EMC / control /run.env
mudassiruddin21's picture
Run r3: recipe comparison, keep-the-better rule, extra measurements
a64b4b7 verified
Raw History Blame Contribute Delete
725 Bytes
# KEY=VALUE settings passed to scripts/pipeline.sh by scripts/supervise.py
# run r3: 50% more data (earlier answers and features kept), four training recipes compared in
# short runs, the best one trained from the r2 head; the new draft replaces r2's only if it
# scores higher on this machine; then the batch / long-context / draft-length measurements
RUN_ID_FORCE=r3
GEN_TOKENS=6000000
GEN_MINUTES=75
GEN_APPEND=1
EXTRACT_MINUTES=50
EXTRACT_APPEND=1
ABLATE=1
ABLATE_STEPS=300
TRAIN_MINUTES=60
RUN_NAME=main3
TRAIN_INIT=runs/main/sub_bf16
TRAIN_INIT_HF=mudassiruddin21/EMC:submissions/r2/main-bf16
PREV_TAG=eval_draft_fp8
PREV_SUB=runs/main/sub_fp8
PREV_SUB_HF=mudassiruddin21/EMC:submissions/r2/main-fp8
EVAL_ALT=0
BENCH=1