# Class: ExecutionConfig _Configuration for computational execution (parallelization, precision, hardware)._
URI: [tvbo:ExecutionConfig](https://w3id.org/tvbo/ExecutionConfig) ```{mermaid} classDiagram class ExecutionConfig click ExecutionConfig href "./ExecutionConfig.html" ExecutionConfig : accelerator ExecutionConfig --> "0..1" String : accelerator click String href "../http://www.w3.org/2001/XMLSchema#string.html" ExecutionConfig : backend ExecutionConfig --> "0..1" String : backend click String href "../http://www.w3.org/2001/XMLSchema#string.html" ExecutionConfig : batch_size ExecutionConfig --> "0..1" Integer : batch_size click Integer href "../http://www.w3.org/2001/XMLSchema#integer.html" ExecutionConfig : find_fixpoint ExecutionConfig --> "0..1" Boolean : find_fixpoint click Boolean href "../http://www.w3.org/2001/XMLSchema#boolean.html" ExecutionConfig : n_threads ExecutionConfig --> "0..1" Integer : n_threads click Integer href "../http://www.w3.org/2001/XMLSchema#integer.html" ExecutionConfig : n_workers ExecutionConfig --> "0..1" Integer : n_workers click Integer href "../http://www.w3.org/2001/XMLSchema#integer.html" ExecutionConfig : precision ExecutionConfig --> "0..1" String : precision click String href "../http://www.w3.org/2001/XMLSchema#string.html" ExecutionConfig : random_seed ExecutionConfig --> "0..1" Integer : random_seed click Integer href "../http://www.w3.org/2001/XMLSchema#integer.html" ExecutionConfig : system_size ExecutionConfig --> "0..1" Float : system_size click Float href "../http://www.w3.org/2001/XMLSchema#float.html" ``` ## Class Properties | Property | Value | | --- | --- | | Class URI | [tvbo:ExecutionConfig](https://w3id.org/tvbo/ExecutionConfig) | ## Slots | Name | Cardinality and Range | Description | Inheritance | | --- | --- | --- | --- | | [n_workers](../slots/n_workers.qmd) | 0..1
[xsd:integer](http://www.w3.org/2001/XMLSchema#integer) | Number of parallel workers (maps to pmap devices in JAX, processes in multipr... | direct | | [n_threads](../slots/n_threads.qmd) | 0..1
[xsd:integer](http://www.w3.org/2001/XMLSchema#integer) | Number of CPU threads per worker (-1 = auto-detect) | direct | | [precision](../slots/precision.qmd) | 0..1
[xsd:string](http://www.w3.org/2001/XMLSchema#string) | Floating point precision: 'float32' or 'float64' | direct | | [accelerator](../slots/accelerator.qmd) | 0..1
[xsd:string](http://www.w3.org/2001/XMLSchema#string) | Hardware accelerator the run targets | direct | | [backend](../slots/backend.qmd) | 0..1
[xsd:string](http://www.w3.org/2001/XMLSchema#string) | Execution backend this experiment runs on when no backend is passed explicitl... | direct | | [batch_size](../slots/batch_size.qmd) | 0..1
[xsd:integer](http://www.w3.org/2001/XMLSchema#integer) | Batch size for vectorized operations (None = auto) | direct | | [random_seed](../slots/random_seed.qmd) | 0..1
[xsd:integer](http://www.w3.org/2001/XMLSchema#integer) | Experiment-global RNG seed (default 0) | direct | | [find_fixpoint](../slots/find_fixpoint.qmd) | 0..1
[xsd:boolean](http://www.w3.org/2001/XMLSchema#boolean) | Whether to find a fixed point (steady state) before time integration | direct | | [system_size](../slots/system_size.qmd) | 0..1
[xsd:float](http://www.w3.org/2001/XMLSchema#float) | System size (van Kampen Ω) for the stochastic 'gillespie' backend: the number... | direct | ## Usages | used by | used in | type | used | | --- | --- | --- | --- | | [Analysis](../classes/Analysis.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [Optimization](../classes/Optimization.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [Exploration](../classes/Exploration.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [Inference](../classes/Inference.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [Algorithm](../classes/Algorithm.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [Continuation](../classes/Continuation.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | | [SimulationExperiment](../classes/SimulationExperiment.qmd) | [execution](../slots/execution.qmd) | range | [ExecutionConfig](../classes/ExecutionConfig.qmd) | ## Identifier and Mapping Information ### Schema Source * from schema: https://w3id.org/tvbo ## Mappings | Mapping Type | Mapped Value | | --- | --- | | self | tvbo:ExecutionConfig | | native | tvbo:ExecutionConfig | ## LinkML Source ### Direct
```yaml name: ExecutionConfig description: Configuration for computational execution (parallelization, precision, hardware). from_schema: https://w3id.org/tvbo attributes: n_workers: name: n_workers description: Number of parallel workers (maps to pmap devices in JAX, processes in multiprocessing) from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(1) domain_of: - ExecutionConfig range: integer n_threads: name: n_threads description: Number of CPU threads per worker (-1 = auto-detect) from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(-1) domain_of: - ExecutionConfig range: integer precision: name: precision description: 'Floating point precision: ''float32'' or ''float64''' from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: string(float64) domain_of: - ExecutionConfig range: string accelerator: name: accelerator description: Hardware accelerator the run targets. 'auto' (default) lets JAX detect the machine — GPU/TPU when present, else CPU; 'cpu', 'gpu', or 'tpu' pin the JAX platform explicitly. tvbo maps this to JAX_PLATFORMS before JAX initialises (an existing JAX_PLATFORMS in the environment still wins); tvboptim then spreads work across that platform's devices via n_pmap/n_vmap. In a workflow kit, 'gpu' also makes the experiment's rule request a SLURM gres unless one is set. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: string(auto) domain_of: - ExecutionConfig range: string backend: name: backend description: Execution backend this experiment runs on when no backend is passed explicitly (e.g. 'tvboptim', 'tvb', 'jax', 'julia', 'brian2'). Declares the engine the experiment requires — a spiking point-neuron network sets 'brian2', a mean-field model leaves it unset. run(format=...) and ``tvbo run --backend`` override it; when unset the default backend (tvboptim) is used. from_schema: https://w3id.org/tvbo rank: 1000 domain_of: - ExecutionConfig range: string batch_size: name: batch_size description: Batch size for vectorized operations (None = auto) from_schema: https://w3id.org/tvbo domain_of: - Dataset - ExecutionConfig range: integer random_seed: name: random_seed description: Experiment-global RNG seed (default 0). Governs every stochastic draw — additive noise, initial-condition and parameter distributions, and n_trials ensembles — unless a specific `Distribution.seed` overrides it for that distribution. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(0) domain_of: - ExecutionConfig range: integer find_fixpoint: name: find_fixpoint description: Whether to find a fixed point (steady state) before time integration. Used as initial condition for ODEProblem. Maps to NLsolve.fixpoint! in ND.jl or similar in other backends. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: 'False' domain_of: - ExecutionConfig range: boolean system_size: name: system_size description: 'System size (van Kampen Ω) for the stochastic ''gillespie'' backend: the number of discrete units per unit of the activity variable, and the sole source of noise. Runs a mean-field rate model as a finite birth-death process (Gillespie SSA): the activity state variable X becomes a discrete count n ≈ Ω·X, its relaxation equation tau*X'' = -X + F(...) is read as birth propensity Ω·F/tau (the gain) and death propensity n/tau (the leak), while the remaining state variables integrate deterministically between events. Larger Ω means finer discreteness and weaker fluctuations, recovering the deterministic mean field as Ω→∞. For an activity already expressed as a fraction in [0,1], Ω is the population size N; for a firing rate it is N divided by the rate at full activation. Required by the ''gillespie'' backend; ignored by every other backend.' from_schema: https://w3id.org/tvbo rank: 1000 domain_of: - ExecutionConfig range: float class_uri: tvbo:ExecutionConfig ```
### Induced
```yaml name: ExecutionConfig description: Configuration for computational execution (parallelization, precision, hardware). from_schema: https://w3id.org/tvbo attributes: n_workers: name: n_workers description: Number of parallel workers (maps to pmap devices in JAX, processes in multiprocessing) from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(1) owner: ExecutionConfig domain_of: - ExecutionConfig range: integer n_threads: name: n_threads description: Number of CPU threads per worker (-1 = auto-detect) from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(-1) owner: ExecutionConfig domain_of: - ExecutionConfig range: integer precision: name: precision description: 'Floating point precision: ''float32'' or ''float64''' from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: string(float64) owner: ExecutionConfig domain_of: - ExecutionConfig range: string accelerator: name: accelerator description: Hardware accelerator the run targets. 'auto' (default) lets JAX detect the machine — GPU/TPU when present, else CPU; 'cpu', 'gpu', or 'tpu' pin the JAX platform explicitly. tvbo maps this to JAX_PLATFORMS before JAX initialises (an existing JAX_PLATFORMS in the environment still wins); tvboptim then spreads work across that platform's devices via n_pmap/n_vmap. In a workflow kit, 'gpu' also makes the experiment's rule request a SLURM gres unless one is set. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: string(auto) owner: ExecutionConfig domain_of: - ExecutionConfig range: string backend: name: backend description: Execution backend this experiment runs on when no backend is passed explicitly (e.g. 'tvboptim', 'tvb', 'jax', 'julia', 'brian2'). Declares the engine the experiment requires — a spiking point-neuron network sets 'brian2', a mean-field model leaves it unset. run(format=...) and ``tvbo run --backend`` override it; when unset the default backend (tvboptim) is used. from_schema: https://w3id.org/tvbo rank: 1000 owner: ExecutionConfig domain_of: - ExecutionConfig range: string batch_size: name: batch_size description: Batch size for vectorized operations (None = auto) from_schema: https://w3id.org/tvbo owner: ExecutionConfig domain_of: - Dataset - ExecutionConfig range: integer random_seed: name: random_seed description: Experiment-global RNG seed (default 0). Governs every stochastic draw — additive noise, initial-condition and parameter distributions, and n_trials ensembles — unless a specific `Distribution.seed` overrides it for that distribution. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: integer(0) owner: ExecutionConfig domain_of: - ExecutionConfig range: integer find_fixpoint: name: find_fixpoint description: Whether to find a fixed point (steady state) before time integration. Used as initial condition for ODEProblem. Maps to NLsolve.fixpoint! in ND.jl or similar in other backends. from_schema: https://w3id.org/tvbo rank: 1000 ifabsent: 'False' owner: ExecutionConfig domain_of: - ExecutionConfig range: boolean system_size: name: system_size description: 'System size (van Kampen Ω) for the stochastic ''gillespie'' backend: the number of discrete units per unit of the activity variable, and the sole source of noise. Runs a mean-field rate model as a finite birth-death process (Gillespie SSA): the activity state variable X becomes a discrete count n ≈ Ω·X, its relaxation equation tau*X'' = -X + F(...) is read as birth propensity Ω·F/tau (the gain) and death propensity n/tau (the leak), while the remaining state variables integrate deterministically between events. Larger Ω means finer discreteness and weaker fluctuations, recovering the deterministic mean field as Ω→∞. For an activity already expressed as a fraction in [0,1], Ω is the population size N; for a firing rate it is N divided by the rate at full activation. Required by the ''gillespie'' backend; ignored by every other backend.' from_schema: https://w3id.org/tvbo rank: 1000 owner: ExecutionConfig domain_of: - ExecutionConfig range: float class_uri: tvbo:ExecutionConfig ```