityLevel.APPROXIMATE,
|
|
106
|
+
ExecutionProfile.HARNESS_MANAGED_BENCHMARK_ENVIRONMENT: CapabilityLevel.UNSUPPORTED,
|
|
107
|
+
},
|
|
108
|
+
notes=[
|
|
109
|
+
"Best external reference for clean reset/step/state client-server interaction.",
|
|
110
|
+
"Task catalog and benchmark orchestration are weaker than in the Synth substrate.",
|
|
111
|
+
],
|
|
112
|
+
)
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
|
|
116
|
+
def archipelago_descriptor() -> FrameworkAdapterDescriptor:
|
|
117
|
+
return FrameworkAdapterDescriptor(
|
|
118
|
+
framework="archipelago",
|
|
119
|
+
runtime_kind=RuntimeKind.SANDBOX,
|
|
120
|
+
noun_fidelity={
|
|
121
|
+
CoreNoun.RUNTIME: CapabilityLevel.NATIVE,
|
|
122
|
+
CoreNoun.ACTOR: CapabilityLevel.NATIVE,
|
|
123
|
+
CoreNoun.ACTION: CapabilityLevel.NATIVE,
|
|
124
|
+
CoreNoun.OBSERVATION: CapabilityLevel.APPROXIMATE,
|
|
125
|
+
CoreNoun.STATE: CapabilityLevel.APPROXIMATE,
|
|
126
|
+
CoreNoun.EXECUTION: CapabilityLevel.NATIVE,
|
|
127
|
+
CoreNoun.OUTCOME: CapabilityLevel.NATIVE,
|
|
128
|
+
CoreNoun.TASK_INSTANCE: CapabilityLevel.NATIVE,
|
|
129
|
+
CoreNoun.ARTIFACT: CapabilityLevel.NATIVE,
|
|
130
|
+
CoreNoun.TOOL: CapabilityLevel.NATIVE,
|
|
131
|
+
CoreNoun.CHECKPOINT: CapabilityLevel.APPROXIMATE,
|
|
132
|
+
},
|
|
133
|
+
protocol_fidelity={
|
|
134
|
+
PrimitiveProtocol.CATALOG_BACKED: CapabilityLevel.APPROXIMATE,
|
|
135
|
+
PrimitiveProtocol.RESETTABLE: CapabilityLevel.APPROXIMATE,
|
|
136
|
+
PrimitiveProtocol.STEPPABLE: CapabilityLevel.APPROXIMATE,
|
|
137
|
+
PrimitiveProtocol.OBSERVABLE: CapabilityLevel.APPROXIMATE,
|
|
138
|
+
PrimitiveProtocol.STATE_READABLE: CapabilityLevel.APPROXIMATE,
|
|
139
|
+
PrimitiveProtocol.CHECKPOINTABLE: CapabilityLevel.APPROXIMATE,
|
|
140
|
+
PrimitiveProtocol.RESTORABLE: CapabilityLevel.APPROXIMATE,
|
|
141
|
+
PrimitiveProtocol.FORKABLE: CapabilityLevel.APPROXIMATE,
|
|
142
|
+
PrimitiveProtocol.ROLLOUT_RUNNABLE: CapabilityLevel.DERIVED,
|
|
143
|
+
PrimitiveProtocol.ASYNC_ROLLOUT_RUNNABLE: CapabilityLevel.NATIVE,
|
|
144
|
+
PrimitiveProtocol.TRACE_EMITTING: CapabilityLevel.NATIVE,
|
|
145
|
+
PrimitiveProtocol.REWARD_EMITTING: CapabilityLevel.APPROXIMATE,
|
|
146
|
+
PrimitiveProtocol.VERIFIER_BACKED: CapabilityLevel.NATIVE,
|
|
147
|
+
PrimitiveProtocol.TOOL_CALLABLE: CapabilityLevel.NATIVE,
|
|
148
|
+
PrimitiveProtocol.TOKEN_TRACE_EMITTING: CapabilityLevel.UNSUPPORTED,
|
|
149
|
+
PrimitiveProtocol.MULTI_ACTOR: CapabilityLevel.APPROXIMATE,
|
|
150
|
+
PrimitiveProtocol.PROXIED_INFERENCE_BACKED: CapabilityLevel.APPROXIMATE,
|
|
151
|
+
},
|
|
152
|
+
profile_fidelity={
|
|
153
|
+
ExecutionProfile.SANDBOXED_MCP_WORLD: CapabilityLevel.NATIVE,
|
|
154
|
+
ExecutionProfile.HARNESS_MANAGED_BENCHMARK_ENVIRONMENT: CapabilityLevel.NATIVE,
|
|
155
|
+
ExecutionProfile.CHECKPOINTABLE_STATEFUL_ENVIRONMENT: CapabilityLevel.APPROXIMATE,
|
|
156
|
+
},
|
|
157
|
+
notes=[
|
|
158
|
+
"Sandbox and MCP gateway patterns are first-class.",
|
|
159
|
+
"Snapshots are often grading-oriented rather than branchable true environment checkpoints.",
|
|
160
|
+
],
|
|
161
|
+
)
|
|
162
|
+
|
|
163
|
+
|
|
164
|
+
|
|
165
|
+
def harbor_descriptor() -> FrameworkAdapterDescriptor:
|
|
166
|
+
return FrameworkAdapterDescriptor(
|
|
167
|
+
framework="harbor",
|
|
168
|
+
runtime_kind=RuntimeKind.HARNESS,
|
|
169
|
+
noun_fidelity={
|
|
170
|
+
CoreNoun.RUNTIME: CapabilityLevel.NATIVE,
|
|
171
|
+
CoreNoun.ACTOR: CapabilityLevel.NATIVE,
|
|
172
|
+
CoreNoun.ACTION: CapabilityLevel.NATIVE,
|
|
173
|
+
CoreNoun.OBSERVATION: CapabilityLevel.APPROXIMATE,
|
|
174
|
+
CoreNoun.STATE: CapabilityLevel.UNSUPPORTED,
|
|
175
|
+
CoreNoun.EXECUTION: CapabilityLevel.NATIVE,
|
|
176
|
+
CoreNoun.OUTCOME: CapabilityLevel.NATIVE,
|
|
177
|
+
CoreNoun.TASK_INSTANCE: CapabilityLevel.NATIVE,
|
|
178
|
+
CoreNoun.ARTIFACT: CapabilityLevel.NATIVE,
|
|
179
|
+
CoreNoun.TRACE: CapabilityLevel.NATIVE,
|
|
180
|
+
CoreNoun.VERIFIER_RESULT: CapabilityLevel.NATIVE,
|
|
181
|
+
CoreNoun.TASK_CATALOG: CapabilityLevel.NATIVE,
|
|
182
|
+
},
|
|
183
|
+
protocol_fidelity={
|
|
184
|
+
PrimitiveProtocol.CATALOG_BACKED: CapabilityLevel.NATIVE,
|
|
185
|
+
PrimitiveProtocol.RESETTABLE: CapabilityLevel.APPROXIMATE,
|
|
186
|
+
PrimitiveProtocol.STEPPABLE: CapabilityLevel.APPROXIMATE,
|
|
187
|
+
PrimitiveProtocol.OBSERVABLE: CapabilityLevel.APPROXIMATE,
|
|
188
|
+
PrimitiveProtocol.STATE_READABLE: CapabilityLevel.UNSUPPORTED,
|
|
189
|
+
PrimitiveProtocol.CHECKPOINTABLE: CapabilityLevel.APPROXIMATE,
|
|
190
|
+
PrimitiveProtocol.RESTORABLE: CapabilityLevel.APPROXIMATE,
|
|
191
|
+
PrimitiveProtocol.FORKABLE: CapabilityLevel.APPROXIMATE,
|
|
192
|
+
PrimitiveProtocol.ROLLOUT_RUNNABLE: CapabilityLevel.NATIVE,
|
|
193
|
+
PrimitiveProtocol.ASYNC_ROLLOUT_RUNNABLE: CapabilityLevel.NATIVE,
|
|
194
|
+
PrimitiveProtocol.TRACE_EMITTING: CapabilityLevel.NATIVE,
|
|
195
|
+
PrimitiveProtocol.REWARD_EMITTING: CapabilityLevel.NATIVE,
|
|
196
|
+
PrimitiveProtocol.VERIFIER_BACKED: CapabilityLevel.NATIVE,
|
|
197
|
+
PrimitiveProtocol.TOOL_CALLABLE: CapabilityLevel.APPROXIMATE,
|
|
198
|
+
PrimitiveProtocol.TOKEN_TRACE_EMITTING: CapabilityLevel.APPROXIMATE,
|
|
199
|
+
PrimitiveProtocol.MULTI_ACTOR: CapabilityLevel.DERIVED,
|
|
200
|
+
PrimitiveProtocol.PROXIED_INFERENCE_BACKED: CapabilityLevel.NATIVE,
|
|
201
|
+
},
|
|
202
|
+
profile_fidelity={
|
|
203
|
+
ExecutionProfile.HARNESS_MANAGED_BENCHMARK_ENVIRONMENT: CapabilityLevel.NATIVE,
|
|
204
|
+
ExecutionProfile.RL_TRAJECTORY_EMITTER: CapabilityLevel.APPROXIMATE,
|
|
205
|
+
ExecutionProfile.CHECKPOINTABLE_LONG_HORIZON_ENVIRONMENT: CapabilityLevel.APPROXIMATE,
|
|
206
|
+
},
|
|
207
|
+
notes=[
|
|
208
|
+
"Strong benchmark/eval/RL orchestration reference.",
|
|
209
|
+
"Harness-centric design means engine state and native checkpoint semantics are weaker than substrate-first systems.",
|
|
210
|
+
],
|
|
211
|
+
)
|
|
212
|
+
|
|
213
|
+
|
|
214
|
+
|
|
215
|
+
def all_framework_descriptors() -> list[FrameworkAdapterDescriptor]:
|
|
216
|
+
return [
|
|
217
|
+
environments_old_descriptor(),
|
|
218
|
+
openenv_descriptor(),
|
|
219
|
+
archipelago_descriptor(),
|
|
220
|
+
harbor_descriptor(),
|
|
221
|
+
]
|