Strata

0 Fork 0

mirror of https://github.com/Niko1221/Strata.git synced 2026-11-09 17:36:14 -10:00

Files

T

Constantin cb87a62e82 ple: read the n-gram table at full BF16 precision (#586 )

The full checkpoint stores the table as BF16 (320-byte rows, 102.4 GB), the source of record; the engine gets it as
IQ4_NL (about 8% off per row) or FP8. A bfloat16 is the top half of a float32, so widening is a shift: exact for
every value, no rounding and no scale. One entry in the PLE format list (type BF16, self-describing: no scale, no
metadata) and bf16_dequant_row; the direct reader takes the 320-byte row like any other width, and
PLE_ROW_BYTES_MAX is 320.

ple_reader_test checks the widening is exact over the patterns a lossy path breaks on (subnormals, +-inf, both NaNs);
the generic every-format test covers a synthetic BF16 GGUF through both readers.

Not in this commit: the PR's rename of tools/ple_fp8_pack.py (nothing needs it) and any setup option. Whether a
table format other than IQ4_NL is worth offering is a quality question (see the measurement in the 0.1.40 notes).

Ported by the maintainer from constantindjonkam's PR #586.

2026-10-05 12:18:48 +02:00

artifact

ple: read Q8_0 and Q5_1 n-gram tables (#651 )

2026-10-05 12:18:47 +02:00

core

verify: the all-resident graph runs only while every expert is in VRAM; a -1 raises a plan error flag (#871 )

2026-10-05 11:03:50 +02:00

hip_compat

Merge branch 'f-139b-spd' into f-139-rel

2026-10-04 10:12:55 +02:00

kernels

ple: read the n-gram table at full BF16 precision (#586 )