Dataset Viewer
Auto-converted to Parquet Duplicate
actions
list
actual_gib
float64
class_counts_this_run
dict
decision_classes
list
dtype
string
edge_types
dict
elapsed_seconds_this_run
float64
generator_version
string
geometry
dict
minimum_rows_requested
int64
name
string
node_types
dict
notes
list
routing_protocols
dict
rows_per_second_this_run
float64
rows_written
int64
schema_version
string
seed
int64
split_ids
dict
status
string
storage_format
string
target_gib
float64
tensor_bytes_per_row_estimate
int64
topology_families
dict
[ "NO_ACTION", "RESTORE_DNS_RECORD", "CORRECT_DNS_RECORD", "BRING_SOURCE_INTERFACE_UP", "RESTORE_LINK", "RESTORE_OSPF_ADJACENCY", "RESTORE_BGP_SESSION", "FIX_BGP_IMPORT_POLICY", "INSTALL_OR_RESTORE_ROUTE", "ALLOW_REQUIRED_FLOW", "RESTORE_NAT_MAPPING", "FIX_PATH_MTU", "RESTORE_SERVICE_ENDPOINTS...
15.212922
{ "ACL_DENY": 61680, "BGP_POLICY_REJECT": 61681, "BGP_SESSION_DOWN": 61681, "DNS_NXDOMAIN": 61681, "DNS_WRONG_ANSWER": 61681, "ENDPOINT_DOWN": 61681, "HEALTHY": 61681, "LINK_DOWN": 61681, "MTU_BLACKHOLE": 61681, "NAT_MISSING": 61681, "NO_ROUTE": 61681, "OSPF_ADJACENCY_DOWN": 61681, "PORT_CLOSE...
[ "HEALTHY", "DNS_NXDOMAIN", "DNS_WRONG_ANSWER", "SOURCE_INTERFACE_DOWN", "LINK_DOWN", "OSPF_ADJACENCY_DOWN", "BGP_SESSION_DOWN", "BGP_POLICY_REJECT", "NO_ROUTE", "ACL_DENY", "NAT_MISSING", "MTU_BLACKHOLE", "SERVICE_SELECTOR_EMPTY", "ENDPOINT_DOWN", "PORT_CLOSED", "TLS_CERT_EXPIRED", "...
float16
{ "DNS_RESOLUTION": 3, "PAD": 0, "PHYSICAL_LINK": 1, "SERVICE_ENDPOINT": 2, "TLS_CERTIFICATE": 4 }
45.958085
netdecision-cuda-v2.0
{ "edge_features": 10, "event_count": 32, "event_features": 8, "flow_count": 48, "flow_features": 12, "max_edges": 96, "max_nodes": 32, "max_path_nodes": 12, "node_features": 12, "proof_bits": 20, "query_features": 10, "route_features": 8, "route_rows": 64, "telemetry_devices": 8, "telemet...
1,000,000
NetDecision CUDA v2
{ "CERTIFICATE": 6, "CLIENT": 1, "DNS": 5, "PAD": 0, "ROUTER": 2, "SERVER": 3, "SERVICE": 4 }
[ "All numeric corpus tensors are generated in batches on the requested device.", "The injected target is never written into input tensors.", "A separate deterministic tensor oracle re-derives decision_id.", "The pre-fault counterfactual is oracle-validated as HEALTHY.", "Off-path decoys are excluded from the...
{ "BGP": 1, "OSPF": 0 }
22,815.920291
1,048,576
netdecision-tensor-schema-v2.0
20,261,007
{ "test": 2, "train": 0, "validation": 1 }
complete
safetensors
15
15,576
{ "CHAIN": 0, "DUAL_CORE": 2, "RING": 1, "SPARSE_MESH": 4, "SPINE_LEAF": 3 }

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

NetDecision CUDA v2

GPU-native typed network-decision dataset.

Task

typed graph + state + telemetry + query -> decision_id + location + action_id

The generator has no LLM labeler and no probabilistic oracle.

Files

Each part-*.safetensors shard contains fixed-shape tensors:

  • node_type
  • node_features
  • edge_index
  • edge_type
  • edge_features
  • query
  • telemetry
  • flows
  • routes
  • events
  • proof_bits
  • decision_id
  • action_id
  • fault_entity_kind
  • fault_entity_index
  • counterfactual_decision_id
  • topology/split metadata

Deterministic quality controls

  1. Build a healthy network state.
  2. Run the tensor oracle and require HEALTHY.
  3. Mutate the network with one target fault.
  4. Add unrelated off-path decoys.
  5. Run the oracle again.
  6. Require oracle diagnosis == injected target.
  7. Write only validated batches.

Shape summary

  • Nodes: 32
  • Edges: 96
  • Telemetry: 8 devices x 60 steps x 10 metrics
  • Flows: 48 x 12
  • Routes: 64 x 8
  • Events: 32 x 8

See manifest.json for vocabularies and exact metadata.

Downloads last month
183