mirror of
https://github.com/harvard-edge/cs249r_book.git
synced 2026-07-16 23:24:55 -05:00
ff614a3682
Update tutorials and landing pages to canonical nested registry paths, fix solver import examples, add Platforms/Datasets/Literature/Ops zoo pages plus provenance docs, and gate doc drift with test_doc_registry_paths.
44 lines
1.4 KiB
Plaintext
44 lines
1.4 KiB
Plaintext
---
|
|
title: "The Platforms Zoo"
|
|
subtitle: "Deployment Paradigm Envelopes"
|
|
---
|
|
|
|
The Platforms Zoo captures **abstract deployment contexts** — latency budgets, RAM envelopes,
|
|
and storage tiers for Cloud, Edge, Mobile, and TinyML. These are teaching anchors, not chip
|
|
specs. Pair them with [`Hardware`](hardware.qmd) entries for concrete silicon.
|
|
|
|
::: {.callout-note}
|
|
## Replaces `Systems.Tiers`
|
|
Platform envelopes moved out of fleet topology into `Platforms.*` so `Systems` stays focused
|
|
on nodes, fabrics, and clusters. See [DATA_MODEL.md](../DATA_MODEL.md) for the full layout.
|
|
:::
|
|
|
|
```{python}
|
|
#| echo: false
|
|
#| output: asis
|
|
import mlsysim
|
|
|
|
print("| Platform | Latency Range | RAM Envelope | Typical Budget |")
|
|
print("|:---|:---|:---|:---:|")
|
|
for p in mlsysim.Platforms.list():
|
|
ram = getattr(p, "ram_range", None) or (
|
|
f"{p.ram.to('GiB').magnitude:.0f} GiB" if p.ram else "---"
|
|
)
|
|
budget = f"{p.typical_latency_budget.to('ms').magnitude:.0f} ms" if p.typical_latency_budget else "---"
|
|
print(f"| **{p.name}** | {p.latency_range_ms} | {ram} | {budget} |")
|
|
```
|
|
|
|
## Python Access
|
|
|
|
```python
|
|
import mlsysim
|
|
|
|
cloud = mlsysim.Platforms.Cloud
|
|
edge = mlsysim.Platforms.Edge
|
|
mobile = mlsysim.Platforms.Mobile
|
|
tiny = mlsysim.Platforms.Tiny
|
|
```
|
|
|
|
Use platform envelopes when comparing deployment paradigms before committing to a specific
|
|
accelerator from the [Silicon Zoo](hardware.qmd).
|