mirror of
https://github.com/harvard-edge/cs249r_book.git
synced 2026-05-07 18:18:42 -05:00
23 lines
428 B
YAML
23 lines
428 B
YAML
version: "1.0"
|
|
name: "Llama-3 70B Distributed Training with Assertions"
|
|
|
|
workload:
|
|
name: "Llama3_70B"
|
|
batch_size: 4096
|
|
seq_len: 2048
|
|
|
|
hardware:
|
|
name: "H100"
|
|
nodes: 64
|
|
precision: "fp16"
|
|
efficiency: 0.45
|
|
|
|
ops:
|
|
region: "Quebec"
|
|
duration_days: 14.0
|
|
|
|
constraints:
|
|
assert:
|
|
- metric: "macro.carbon_footprint"
|
|
max: 1.0 # Force a failure (Quebec is clean but 1.0 tons is too low for 64 H100s over 14 days)
|