Upload scripts/eval_saturate.sh with huggingface_hub
Browse files- scripts/eval_saturate.sh +33 -0
scripts/eval_saturate.sh
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env bash
|
| 2 |
+
# Phase 1: base680 positional-competence GRID -- 8 small jobs, all 4 GPUs.
|
| 3 |
+
# scale=1 (virtual==real): 512 1024 2048 3072 4096 -> does the BASE hold to 4,096?
|
| 4 |
+
# scale=4 (virtual=real/4): 4096 8192 16360 -> same base, compressed
|
| 5 |
+
# If base is flat at scale=1 but the extensions aren't, the damage is in the ladder.
|
| 6 |
+
# If base drops at real 4,096, theta=500,000 at a 4,096 pretrain is the ceiling.
|
| 7 |
+
# Phase 2: x64 arm, 6 big jobs.
|
| 8 |
+
set -uo pipefail; cd "$(dirname "$0")"
|
| 9 |
+
export LD_LIBRARY_PATH="$(cd ../.. && pwd)/Tensor-Implementations/lib:$(cd ../.. && pwd)/Profiler/lib:${LD_LIBRARY_PATH:-}"
|
| 10 |
+
export PATH=/usr/local/cuda/bin:$PATH; export PY=/root/Adhi/BluTrain/envv/bin/python
|
| 11 |
+
A680="CP_N_EMBD=1536 CP_N_LAYER=24 CP_N_HEAD=12 CP_N_KVHEAD=4 CP_FFN=4096 CP_WEIGHT_TYING=1"
|
| 12 |
+
L=eval_sat_logs; mkdir -p "$L"; M="$L/master.log"
|
| 13 |
+
say(){ echo "[$(date -u +%H:%M:%S)] $*" | tee -a "$M" >&2; }
|
| 14 |
+
one(){ # scale T promptlen gpu tag
|
| 15 |
+
local S=$1 TT=$2 PL=$3 G=$4 O="eb_s${S}_${PL}"
|
| 16 |
+
A1B="$A680" HAY="data/mix128_b6_60/val_shard_00000.bin" GEN_DATA_ROOT="data/mix128_b6_60" \
|
| 17 |
+
CKPT_EXTRA="base=base680.ckpt" CKPTS="base" TASKS="niah_single" N=100 \
|
| 18 |
+
T=$TT YARN=$S YARN_ORIG=4096 PROMPT_LEN=$PL GPUS="$G" NP=1 JOBS_PER_GPU=1 OUT="$O" \
|
| 19 |
+
./run_longeval.sh > "$L/s${S}_${PL}.log" 2>&1
|
| 20 |
+
say " base680 scale=$S len=$PL (virtual $((PL/S))) -> $(grep -h '^niah_single,' $O/summary.csv 2>/dev/null|cut -d, -f2|tr -d '\r')"
|
| 21 |
+
}
|
| 22 |
+
say "=== PHASE 1: base680 grid, 8 jobs across 4 GPUs ==="
|
| 23 |
+
one 1 8192 512 0 & one 1 8192 1024 1 & one 1 8192 2048 2 & one 1 8192 3072 3 &
|
| 24 |
+
one 1 8192 4096 0 & one 4 16384 4096 1 & one 4 16384 8192 2 & one 4 16384 16360 3 &
|
| 25 |
+
wait
|
| 26 |
+
say "=== BASE680 GRID (niah_single, N=100) ==="
|
| 27 |
+
for c in "1 512" "1 1024" "1 2048" "1 3072" "1 4096" "4 4096" "4 8192" "4 16360"; do
|
| 28 |
+
set -- $c; v=$(grep -h '^niah_single,' eb_s$1_$2/summary.csv 2>/dev/null|cut -d, -f2|tr -d '\r')
|
| 29 |
+
printf " scale=%-2s len=%-6s virtual=%-5s -> %s\n" "$1" "$2" "$(($2/$1))" "${v:-.}" | tee -a "$M"
|
| 30 |
+
done
|
| 31 |
+
say "=== PHASE 2: x64 arm ==="
|
| 32 |
+
./eval_x64.sh >> eval_x64.boot 2>&1
|
| 33 |
+
say "SATURATE DONE"
|