biolm.finetune

XGBoost finetune

from biolm.finetune import Finetune

result = Finetune.xgboost(
    train_data=[{"sequence": "MKTAYIAKQRQ", "label": 1}],
    embedding_models=["esm2-8m"],
    task_type="classification",
    run_name="my-xgb-run",
)
print(result["run_id"])

DSM finetune

DSM is a two-stage workflow (masked-LM pretrain, then RL). Launch stage 1:

from biolm.finetune import Finetune

result = Finetune.dsm_stage1(
    train_data=[{"sequence": "MKTAYIAKQRQ"}],
    run_name="my-dsm-stage1",
)
run_id = result["run_id"]
Finetune.wait(run_id)

All methods have async variants (e.g. Finetune.xgboost_async, Finetune.dsm_stage1_async). Pass api_key= or set BIOLM_TOKEN for authentication.

API

class biolm.finetune.Finetune

Launch and track BioLM finetuning runs.

All methods are classmethods returning plain dicts. *_data arguments accept a list of row dicts ([{"sequence": ..., "label": ...}, ...]) or a raw CSV string; they are sent inline as JSON.

classmethod cancel(run_id: str, **kwargs) → dict

Synchronous wrapper for cancel_async().

classmethod dsm_rl(**kwargs) → dict

Synchronous wrapper for dsm_rl_async().

classmethod dsm_stage1(**kwargs) → dict

Synchronous wrapper for dsm_stage1_async().

classmethod dsm_stage2(**kwargs) → dict

Synchronous wrapper for dsm_stage2_async().

classmethod get_run(run_id: str, **kwargs) → dict

Synchronous wrapper for get_run_async().

classmethod progress(run_id: str, **kwargs) → dict

Synchronous wrapper for progress_async().

classmethod wait(run_id: str, *, poll_interval: float = 15.0, timeout: float | None = None, api_key: str | None = None, base_url: str | None = None) → dict

Block until run_id reaches a terminal state, returning its detail.

classmethod xgboost(**kwargs) → dict

Synchronous wrapper for xgboost_async().

See also

• biolm package — biolm.finetune module reference