Slot: n_parallel

Grid batch size for parameter exploration — how many sweep cells the backend evaluates as one vectorised chunk. auto (the default) vectorises the grid, bounded by a cell-count cap (min(grid_size, 64), env TVBO_NVMAP_AUTO_CAP) and a working-memory budget (env TVBO_NVMAP_MEM_BUDGET_GB, default 2 GB) estimated from the per-cell output and live state: fast on every backend, and the memory budget keeps a large-per-cell grid (e.g. a whole-brain delay network) from inflating peak memory. Vectorising still holds n_vmap cells at once, so auto uses more working memory than a sequential run — the budget bounds it. An integer fixes the chunk size and bypasses both bounds (1 = fully sequential, minimal memory; a larger value trades memory for fewer, wider batches). Backends translate this into their own execution primitive (e.g. tvboptim’s vmap chunk width) without exposing any backend-specific names in the schema.

URI: tvbo:slot/n_parallel

Applicable Classes

Name Description Modifies Slot
Exploration Parameter space exploration (grid search, sweep) no

Properties

Type and Range

Property Value
Range xsd:string or 
xsd:integer
Domain Of Exploration

Cardinality and Requirements

Property Value

Slot Characteristics

Property Value
If Absent string(auto)
Owner Exploration
Expressions & Logic

Any Of

Value must satisfy at least one of: - AnonymousSlotExpression({‘range’: ‘integer’}) - AnonymousSlotExpression({‘range’: ‘string’})

Identifier and Mapping Information

Schema Source

  • from schema: https://w3id.org/tvbo

Mappings

Mapping Type Mapped Value
self tvbo:n_parallel
native tvbo:n_parallel

LinkML Source

name: n_parallel
description: 'Grid batch size for parameter exploration — how many sweep cells the
  backend evaluates as one vectorised chunk. ``auto`` (the default) vectorises the
  grid, bounded by a cell-count cap (``min(grid_size, 64)``, env ``TVBO_NVMAP_AUTO_CAP``)
  and a working-memory budget (env ``TVBO_NVMAP_MEM_BUDGET_GB``, default 2 GB) estimated
  from the per-cell output and live state: fast on every backend, and the memory budget
  keeps a large-per-cell grid (e.g. a whole-brain delay network) from inflating peak
  memory. Vectorising still holds ``n_vmap`` cells at once, so auto uses more working
  memory than a sequential run — the budget bounds it. An integer fixes the chunk
  size and bypasses both bounds (``1`` = fully sequential, minimal memory; a larger
  value trades memory for fewer, wider batches). Backends translate this into their
  own execution primitive (e.g. tvboptim''s vmap chunk width) without exposing any
  backend-specific names in the schema.'
from_schema: https://w3id.org/tvbo
rank: 1000
ifabsent: string(auto)
owner: Exploration
domain_of:
- Exploration
range: string
any_of:
- range: integer
- range: string