Deployment¶

Deploy¶

$ threading deploy --target gcp-a100
[deploy] Containerizing workspace...
[deploy] Provisioning cluster [4x A100-80GB]
[deploy] MCP Server live: https://api.threading.cloud/v1/exp-92a

Targets¶

Target	GPUs	Best for
`gcp-a100`	4× A100-80GB	Large models
`gcp-v100`	4× V100-16GB	Standard ML
`aws-p4d`	8× A100-40GB	Largest scale
`aws-p3`	4× V100-16GB	Standard ML
`azure-a100`	4× A100-80GB	Large models
`local-gpu`	Your GPU(s)	Development

Load data¶

$ threading ld s3://bio-data/microbiome_full_v2
[dataset] Mounting to /mnt/data...
[dataset] Sharding across 4 nodes

Scheme	Example
`s3://`	`s3://bucket/path/`
`gs://`	`gs://bucket/path/`
`az://`	`az://container/path/`

Start jobs¶

$ threading start
[job] Spawning 512 workers...
[throughput] 12,500 samples/sec
[estimate] 2m 18s

Parameter sweeps¶

threading start --sweep "pca.n_components=[10,20,30,40,50]"

Monitor¶

threading status
threading logs -f

Teardown¶

threading stop --teardown

Cost optimization¶

Use spot instances:

threading deploy --target gcp-a100 --spot

Auto-shutdown:

threading deploy --target gcp-a100 --auto-shutdown 30m