Plan data and compute¶
Level: Engineer → Research · Time: 55 minutes
A training plan is a coupled budget across tokens, model operations, device memory, communication, storage, reliability, and evaluation. A single “GPU count” is not a plan.
Token budget and mixture¶
Define the number of tokens consumed by training, then derive sampling weights from the intended mixture. If sources are sampled with replacement, a small high-weight source may repeat for many effective epochs while a large low-weight source is barely seen.
For source \(i\) with available tokens \(N_i\), mixture probability \(p_i\), and total consumed tokens \(N\), an approximate expected pass count is:
This simplification assumes token-level sampling and ignores packing, filters, and sampling implementation. Report the actual sampler contract.
A first compute estimate¶
A commonly used back-of-the-envelope estimate for dense Transformer training is proportional to parameter count times training tokens:
Here \(P\) is an appropriate active non-embedding parameter approximation and \(N\) is training tokens. The factor is a heuristic, not a hardware forecast: attention at long sequence length, embeddings, sparse routing, activation recomputation, optimizer, and implementation details matter.
Convert required operations to time only after applying an evidence-based model FLOP utilization:
Peak vendor throughput is not sustained training throughput.
Memory ledger¶
Budget separately:
| Category | Scales with | Possible controls |
|---|---|---|
| Weights | total resident parameters | sharding, lower precision, offload |
| Gradients | trainable parameters | sharding, lower precision |
| Optimizer states | parameters and optimizer | sharding, lower-state optimizers, offload |
| Activations | batch, sequence, width, layers | checkpointing, sequence parallelism |
| Attention temporaries | sequence and kernel | memory-efficient kernels |
| Communication buffers | parallel topology | bucket sizing, overlap, implementation |
| Fragmentation/runtime | allocator and workload | headroom, profiling |
Do not allocate 100% of nominal memory on paper. Recovery, evaluation, compilation, and occasional shape tails need headroom.
Parallelism selection¶
flowchart TD
Q{Does one model replica fit?}
Q -- yes --> DP[Data parallelism]
Q -- no --> FS[Shard states with FSDP or ZeRO]
FS --> T{Does one layer fit and run efficiently?}
T -- no --> TP[Tensor or sequence parallelism]
T -- yes --> PP[Consider pipeline parallelism at scale]
TP --> PP
E{MoE experts?} --> EP[Expert parallelism and all-to-all]
DP --> E
PP --> E Real meshes compose dimensions. The best topology depends on intra-node bandwidth, inter-node fabric, layer shapes, batch, sequence, experts, framework, and failure domain. Benchmark the intended configuration.
Storage and network¶
Plan for:
- raw and processed data snapshots;
- tokenizer and manifests;
- training shards and indices;
- complete and lightweight checkpoints;
- optimizer state and RNG/data-loader state;
- logs and raw evaluation outputs;
- checkpoint upload bandwidth and retention;
- restoration drills, not only backups.
A checkpoint cadence is a reliability/economics decision. Lost compute from a failure grows with the interval; checkpoint overhead grows as the interval shrinks.
Pilot ladder¶
- Unit: shapes, masks, losses, serialization.
- One batch: overfit and verify loss decreases.
- One device: measure memory, throughput, data correctness.
- One node: validate distributed equivalence and checkpoint resume.
- Few nodes: measure communication, failure recovery, stragglers.
- Short scaled run: validate slopes, stability, evaluations, and operations.
- Full run: only after exit criteria are met.
Each rung should have explicit pass/fail thresholds.
Training readiness review¶
- immutable configs and code revision captured;
- source and processed-data manifests signed off;
- tokenizer checksum and test vectors recorded;
- sample batches decoded and manually inspected;
- evaluation contamination and prompt versions documented;
- memory and throughput measured on target hardware;
- loss scale, gradient norm, and update norm dashboards tested;
- checkpoint resume including data position demonstrated;
- stop conditions, escalation, and ownership defined;
- cost and capacity buffer approved.