Commits

History for

onfix/473-downsample-hard-cap

Commits on Aug 20, 2026

feat(stream): support heterogeneous decoder layers and admit Qwen3.5 MoE
Shutaru
authored
82bce26
feat(data): wire data.interleave into training-time dataset loading (#443)
blackcoderx
authored
a50264d
feat(stream): make layer-streaming pinning configurable via training.stream_pin (#366)
ousamabenyounes
authored
9d1ce2c

Commits on Aug 17, 2026

fix(quant): honour training.bnb_4bit_use_double_quant across every 4-bit path (#321)
ousamabenyounes
authored
c382c80

Commits on Aug 16, 2026

Commits on Aug 14, 2026

feat(stream): measure the VRAM fit instead of predicting it (#349)
MakazhanAlpamys
committed
36e9f71

Commits on Aug 13, 2026

feat(stream): let training.stream_vram_override replace the measured free VRAM (#386)
AmirF194
authored
d31c1e5

Commits on Aug 8, 2026

feat(train): #341 #340 - a real training seed, and full fine-tuning as lora.r=0
MakazhanAlpamys
committed
dcb5eb5

Commits on Aug 3, 2026

feat(streaming): preference losses over layer streaming (v0.72.4)
MakazhanAlpamys
committed
4fb2540

Commits on Jul 28, 2026

Commits on Jul 27, 2026

fix(train): streamed adapters were saved unloadable (v0.72.1)
MakazhanAlpamys
committed
af8683b

Commits on Jul 26, 2026

feat(train): layer streaming — fine-tune models larger than VRAM (v0.72.0 BETA)
MakazhanAlpamys
committed
00833ac

Commits on Jul 19, 2026

Commits on Jul 17, 2026

fix(cli): quote install hints so `pip install soup-cli[extra]` works on cmd.exe (v0.71.37)

MakazhanAlpamys
and
claude
committed
a95fede

Commits on Jul 16, 2026

Commits on Jul 15, 2026

Commits on Jul 8, 2026

fix: resolve full-tree code-review findings (6 HIGH + MEDIUM/LOW)

MakazhanAlpamys
and
claude
committed
3a549d3

Commits on Jul 6, 2026

Commits on Jul 5, 2026

Commits on Jul 1, 2026