File size: 3,290 Bytes
506d5cf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 | ---
license: mit
pretty_name: BoLT Parallelism Configuration (32 GPUs)
task_categories:
- other
tags:
- bayesian-optimization
- benchmark
- parallelism-configuration
- tabular
size_categories:
- n<1K
---
# pco32
Throughput of distributed LLM training as a function of the parallelism
configuration, on 32 GPUs across 8 hosts.
Tabular benchmark for the `bolt` problem `pco32`: the 379
rows are the full candidate set.
- **Objective** `throughput_mean`, to maximise. `NaN` where the run OOMed, since
no throughput is observed at all -- a hidden (crash) constraint, not a bad value.
- **Constraint** `ran_successfully`. 339 of 379
configurations run; the rest exhaust GPU memory. Feasibility is only learnable
by attempting the run, so it cannot be decoupled from the objective.
- **Auxiliary** `peak_mem` (GB, `inf` when infeasible), `peak_mem_normalised`
(the run's own feasibility margin, `< 0` iff feasible, a fixed pad value when
infeasible) and `overall_time` (wall clock in seconds, `NaN` when infeasible).
`peak_mem` does **not** define the constraint: feasible runs here reach
88.5 GB, so no memory threshold separates the
classes. `peak_mem_normalised` does, by construction.
## Instance context
Constant for every row; these describe the instance, not the search space.
| field | value |
|---|---|
| `num_layers` | 40 |
| `num_attention_heads` | 40 |
| `hidden_size` | 5120 |
| `ffn_hidden_size` | 17408 |
| `batch_size` | 512 |
| `seq_length` | 8192 |
| `log2_num_gpus` | 5 |
| `log2_num_hosts` | 3 |
| `log2_grad_accum_steps` | 0 |
## Columns
The table is the raw sweep export, unchanged. Every knob halves or doubles, so
the sizes arrive as `log2_*` and their levels are evenly spaced. `zero_stage` is
the raw stage (0/2/3); `bolt` searches it as an ordinal 0/1/2.
| column | role | meaning |
|---|---|---|
| `log2_dp_size` | input | log2 of data-parallel size |
| `log2_tp_size` | input | log2 of tensor-parallel size |
| `log2_pp_size` | input | log2 of pipeline-parallel size |
| `log2_cp_size` | input | log2 of context-parallel size |
| `log2_sp_size` | input | log2 of sequence-parallel size |
| `log2_dp_bucket_size_mb` | input | log2 of DDP gradient bucket size in MB |
| `zero_stage` | input | ZeRO stage, one of [0, 2, 3] |
| `log2_num_model_chunks` | input | log2 of number of model chunks |
| `log2_grad_accum_steps` | context | log2 of gradient accumulation steps |
| `ran_successfully` | output | ground-truth constraint label |
| `throughput_mean` | output | objective, `NaN` when infeasible |
| `throughput_std` | output | spread over the timed steps, `NaN` when infeasible |
| `peak_mem_normalised` | output | feasibility margin, `< 0` iff feasible |
| `overall_time` | output | wall clock in seconds, `NaN` when infeasible |
| `all_time_steps` | output | per-step times, as a string list |
| `peak_mem` | output | peak GPU memory in GB, `inf` when infeasible |
| `peak_mem_per_gpu` | output | per-GPU peak memory, as a string list |
## Best observed
`throughput_mean` = 0.46958 at `log2_dp_size`=5, `log2_tp_size`=0, `log2_pp_size`=0, `log2_cp_size`=0, `log2_sp_size`=0, `log2_dp_bucket_size_mb`=8, `zero_stage`=2, `log2_num_model_chunks`=0
(peak memory 82.6 GB).
## Provenance
`data_40depth-32gpus.csv`, converted to parquet unchanged.
|