73 lines
3.1 KiB
Bash
Executable File
73 lines
3.1 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# kld-gate-swift15.sh — RUNS ON DANKSERVER. Score the quantized candidate
|
|
# against the BF16 reference and apply the pre-approved gate:
|
|
# mean KL <= 0.03 AND top-1 agreement >= 95%.
|
|
# Same engine (1.0.13) and same boot shape as the reference RECORD
|
|
# (run-swift15-bf16.sh kld): disk-tier, no speculation, deterministic, fp8 KV.
|
|
# The engine compares its own logits against /data/kld-ref and prints the
|
|
# score table at exit. Gate failure exits 1 (escalation owner: recipe author);
|
|
# the script never touches the production unit.
|
|
set -uo pipefail
|
|
PIN=sha256:4c3f484564bb2872784e46b0a59cbd71becc9c1e746e301a3596ffae253c322e
|
|
RAD=/home/dank/models/rad/qwen4exp-swift15-w4dn8-i8-hc8m.rad
|
|
C=swift15gate
|
|
LOG=/home/dank/logs/swift15_kldgate.log
|
|
KLD_REF_DIR=/home/dank/corpora/kld-ref
|
|
KLD_CORPUS=/home/dank/corpora/out/kld.jsonl
|
|
KL_MAX=0.03
|
|
TOP1_MIN=95.0
|
|
|
|
[ -d "$KLD_REF_DIR" ] || { echo "missing reference: $KLD_REF_DIR"; exit 2; }
|
|
[ -f "$KLD_CORPUS" ] || { echo "missing corpus: $KLD_CORPUS"; exit 2; }
|
|
|
|
cp -f /home/dank/serverAppConfigs/scripts/qwen38-froggeric-v22.5.jinja \
|
|
/home/dank/models/tpl.jinja 2>/dev/null || true
|
|
sudo systemctl stop frank-primary >/dev/null 2>&1 || true
|
|
podman rm -f "$C" >/dev/null 2>&1 || true
|
|
podman run --rm --name "$C" \
|
|
--network host --init \
|
|
--security-opt seccomp=unconfined \
|
|
--ulimit memlock=-1:-1 \
|
|
--device /dev/kfd --device /dev/dri \
|
|
-v /home/dank/models:/models:ro \
|
|
-v /home/dank/radiance-state:/data \
|
|
-v "$KLD_CORPUS:/data/kld.jsonl:ro" \
|
|
-v "$KLD_REF_DIR:/data/kld-ref:ro" \
|
|
"stilldeadcode/radiance@$PIN" \
|
|
--model /models/rad/qwen4exp-swift15-w4dn8-i8-hc8m.rad \
|
|
--tp 2 --placement expert_tiered --weights-disk-tier \
|
|
--host-pool-mib 12288 --deterministic --gpu-headroom-mib 768 \
|
|
--num-speculative-tokens 0 \
|
|
--kv-cache-dtype fp8 --max-model-len 16384 \
|
|
--ngram-placement ram \
|
|
--override-chat-template /models/tpl.jinja \
|
|
--kld-ref /data/kld-ref --kld-corpus /data/kld.jsonl \
|
|
--vram-kv-mib 3072 \
|
|
--max-num-seqs 4 --max-num-batched-tokens 32768 \
|
|
2>&1 | tee "$LOG"
|
|
rc=${PIPESTATUS[0]}
|
|
[ "$rc" -eq 0 ] || { echo "KLD GATE BOOT FAILED rc=$rc; log: $LOG"; exit "$rc"; }
|
|
|
|
# Engine prints its score table on stdout (captured in $LOG). Pull the two
|
|
# gate numbers; the summary line is "mean KL"/"top-1" in the radiance
|
|
# scorecard; fall back to any numeric KL/top1 lines.
|
|
python3 - "$LOG" <<'PY'
|
|
import re, sys
|
|
log = open(sys.argv[1], errors="replace").read()
|
|
def num(pat):
|
|
m = re.findall(pat, log, re.I)
|
|
return float(m[-1]) if m else None
|
|
kl = num(r"mean[_ ]kl[^0-9-]*([0-9]*\.?[0-9]+(?:e[-+][0-9]+)?)")
|
|
top1 = num(r"top[-_ ]?1[^0-9]*([0-9]*\.?[0-9]+)")
|
|
if kl is None or top1 is None:
|
|
print("KLD GATE PARSE FAILED: mean_kl=%s top1=%s — inspect %s" % (kl, top1, sys.argv[1]))
|
|
sys.exit(2)
|
|
print("KLD GATE: mean KL = %.4f (max %.2f) | top-1 = %.2f%% (min %.1f)" % (kl, 0.03, top1, 95.0))
|
|
ok = (kl <= 0.03) and (top1 >= 95.0)
|
|
print("KLD GATE " + ("PASS" if ok else "FAIL"))
|
|
sys.exit(0 if ok else 1)
|
|
PY
|
|
gate=$?
|
|
echo "kld-gate complete rc=$gate; log: $LOG"
|
|
exit "$gate"
|