Upload scripts/t128_train.sh with huggingface_hub
Browse files- scripts/t128_train.sh +4 -4
scripts/t128_train.sh
CHANGED
|
@@ -7,7 +7,7 @@
|
|
| 7 |
# per virtual unit. The model re-conditioned to that compressed regime and lost
|
| 8 |
# the positional sensitivity that made compression useful: 90 -> 49 by step 800.
|
| 9 |
#
|
| 10 |
-
# HERE THE INVARIANT HOLDS: SCALE x ORIG == CP_T ->
|
| 11 |
# -> positions span the FULL virtual range 0..4,095
|
| 12 |
# -> 16 tokens per virtual unit, the minimum compression 64k allows
|
| 13 |
# -> this is what p1final was trained under (4 x 4,096 == 16,384) and p1final
|
|
@@ -40,7 +40,7 @@ STABLE=$(( STEPS - WARM - TAIL ))
|
|
| 40 |
GB=524288
|
| 41 |
DATA=Data_Loader/Data/mix128
|
| 42 |
CK=checkpoints_bluscriptcp/rank_0
|
| 43 |
-
PARENT="$CK/
|
| 44 |
say(){ echo "[$(date -u +%H:%M:%S)] $*" | tee -a "$M" >&2; }
|
| 45 |
die(){ say "FATAL: $*"; exit 1; }
|
| 46 |
|
|
@@ -74,7 +74,7 @@ CP_ZERO=1 CP_ZERO_V2=1 CP_BF16=1 CP_OPTIMIZER=muon \
|
|
| 74 |
CP_WARMUP=$WARM CP_REWARMUP=$WARM CP_REWARMUP_PEAK=0.3 \
|
| 75 |
CP_STABLE=$STABLE CP_CKPT=1 CP_CKPT_FREQ=$FREQ CP_CKPT_KEEP=$KEEP"
|
| 76 |
|
| 77 |
-
say "=== 680M 131k extension, 8x
|
| 78 |
say "parent : t64 (98 @65,512 -- the 64k model, staged ladder)"
|
| 79 |
say "YaRN : SCALE=$SCALE x ORIG=$ORIG = $((SCALE*ORIG)) == CP_T INVARIANT HOLDS"
|
| 80 |
say " virtual range 0..$(( (T-1)/SCALE )) (full), $SCALE tokens/virtual unit"
|
|
@@ -90,7 +90,7 @@ if [ ! -f "$LOG/probe_cp${CPSZ}.ok" ]; then
|
|
| 90 |
grep -qaiE 'out of memory|bad_alloc' "$LOG/probe_cp${CPSZ}.log" && die "PROBE OOM at cp=$CPSZ"
|
| 91 |
N=$(grep -acE '^step +[0-9]+ \|' "$LOG/probe_cp${CPSZ}.log"); P=$(grep -aoE 'mem: [0-9]+MB' "$LOG/probe_cp${CPSZ}.log"|tail -1)
|
| 92 |
[ "$RC" -eq 0 ] && [ "$N" -ge 1 ] && [ -n "$P" ] || { say "PROBE FAILED rc=$RC steps=$N"; tail -14 "$LOG/probe_cp${CPSZ}.log" >&2; exit 1; }
|
| 93 |
-
say "PROBE OK cp=$CPSZ: peak $P of
|
| 94 |
touch "$LOG/probe_cp${CPSZ}.ok"
|
| 95 |
fi
|
| 96 |
|
|
|
|
| 7 |
# per virtual unit. The model re-conditioned to that compressed regime and lost
|
| 8 |
# the positional sensitivity that made compression useful: 90 -> 49 by step 800.
|
| 9 |
#
|
| 10 |
+
# HERE THE INVARIANT HOLDS: SCALE x ORIG == CP_T -> 32 x 4,096 == 131,072
|
| 11 |
# -> positions span the FULL virtual range 0..4,095
|
| 12 |
# -> 16 tokens per virtual unit, the minimum compression 64k allows
|
| 13 |
# -> this is what p1final was trained under (4 x 4,096 == 16,384) and p1final
|
|
|
|
| 40 |
GB=524288
|
| 41 |
DATA=Data_Loader/Data/mix128
|
| 42 |
CK=checkpoints_bluscriptcp/rank_0
|
| 43 |
+
PARENT="$CK/t64.ckpt" # the 64k model (98 @65,512)
|
| 44 |
say(){ echo "[$(date -u +%H:%M:%S)] $*" | tee -a "$M" >&2; }
|
| 45 |
die(){ say "FATAL: $*"; exit 1; }
|
| 46 |
|
|
|
|
| 74 |
CP_WARMUP=$WARM CP_REWARMUP=$WARM CP_REWARMUP_PEAK=0.3 \
|
| 75 |
CP_STABLE=$STABLE CP_CKPT=1 CP_CKPT_FREQ=$FREQ CP_CKPT_KEEP=$KEEP"
|
| 76 |
|
| 77 |
+
say "=== 680M 131k extension, 8x H200, cp=$CPSZ dp=$DP ==="
|
| 78 |
say "parent : t64 (98 @65,512 -- the 64k model, staged ladder)"
|
| 79 |
say "YaRN : SCALE=$SCALE x ORIG=$ORIG = $((SCALE*ORIG)) == CP_T INVARIANT HOLDS"
|
| 80 |
say " virtual range 0..$(( (T-1)/SCALE )) (full), $SCALE tokens/virtual unit"
|
|
|
|
| 90 |
grep -qaiE 'out of memory|bad_alloc' "$LOG/probe_cp${CPSZ}.log" && die "PROBE OOM at cp=$CPSZ"
|
| 91 |
N=$(grep -acE '^step +[0-9]+ \|' "$LOG/probe_cp${CPSZ}.log"); P=$(grep -aoE 'mem: [0-9]+MB' "$LOG/probe_cp${CPSZ}.log"|tail -1)
|
| 92 |
[ "$RC" -eq 0 ] && [ "$N" -ge 1 ] && [ -n "$P" ] || { say "PROBE FAILED rc=$RC steps=$N"; tail -14 "$LOG/probe_cp${CPSZ}.log" >&2; exit 1; }
|
| 93 |
+
say "PROBE OK cp=$CPSZ: peak $P of 143771 MiB (H200), dt $(grep -aoE 'dt: [0-9.]+ms' "$LOG/probe_cp${CPSZ}.log"|tail -1)"
|
| 94 |
touch "$LOG/probe_cp${CPSZ}.ok"
|
| 95 |
fi
|
| 96 |
|