RomanNum3ral
/
soup
mirror of https://github.com/razor-ai/soup.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
soup/soup_cli/utils
History
Alpamys 06fbd15ec4 feat(multipack): v0.37.0 — Multipack (5 Parts A/B/C/D/E) 
FFD bin-packing sampler closes the throughput gap with Axolotl on
uneven-length chat data. Five focused parts:

- Part A: MultipackBatchSampler (FFD + 18-arch allowlist + loud-fail
  vs Axolotl silent miss + _MAX_FFD_ITEMS=1M DoS cap)
- Part B: schema gate (sft/pretrain only on transformers backend,
  multipack/packing mutually exclusive, distinct mlx error),
  build_multipack_sampler_for_lengths helper
- Part C: neat_packing 4D attention mask + FA-vs-4D strategy picker,
  _MAX_MASK_ELEMENTS=2**31 / _MAX_BOUNDARY_SEGMENTS=1M caps
- Part D: JinjaTemplateAnalyzer (parse-only AST walker, 128KB cap)
- Part E: cross-module property tests (4-seed x 200 samples, 5k stress,
  FFD-to-4D-mask coherence)

All five review-agent waves clean before tag (python / code /
security / tdd / verification-loop).
Net +125 tests (4249 -> 4374), 121 test files (+5).

Live HF Trainer sampler-swap wiring deferred to v0.37.1 (mirrors v0.27.0
MII stub-then-live pattern). Schema gate + helper ship now.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 		2026-04-30 13:48:16 +05:00
..
__init__.py Initial project setup: CLI skeleton + config + trainer + data pipeline 2026-02-20 16:14:56 +05:00
activation_offload.py feat(training): Training Speed & Memory — CCE, FP8, grad-ckpt tiers, kernel picker, cross-doc attn, activation offload (v0.28.0) 2026-04-22 22:51:15 +05:00
auto_quant.py feat(trainers): v0.35.0 — Trainer Coverage (closes #60, #61, #45) 2026-04-28 15:01:42 +05:00
batch_probe.py feat(correctness): v0.36.0 — Correctness First (4 Parts: A/B/C/D) 2026-04-30 11:52:51 +05:00
constants.py feat: add GitHub repo link to CLI output, bump version to v0.17.3 2026-03-26 15:48:04 +05:00
convergence.py feat(training): v0.32.0 — Training Stability & Auto-Tuning 2026-04-26 15:24:31 +05:00
crash.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00
cross_doc_attn.py feat(training): Training Speed & Memory — CCE, FP8, grad-ckpt tiers, kernel picker, cross-doc attn, activation offload (v0.28.0) 2026-04-22 22:51:15 +05:00
curriculum.py feat: v0.23.0 — AWQ/GPTQ Export, Sample Packing, Data Split, Curriculum Learning 2026-04-03 13:55:01 +05:00
cut_ce.py feat(training): Training Speed & Memory — CCE, FP8, grad-ckpt tiers, kernel picker, cross-doc attn, activation offload (v0.28.0) 2026-04-22 22:51:15 +05:00
deepspeed.py feat(v0.27.0): Multi-GPU Mastery — topology, ZeRO++, FSDP2+compile, MII, PP, recipes 2026-04-21 14:54:34 +05:00
errors.py v0.10.3: Fix PPO trl >=0.28 API compat, improve GRPO CPU error message 2026-03-24 14:27:39 +05:00
flash_attn.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
fp8.py feat(fp8): add rowwise and rowwise_with_gw_hp scaling recipes for FP8 training (#62) 2026-04-28 20:16:43 +05:00
freeze.py feat: v0.24.0 — Dataset Hub, Freeze Training, Loss Watchdog, Dataset Registry 2026-04-03 16:35:23 +05:00
fsdp.py feat(v0.27.0): Multi-GPU Mastery — topology, ZeRO++, FSDP2+compile, MII, PP, recipes 2026-04-21 14:54:34 +05:00
galore.py v0.12.0: ORPO/SimPO/IPO trainers + DoRA/LoRA+/GaLore 2026-03-25 18:12:36 +05:00
gpu.py v0.10.1: Fix 6 bugs from manual testing report 2026-03-24 12:31:37 +05:00
grad_accum.py feat(training): v0.32.0 — Training Stability & Auto-Tuning 2026-04-26 15:24:31 +05:00
gradient_ckpt.py feat(speed,memory): v0.28.0 features go multi-trainer (v0.33.0 Part C) 2026-04-27 18:55:50 +05:00
hf.py feat(hf): v0.29.0 — HuggingFace Hub Deep Integration 2026-04-23 16:17:28 +05:00
jinja_analyzer.py feat(multipack): v0.37.0 — Multipack (5 Parts A/B/C/D/E) 2026-04-30 13:48:16 +05:00
kernel_picker.py feat(trainers): v0.35.0 — Trainer Coverage (closes #60, #61, #45) 2026-04-28 15:01:42 +05:00
launcher.py feat(v0.27.0): Multi-GPU Mastery — topology, ZeRO++, FSDP2+compile, MII, PP, recipes 2026-04-21 14:54:34 +05:00
liger.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
log_level.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00
long_context.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
lr_finder.py fix(v0.33.0): review-wave findings (CRITICAL + HIGH + MEDIUM + LOW) 2026-04-27 19:57:57 +05:00
metrics.py feat(inference): v0.30.0 — Inference Excellence 2026-04-24 23:39:04 +05:00
mii.py fix(v0.33.0): review-wave findings (CRITICAL + HIGH + MEDIUM + LOW) 2026-04-27 19:57:57 +05:00
mixed_precision.py feat(training): v0.32.0 — Training Stability & Auto-Tuning 2026-04-26 15:24:31 +05:00
mlx.py feat(v0.25.0): Beyond the Wrapper — 8 major features 2026-04-13 12:58:11 +05:00
moe.py feat: v0.14.0 — pre-training + MoE support 2026-03-25 22:26:01 +05:00
multipack.py feat(multipack): v0.37.0 — Multipack (5 Parts A/B/C/D/E) 2026-04-30 13:48:16 +05:00
multipack_sampler.py feat(multipack): v0.37.0 — Multipack (5 Parts A/B/C/D/E) 2026-04-30 13:48:16 +05:00
neat_packing.py feat(multipack): v0.37.0 — Multipack (5 Parts A/B/C/D/E) 2026-04-30 13:48:16 +05:00
ollama.py feat: add Ollama integration — deploy GGUF models in one command (v0.18.0) 2026-04-01 13:47:40 +05:00
paths.py feat(v0.26.0): Parts B-E — Eval Gate, Trace-to-Pref, Quant-Check, Soup Cans 2026-04-20 21:37:05 +05:00
peft_builder.py feat(v0.25.0): Beyond the Wrapper — 8 major features 2026-04-13 12:58:11 +05:00
pipeline.py feat(v0.27.0): Multi-GPU Mastery — topology, ZeRO++, FSDP2+compile, MII, PP, recipes 2026-04-21 14:54:34 +05:00
profiler.py fix: v0.24.1 — Windows Unicode fix, AWQ/GPTQ output path traversal 2026-04-03 23:41:44 +05:00
profiling.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00
qat.py Add Quantization-Aware Training support (Phase 7) — v0.6.0 2026-03-23 20:35:35 +05:00
quality.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
registry.py feat: v0.24.0 — Dataset Hub, Freeze Training, Loss Watchdog, Dataset Registry 2026-04-03 16:35:23 +05:00
replay.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00
ring_attention.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
run_cost.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00
sglang.py fix: rename APIs to match test plan, fix RoPE factor detection 2026-03-26 15:14:24 +05:00
spec_pairing.py feat(inference): v0.30.0 — Inference Excellence 2026-04-24 23:39:04 +05:00
spike_recovery.py feat(training): v0.32.0 — Training Stability & Auto-Tuning 2026-04-26 15:24:31 +05:00
structured_output.py feat(serve): structured-output + auto-quant live (v0.33.0 Part D) 2026-04-27 18:43:07 +05:00
topology.py feat(v0.27.0): Multi-GPU Mastery — topology, ZeRO++, FSDP2+compile, MII, PP, recipes 2026-04-21 14:54:34 +05:00
tracing.py feat(inference): v0.30.0 — Inference Excellence 2026-04-24 23:39:04 +05:00
trust_remote.py feat(correctness): v0.36.0 — Correctness First (4 Parts: A/B/C/D) 2026-04-30 11:52:51 +05:00
unsloth.py Add Unsloth backend for 2-5x faster training (Phase 5) — v0.4.3 2026-03-23 16:55:44 +05:00
v028_features.py fix(fp8): wire fp8_recipe through v028_features (covers 10 trainers) 2026-04-28 20:19:58 +05:00
vllm.py feat(trainers): v0.35.0 — Trainer Coverage (closes #60, #61, #45) 2026-04-28 15:01:42 +05:00
warmup.py feat(training): v0.32.0 — Training Stability & Auto-Tuning 2026-04-26 15:24:31 +05:00
why.py feat(observability): v0.34.0 — Observability & Dev UX (7 Parts) 2026-04-28 13:14:49 +05:00