ce_score_job: multiple inputs, OUT_TAG

ce_train_job: INIT/EXTRA/ORIG_N/SKIP_HELDOUT, per-source val

cross-encoder jobs

cross-encoder jobs

frsyn bench: ENCODERS env, w50 blend

encode: ONLY_VARIANT

variant support: WISE_ALPHA, EMB_DIR/OUT_PREFIX

variant support: WISE_ALPHA, EMB_DIR/OUT_PREFIX

Upload code/hf_jobs/ckpt_diag_job.py with huggingface_hub

Upload code/hf_jobs/frsyn_bench_job.py with huggingface_hub

dense_ft_retrieval: SPLITS env, per-split downloads

encode_e5_ft: pre-download private model revision with token

eval_ft_job: pre-download private model with token

status helper

e5 fine-tune chain: all scripts (train, encode, eval driver, prep, launch, log helpers)

dense_retrieval.py

dense_ft retrieval job

eval job: pandas<3, RESULTS_DIR

e5 fine-tune: encode/eval job scripts, prep scripts, eval scripts with e5-ft

e5 fine-tune job script

Upload code/hf_jobs/encode_e5.py with huggingface_hub

test blocking job

Upload code/hf_jobs/encode_e5.py with huggingface_hub

study uses reverse-cap 20 like test

fix train job download

code snapshot for train job

code snapshot