compilation resource params

General
1

Hi there,

I’ve successfully converted a model i’ve successfully compiled using when setting thecompiler_optimization_level to max.

Here is the utilisation table:

[info] ±----------±--------------------±--------------------±-------------------+
[info] | Cluster | Control Utilization | Compute Utilization | Memory Utilization |
[info] ±----------±--------------------±--------------------±-------------------+
[info] | cluster_0 | 100% | 60.9% | 35.9% |
[info] | cluster_1 | 100% | 68.8% | 65.6% |
[info] | cluster_2 | 87.5% | 100% | 61.7% |
[info] | cluster_3 | 100% | 87.5% | 97.7% |
[info] | cluster_4 | 81.3% | 92.2% | 96.9% |
[info] | cluster_5 | 93.8% | 98.4% | 93.8% |
[info] | cluster_6 | 100% | 60.9% | 78.1% |
[info] | cluster_7 | 81.3% | 84.4% | 50% |
[info] ±----------±--------------------±--------------------±-------------------+
[info] | Total | 93% | 81.6% | 72.5% |
[info] ±----------±--------------------±--------------------±-------------------+

This process took many many hours. How would I configure the model script to ensure the same compilation, without needing to exhaustively search for optimal params.

Could I use commands like these:

performance_param(compiler_optimization_level=1)
resources_param(strategy=greedy, max_control_utilization=1.0, max_compute_utilization=1.0, max_memory_utilization=0.97)
context_switch_param(mode=disabled)

If so, based on the above utilisation table, what should I set the utilisation thresholds as? Is the utilisation threshold the cluster utilisation or the total utilisation for each type of utilisation?

Thanks

2

Hey @piers.turner ,

Welcome back to the Hailo Community!

If you want to keep this cluster and utilization as-is, you can skip ahead to just the compilation step—since you’ve already got your optimized HAR file, you can make any adjustments you need during compilation. But if you decide you want to run the optimization step again, you’ll need to redo everything from the start.