InstantMLDocs

Configs, tags, and notes

Attach config, tags, and notes so runs stay searchable and comparable.

Open .md

Give each run enough identity — config, tags, and notes — to explain what changed without rereading the code.

Set config

Pass config at run creation:

python
import instantml as im

run = im.init(
    project="mnist",
    name="cnn-seed-7",
    config={
        "seed": 7,
        "model": "small-cnn",
        "optimizer": {"name": "adam", "lr": 3e-4},
        "batch_size": 128,
    },
)

Add or update config later:

python
run.log_config({"schedule": {"warmup_steps": 500}})

Use config for values that should appear in Run Detail and Compare.

Tag runs

Use set_tags(...) for the tags you want the Runs list, search, and Compare views to treat as run identity:

python
run.set_tags(["baseline", "ready-for-compare"])

Use add_tags(...) when you want to append tag-style history:

python
run.add_tags(["manual-review"])

Write notes

Notes are searchable free text:

python
run.set_notes("New tokenizer improved eval accuracy but increased train loss.")

Good notes explain what a future reviewer would otherwise have to reconstruct from memory.

Find them in the UI

SDK inputDashboard surface
configRun Detail, Compare config rows, search/filter context
tagsRun rows, run selector chips, Run Detail, Compare, run search
notesRuns list, Run Detail, Compare, run search

Put it together

python
run = im.init(
    project="finetune",
    name="llama8b-lora-seed-12",
    config={
        "seed": 12,
        "base_model": "llama-8b",
        "adapter": "lora",
        "dataset": "support-tickets-v3",
    },
    tags=["lora", "support", "baseline"],
    notes="Baseline before adding retrieval examples.",
)

Next steps