Soup turns the pain of LLM fine-tuning into a simple workflow. One config, one command, done.
active 2026-02-20 → 2026-08-09 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 171 days from 2026-02-20 to 2026-08-09. Pushes: 165 total, peak 11 in a day. Pull requests: 3 total, peak 2 in a day. Issues: 58 total, peak 15 in a day. Comments: 17 total, peak 4 in a day. Stars: 17 total, peak 3 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| MakazhanAlpamys | 238 | 165 | 0 | 11 |
| Deadpool2000 | 8 | 0 | 3 | 4 |
| snakefood3232 | 2 | 0 | 0 | 2 |
Recent activity
Latest issues, pull requests and releases
- Issue#335MakazhanAlpamys2026-08-07 16:40use_fsdp2_compile writes a dead adapter: torch.compile's _orig_mod. prefix, exit 0, all lora_B zero on reload
- Issue comment#244MakazhanAlpamys2026-06-07 09:11Thread --track-energy measurement into `soup bom emit` (energy on ML-BOM)
- Issue#170MakazhanAlpamys2026-06-07 09:11soup train --diagnose-gate: multi-node RANK awareness
- Issue comment#193MakazhanAlpamys2026-06-03 12:41v0.61.1: Live NPO / SimNPO / RMU loss kernels + UnlearnTrainerWrapper.setup
- ReleaseMakazhanAlpamys2026-06-01 07:53v0.71.0 — Lighter install: heavy deps split into [train]
- Issue#214MakazhanAlpamys2026-05-21 11:02v0.65.x: judge_calibration registry artifact kind for ensure_judge_calibrated
- Issue#211MakazhanAlpamys2026-05-21 11:02v0.65.1: Live lm-eval-harness invocation in soup eval capability
- Issue#177MakazhanAlpamys2026-05-15 18:07v0.58.1: Pack each soup loop iteration as a v0.26 Soup Can + append to Registry lineage DAG
- ReleaseMakazhanAlpamys2026-05-14 17:11v0.54.0 — soup advise: the pre-flight decision
- Issue comment#155MakazhanAlpamys2026-05-14 09:08v0.54.x: Wire SFT trainer-side tool-call emit into SoupTrainerCallback
- Issue#150MakazhanAlpamys2026-05-14 09:08v0.53.x+: [mix] pyproject extra — bundle scikit-optimize for soup data mix
- Issue#155MakazhanAlpamys2026-05-14 07:24v0.54.x: Wire SFT trainer-side tool-call emit into SoupTrainerCallback
- Issue#28MakazhanAlpamys2026-05-14 07:22soup bench: auto-detect backend (Transformers/vLLM/SGLang/MLX)
- Issue#153MakazhanAlpamys2026-05-13 19:28v0.53.9: soup data download --hub <non-hf> live dataset SDK wiring
- ReleaseMakazhanAlpamys2026-05-13 12:29v0.53.6 — Plugin + Agent + Anthropic API
- Issue comment#129MakazhanAlpamys2026-05-13 08:52v0.50.x: vision_grpo — verify base model is actually a VLM
- Issue comment#109MakazhanAlpamys2026-05-12 19:37v0.46.1: Live Quant-Lobotomy auto-measure for soup deploy autopilot
- Issue comment#142MakazhanAlpamys2026-05-12 19:37v0.53.1: Live merge --save-format 4bit + torchao export writers
- Issue#135MakazhanAlpamys2026-05-12 08:50v0.52.1: EBFT + GDPO live loss kernels
- Issue#127MakazhanAlpamys2026-05-11 19:41v0.50.1: Live HF Trainer callback for the 7 GRPO stability/efficiency knobs
- Issue#109MakazhanAlpamys2026-05-11 12:26v0.46.1: Live Quant-Lobotomy auto-measure for soup deploy autopilot
- Issue#105MakazhanAlpamys2026-05-10 16:31v0.45.1: Advanced trainer plugins — actual upstream callback wiring
- Issue#101MakazhanAlpamys2026-05-10 16:31v0.45.1: trainer callback wiring for plugin pre/post-train + pre/post-step hooks
- Issue#101MakazhanAlpamys2026-05-10 16:31v0.45.1: trainer callback wiring for plugin pre/post-train + pre/post-step hooks
- Issue#101MakazhanAlpamys2026-05-10 16:31v0.45.1: trainer callback wiring for plugin pre/post-train + pre/post-step hooks
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 17 stars here means stars gained during the window, not the repo's star count.