(
&self,
workers: &[&dyn super::WorkerCapacity],
num_layers: usize,
master_tflops: f64,
layer_size_bytes: u64,
master_max_layers: usize,
layer_p
| 9 | |
| 10 | impl super::Strategy for DefaultStrategy { |
| 11 | fn assign_layers( |
| 12 | &self, |
| 13 | workers: &[&dyn super::WorkerCapacity], |
| 14 | num_layers: usize, |
| 15 | master_tflops: f64, |
| 16 | layer_size_bytes: u64, |
| 17 | master_max_layers: usize, |
| 18 | layer_prefix: &str, |
| 19 | ) -> Vec<(usize, Vec<String>)> { |
| 20 | if workers.is_empty() || num_layers == 0 { |
| 21 | return vec![]; |
| 22 | } |
| 23 | |
| 24 | // Include master TFLOPS in total so layers are split proportionally |
| 25 | let total_tflops: f64 = |
| 26 | workers.iter().map(|w| w.total_tflops()).sum::<f64>() + master_tflops; |
| 27 | |
| 28 | if total_tflops <= 0.0 { |
| 29 | // No compute info — give half to workers, half to master |
| 30 | let worker_layers = num_layers / 2; |
| 31 | let per_worker = worker_layers / workers.len(); |
| 32 | let mut assignments = vec![]; |
| 33 | let mut offset = 0; |
| 34 | for (i, _) in workers.iter().enumerate() { |
| 35 | let count = if i == workers.len() - 1 { |
| 36 | worker_layers - offset |
| 37 | } else { |
| 38 | per_worker |
| 39 | }; |
| 40 | let layers: Vec<String> = (offset..offset + count) |
| 41 | .map(|l| format!("{layer_prefix}.{l}")) |
| 42 | .collect(); |
| 43 | assignments.push((i, layers)); |
| 44 | offset += count; |
| 45 | } |
| 46 | return assignments; |
| 47 | } |
| 48 | |
| 49 | // Sort worker indices by TFLOPS descending |
| 50 | let mut indices: Vec<usize> = (0..workers.len()).collect(); |
| 51 | indices.sort_by(|a, b| { |
| 52 | workers[*b] |
| 53 | .total_tflops() |
| 54 | .partial_cmp(&workers[*a].total_tflops()) |
| 55 | .unwrap_or(std::cmp::Ordering::Equal) |
| 56 | }); |
| 57 | |
| 58 | // Total layers for all workers combined (master keeps its share) |
| 59 | let workers_tflops: f64 = workers.iter().map(|w| w.total_tflops()).sum(); |
| 60 | let total_worker_layers = |
| 61 | (workers_tflops / total_tflops * num_layers as f64).round() as usize; |
| 62 | let total_worker_layers = total_worker_layers.min(num_layers); |
| 63 | |
| 64 | log::info!( |
| 65 | "master: {:.1} TFLOPS — workers: {:.1} TFLOPS — assigning {} of {} layers to workers", |
| 66 | master_tflops, |
| 67 | workers_tflops, |
| 68 | total_worker_layers, |
no test coverage detected