soup-cli 0.71.29__tar.gz → 0.71.31__tar.gz

{soup_cli-0.71.29 → soup_cli-0.71.31}/.gitignore

@@ -40,9 +40,12 @@ report.xml
 .env
 *.key
 
-# Internal plan + Claude Code local dev instructions (not for repo)
+# Internal plan + local dev instructions (not for repo)
 .claude/
 .coverage
 
+# Internal brainstorming plans + design specs (local-only, not for repo)
+docs/superpowers/
+
 # Local scratch / temp scripts at repo root (e.g. _smoke.py, _scratch.py) — never committed
 /_*.py

{soup_cli-0.71.29 → soup_cli-0.71.31}/CHANGELOG.md

@@ -12,6 +12,76 @@ reproducing 70+ versions of notes.
 
 ## [Unreleased]
 
+## [0.71.31] - 2026-07-06
+
+### Added
+- **Judge-in-the-loop suite** — put an LLM judge in the loop across the workflow:
+  - **`task='online_dpo'`** — Online DPO training (wraps TRL `OnlineDPOTrainer`):
+    the model generates two completions per prompt on-policy each step and a
+    *judge* (a pairwise LLM judge over the existing ollama/openai-compatible
+    backend) OR a `reward_model` picks the winner. Config:
+    `training.online_dpo_judge: "ollama://model"` (or set `reward_model` —
+    exactly one), `online_dpo_loss_type: sigmoid|ipo`, `online_dpo_max_new_tokens`;
+    `beta` reuses `dpo_beta`. Transformers + text only. Recipe:
+    `online-dpo-smollm2-135m`. Adapts to the installed TRL: on trl 0.19.x the
+    judge is a swap-debiased *pairwise* comparison; on trl 1.x (which removed
+    pairwise judges) the same `JudgeEvaluator` is used as a *pointwise*
+    reward function — a documented per-version behaviour difference.
+  - **`soup data best-of-n`** — Best-of-N rejection sampling (BOND-lite): sample
+    N completions from `--base` locally, a `--judge` scores each pointwise, and
+    the winner is written as an SFT chat row (with provenance). `--emit-pairs`
+    also writes winner-vs-loser DPO pairs.
+  - **`soup data evolve`** — Evol-Instruct instruction evolution (WizardLM depth
+    / breadth) over an ollama/vllm provider, completing the synthetic-data suite
+    (Magpie / Forge / Persona / evolve).
+  - **`soup ship --task-mode pairwise`** — a true pairwise judge win-rate as the
+    ship leg-1 task-win (base = 0.5 coin-flip, tuned = its win-rate; swap-debiased),
+    fusing with the catastrophic-forgetting guard into one SHIP / DON'T-SHIP verdict.
+
+### Security
+- `soup data best-of-n` / `evolve` write outputs via atomic `mkstemp` + `os.replace`
+  (re-validated cwd containment), closing the TOCTOU symlink-swap window between the
+  containment check and the write. All judge/provider URLs are SSRF-validated; model
+  loads probe `trust_remote_code`.
+
+## [0.71.30] - 2026-07-05
+
+### Added
+- **PRM-guided GRPO** — use a trained Process Reward Model as the *per-step*
+  reward inside GRPO (the o1-era process-supervision signal). Set
+  `training.prm_reward: <PRM dir|id>` (a model produced by `soup train`
+  `task=prm`) and `training.prm_aggregate: min|prod|last`; the PRM splits each
+  generated completion into reasoning steps, scores every step with its reward
+  head, and folds the per-step scores into one scalar reward that GRPO
+  optimises. The PRM reward *replaces* `reward_fn` and rides the existing
+  reward-shaping + reward-hack-mitigation seam, so the v0.71.26 controller
+  observes it. Cross-validators gate `task='grpo'` + `backend='transformers'` +
+  `modality='text'`. Default aggregation is `min` (weakest-link); `prod`
+  assumes calibrated `[0,1]` step scores.
+- **Bundled rollout environments** — three pure-Python toy environments
+  (`soup_cli.envs.calculator` / `retrieval_qa` / `guess_number`) exposing a
+  `rollout(prompts)` entry point so the live `openenv` GRPO rollout path runs
+  out-of-the-box: `training.rollout_backend=openenv` +
+  `training.rollout_func=soup_cli.envs.calculator:rollout`. Three ready-made
+  recipes added (`grpo-env-calculator` / `grpo-env-retrieval-qa` /
+  `grpo-env-guess-number`); catalog 134 → 137.
+
+### Fixed
+- **`soup train task=prm` producer conformance** (surfaced by the v0.71.30 live
+  smoke): the PRM trainer now casts its reward head to the base-model dtype
+  (bf16 CUDA runs previously crashed on the first `compute_loss`), saves the
+  tokenizer alongside the model (so a PRM checkpoint is loadable standalone),
+  and returns the standard trainer-result shape (previously the CLI crashed with
+  a `KeyError: 'initial_loss'` right after saving).
+
+### Notes
+- Proof-of-mechanism only: validated on a tiny model (SmolLM2-135M) with a tiny
+  synthetic PRM and synthetic reward — not a production reward-model claim
+  (scale ask tracked in #286). The bundled environments are deterministic
+  single-shot *seeders*, not interactive multi-turn model-in-the-loop episodes
+  (the live `openenv` contract passes only prompts). Step split is a newline
+  heuristic; PRM completions are scored one forward pass each.
+
 ## [0.71.29] - 2026-07-05
 
 ### Added

{soup_cli-0.71.29 → soup_cli-0.71.31}/CONTRIBUTING.md

@@ -110,17 +110,17 @@ src/soup_cli/
   experiment/         - SQLite experiment tracking
   eval/               - Eval platform (custom tasks, LLM judge, human eval, leaderboard)
   migrate/            - Config migration (LLaMA-Factory, Axolotl, Unsloth)
-  recipes/            - Ready-made configs for popular models (134 recipes)
+  recipes/            - Ready-made configs for popular models (138 recipes)
   autopilot/          - Zero-config decision engine (v0.25.0)
   registry/           - Model Registry (hashing, store, diff, attach) (v0.26.0 + v0.33.0)
   cans/               - Shareable .can artifact format + run/publish orchestrator (v0.26.0 + v0.33.0)
   data/traces/        - Trace-to-Preference harvester (v0.26.0)
   data/collators.py   - CrossDocCollator for sample packing (v0.33.0)
-  utils/              - GPU, errors, MoE, GaLore, QAT, Unsloth, vLLM, SGLang, Liger, FlashAttn, FSDP, Ring Attention, long-context, quality, curriculum, freeze, dataset-registry, mlx, peft_builder, paths, topology, launcher, mii, pipeline, cut_ce, fp8, gradient_ckpt, kernel_picker, cross_doc_attn, activation_offload, hf, spec_pairing, structured_output, metrics, tracing, auto_quant, lr_finder, grad_accum, mixed_precision, warmup, spike_recovery, convergence, v028_features, multipack_sampler, multipack, neat_packing, jinja_analyzer, quant_menu, relora, peft_patches, peft_wiring, dpo_variants, optimizer_zoo, lr_groups, loftq_init, block_expansion, tts, classifier, distill, bitnet, ebft_gdpo, moe_quant, reasoning_effort, gguf_quant, kv_cache, advanced_precision, save_formats, deploy_measure, advise, advise_history, adapter_diff, adapter_merge, blame, adapter_branch, unlearning, unlearning_eval, knowledge_edit, edit_governor, edit_diff, ra_dit, steering, citation_faithful, grace_codebook, ingest_sources, prune_prompt, active_sampler, ab_test, drift_alarm, tunability, terraform_plan, env_lock, hardware_fit, completions, license_advisor, behavior_battery, capability_suite, checklist_dsl, irt, sae_diff, sleeper_probe, interference, probe_pack, cmaes_merge, vector_bank, mole_routing, adapter_pr, soup_lock, adapter_bisect, prompt_compile, prompt_distill, compile_tools, apple_adapter, local_rl, build_dag, expectations, magpie, persona_hub, brain_rot, reward_hacking, uld, minillm, rl_checkpoint, iterative_dpo, echo_trap, mod, local_rl_scheduler, spectrum_scan, ship_verdict, reward_hack_control, data_doctor, data_lint
+  utils/              - GPU, errors, MoE, GaLore, QAT, Unsloth, vLLM, SGLang, Liger, FlashAttn, FSDP, Ring Attention, long-context, quality, curriculum, freeze, dataset-registry, mlx, peft_builder, paths, topology, launcher, mii, pipeline, cut_ce, fp8, gradient_ckpt, kernel_picker, cross_doc_attn, activation_offload, hf, spec_pairing, structured_output, metrics, tracing, auto_quant, lr_finder, grad_accum, mixed_precision, warmup, spike_recovery, convergence, v028_features, multipack_sampler, multipack, neat_packing, jinja_analyzer, quant_menu, relora, peft_patches, peft_wiring, dpo_variants, optimizer_zoo, lr_groups, loftq_init, block_expansion, tts, classifier, distill, bitnet, ebft_gdpo, moe_quant, reasoning_effort, gguf_quant, kv_cache, advanced_precision, save_formats, deploy_measure, advise, advise_history, adapter_diff, adapter_merge, blame, adapter_branch, unlearning, unlearning_eval, knowledge_edit, edit_governor, edit_diff, ra_dit, steering, citation_faithful, grace_codebook, ingest_sources, prune_prompt, active_sampler, ab_test, drift_alarm, tunability, terraform_plan, env_lock, hardware_fit, completions, license_advisor, behavior_battery, capability_suite, checklist_dsl, irt, sae_diff, sleeper_probe, interference, probe_pack, cmaes_merge, vector_bank, mole_routing, adapter_pr, soup_lock, adapter_bisect, prompt_compile, prompt_distill, compile_tools, apple_adapter, local_rl, build_dag, expectations, magpie, persona_hub, brain_rot, reward_hacking, uld, minillm, rl_checkpoint, iterative_dpo, echo_trap, mod, local_rl_scheduler, spectrum_scan, ship_verdict, reward_hack_control, data_doctor, data_lint, shrink, best_of_n, evolve
   templates/          - 17 built-in soup.yaml templates (YAML + manifest.json) with load_template loader (v0.39.0, +bco v0.40.0)
   ui/                 - Web UI (FastAPI + HTML/JS SPA)
 
-tests/                - Test suite (306 files, 15224 tests)
+tests/                - Test suite (308 files, 15421 tests)
 examples/             - Real-world config examples and datasets
 ```
 
@@ -150,10 +150,9 @@ pytest tests/test_data.py::test_detect_alpaca_format -v
 pytest tests/ --cov=soup_cli --cov-report=html
 ```
 
-### Test Files (306 files)
+### Test Files (307 files)
 
-> A representative sample of the suite below. The full table lives in
-> [`.claude/CLAUDE.md`](.claude/CLAUDE.md); run `pytest tests/ -v` for the complete list.
+> A representative sample of the suite below. Run `pytest tests/ -v` for the complete list.
 
 
 | File | Covers |

{soup_cli-0.71.29 → soup_cli-0.71.31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: soup-cli
-Version: 0.71.29
+Version: 0.71.31
 Summary: Fine-tune and post-train LLMs in one command. No SSH, no config hell.
 Project-URL: Homepage, https://github.com/MakazhanAlpamys/Soup
 Project-URL: Repository, https://github.com/MakazhanAlpamys/Soup
@@ -193,22 +193,29 @@ infrastructure instead of improving models. Soup fixes that.
 
 ## What's New
 
-**v0.71.29 — Make your model smaller, locally: `soup shrink`.** Depth-prune the least-important layers, then distill-heal — with a binary SHIP / DON'T-SHIP verdict. Based on "The Unreasonable Ineffectiveness of the Deeper Layers" (arXiv:2403.17887).
+**v0.71.31 — Judge-in-the-loop suite.** Put an LLM judge in the loop across the whole workflow: train against a judge, mine winners from a base model, grow instruction diversity, and decide SHIP with a true pairwise judge win-rate.
+
+- **`task='online_dpo'`.** Online DPO (wraps TRL `OnlineDPOTrainer`): the model generates two
+  completions per prompt on-policy each step and a *judge* (a pairwise LLM judge over your local
+  ollama / OpenAI-compatible endpoint) — or a `reward_model` — picks the winner. Set
+  `training.online_dpo_judge: "ollama://llama3.1"` (or `reward_model`, exactly one).
+- **`soup data best-of-n`.** Best-of-N rejection sampling: sample N completions from `--base`
+  locally, a `--judge` scores each, and the winner becomes an SFT row (`--emit-pairs` also writes
+  winner-vs-loser DPO pairs).
+- **`soup data evolve`.** WizardLM Evol-Instruct (depth / breadth) over an ollama / vllm provider —
+  completing the synthetic-data suite (Magpie / Forge / Persona / evolve).
+- **`soup ship --task-mode pairwise` (#284).** A true swap-debiased judge win-rate as the ship
+  leg-1 task-win, fused with the catastrophic-forgetting guard into one SHIP / DON'T-SHIP verdict.
+- **Proof-of-mechanism, honestly.** Live-validated on SmolLM2-135M: online-DPO trains with a
+  synthetic judge (`rewards/*` + `objective/kl` logged), best-of-N samples + picks, evolve loops.
+  Not a production RLHF claim (scale help wanted: #286).
 
-- **Importance-ranked depth pruning.** One forward pass per calibration prompt scores each
-  contiguous layer block by the angular distance of the residual stream across it; the
-  least-important block is dropped (the first and last layer are always protected).
-- **Distill-heal, Minitron-style.** `--heal` distills the original model into the pruned
-  student (LoRA logit-KD) as an isolated `soup train` run, then fuses the adapter back so you
-  ship a single dense smaller model — not a base + adapter.
-- **A verdict, not a dashboard.** Every run ends in **SHIP** or **DON'T SHIP** on a
-  before/after perplexity ratio (`exit 0 = SHIP`, `2 = DON'T`, `1 = error`), so it drops into CI.
-- **Runs on a laptop GPU.** Live-validated on Windows + RTX 3050 with SmolLM2-135M: drop 25 %
-  (30 → 22 layers, 21 % params) and drop-4 + CPU heal (perplexity recovered to ×1.35).
-
-```bash
-soup shrink --model HuggingFaceTB/SmolLM2-135M-Instruct --drop-ratio 0.25 \
-    --calib calib.jsonl --heal heal.jsonl --heal-steps 200 -o shrunk --device cpu
+```yaml
+task: online_dpo
+training:
+  online_dpo_judge: "ollama://llama3.1"   # a pairwise judge (or set reward_model)
+  online_dpo_loss_type: sigmoid           # sigmoid | ipo
+  online_dpo_max_new_tokens: 64
 ```
 
 Full history: [CHANGELOG.md](CHANGELOG.md) · [GitHub Releases](https://github.com/MakazhanAlpamys/Soup/releases).

{soup_cli-0.71.29 → soup_cli-0.71.31}/README.md

@@ -49,22 +49,29 @@ infrastructure instead of improving models. Soup fixes that.
 
 ## What's New
 
-**v0.71.29 — Make your model smaller, locally: `soup shrink`.** Depth-prune the least-important layers, then distill-heal — with a binary SHIP / DON'T-SHIP verdict. Based on "The Unreasonable Ineffectiveness of the Deeper Layers" (arXiv:2403.17887).
-
-- **Importance-ranked depth pruning.** One forward pass per calibration prompt scores each
-  contiguous layer block by the angular distance of the residual stream across it; the
-  least-important block is dropped (the first and last layer are always protected).
-- **Distill-heal, Minitron-style.** `--heal` distills the original model into the pruned
-  student (LoRA logit-KD) as an isolated `soup train` run, then fuses the adapter back so you
-  ship a single dense smaller model — not a base + adapter.
-- **A verdict, not a dashboard.** Every run ends in **SHIP** or **DON'T SHIP** on a
-  before/after perplexity ratio (`exit 0 = SHIP`, `2 = DON'T`, `1 = error`), so it drops into CI.
-- **Runs on a laptop GPU.** Live-validated on Windows + RTX 3050 with SmolLM2-135M: drop 25 %
-  (30 → 22 layers, 21 % params) and drop-4 + CPU heal (perplexity recovered to ×1.35).
+**v0.71.31 — Judge-in-the-loop suite.** Put an LLM judge in the loop across the whole workflow: train against a judge, mine winners from a base model, grow instruction diversity, and decide SHIP with a true pairwise judge win-rate.
+
+- **`task='online_dpo'`.** Online DPO (wraps TRL `OnlineDPOTrainer`): the model generates two
+  completions per prompt on-policy each step and a *judge* (a pairwise LLM judge over your local
+  ollama / OpenAI-compatible endpoint) — or a `reward_model` — picks the winner. Set
+  `training.online_dpo_judge: "ollama://llama3.1"` (or `reward_model`, exactly one).
+- **`soup data best-of-n`.** Best-of-N rejection sampling: sample N completions from `--base`
+  locally, a `--judge` scores each, and the winner becomes an SFT row (`--emit-pairs` also writes
+  winner-vs-loser DPO pairs).
+- **`soup data evolve`.** WizardLM Evol-Instruct (depth / breadth) over an ollama / vllm provider —
+  completing the synthetic-data suite (Magpie / Forge / Persona / evolve).
+- **`soup ship --task-mode pairwise` (#284).** A true swap-debiased judge win-rate as the ship
+  leg-1 task-win, fused with the catastrophic-forgetting guard into one SHIP / DON'T-SHIP verdict.
+- **Proof-of-mechanism, honestly.** Live-validated on SmolLM2-135M: online-DPO trains with a
+  synthetic judge (`rewards/*` + `objective/kl` logged), best-of-N samples + picks, evolve loops.
+  Not a production RLHF claim (scale help wanted: #286).
 
-```bash
-soup shrink --model HuggingFaceTB/SmolLM2-135M-Instruct --drop-ratio 0.25 \
-    --calib calib.jsonl --heal heal.jsonl --heal-steps 200 -o shrunk --device cpu
+```yaml
+task: online_dpo
+training:
+  online_dpo_judge: "ollama://llama3.1"   # a pairwise judge (or set reward_model)
+  online_dpo_loss_type: sigmoid           # sigmoid | ipo
+  online_dpo_max_new_tokens: 64
 ```
 
 Full history: [CHANGELOG.md](CHANGELOG.md) · [GitHub Releases](https://github.com/MakazhanAlpamys/Soup/releases).

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/README.md

@@ -8,7 +8,7 @@ feature reference — every `soup` capability, grouped by area.
 | Guide | Covers |
 |---|---|
 | [Training tasks & methods](training.md) | SFT, DPO/GRPO/PPO/KTO/ORPO/SimPO/IPO/BCO, tool-calling, PRM, pre-training, distillation, classification, vision/audio/TTS, unlearning, RAFT/RA-DIT, loop-hardening detectors |
-| [PEFT, long context & efficiency](peft-and-efficiency.md) | DoRA, LoRA+, rsLoRA, VeRA, OLoRA, NEFTune, PiSSA, ReLoRA, optimizer & PEFT zoo, LLaMA Pro, GaLore, YaRN/LongLoRA, packing, curriculum, auto-tuning |
+| [PEFT, long context & efficiency](peft-and-efficiency.md) | DoRA, LoRA+, rsLoRA, VeRA, OLoRA, NEFTune, PiSSA, ReLoRA, optimizer & PEFT zoo, LLaMA Pro, GaLore, YaRN/LongLoRA, packing, curriculum, auto-tuning, depth pruning + distill-heal (`soup shrink`) |
 | [Performance & quantization](performance-and-quantization.md) | QAT, FP8, Quant Menu (I + II), KV-cache, NVFP4, save formats, Cut Cross-Entropy, gradient checkpointing, kernels, activation offloading, multi-GPU / DeepSpeed / FSDP |
 | [Data engineering](data.md) | Formats, the Axolotl/LF-parity pipeline, data tools, synthetic generation & forge, quality scorecards, trace tooling, remote datasets, mixing, recipe DAGs |
 | [Evaluation & probes](evaluation.md) | Eval design/gate, eval-gated training, benchmarks, NLG metrics, calibration, Elo arena, diagnose, `soup ship` verdict, post-train X-ray probes, A/B, drift, tunability, `soup advise` |

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/commands.md

@@ -133,7 +133,7 @@ soup migrate --from llamafactory config.yaml  Import config from LLaMA-Factory
 soup migrate --from axolotl config.yml        Import config from Axolotl
 soup migrate --from unsloth notebook.ipynb    Import config from Unsloth notebook
 soup migrate --from llamafactory c.yaml --dry-run  Preview without writing
-soup recipes list                             List all 134 ready-made recipes
+soup recipes list                             List all 138 ready-made recipes
 soup recipes show llama3.1-8b-sft            Print recipe YAML
 soup recipes use llama3.1-8b-sft             Copy recipe to soup.yaml
 soup recipes search "reasoning"              Search by keyword/task/size
@@ -155,6 +155,7 @@ soup why [run_id]                             Explain training anomalies (heuris
 soup ship --base <m> --adapter <lora> --task-eval t.jsonl  SHIP / DON'T-SHIP verdict: task win AND no forgetting (exit 0=SHIP / 2=DON'T / 1=error) (v0.71.25)
 soup ship --evidence ev.json [--output v.json]  Decide offline from pre-computed scores (no model load)
 soup ship ... --task-mode judge_score --judge-model ollama://llama3.1  Leg-1 via LLM-as-a-judge
+soup ship ... --task-mode pairwise --judge-model ollama://llama3.1  Leg-1 via swap-debiased judge win-rate (base=0.5) (v0.71.31)
 soup ship ... --general-suite mmlu,gsm8k --baseline base.json  lm-eval leg-2 + recorded base scores
 soup mcp serve                                MCP server over stdio (drive Soup from Claude Code / Cursor / Cline; requires [mcp] extra) (v0.71.28)
 soup mcp serve --allow-mutating               Also expose plan-only train_start / export tools (never execute) (v0.71.28)
@@ -250,6 +251,8 @@ soup local-rl train --db <path> --model <id> [--scheduler-dir <dir>] [--hour H]
 soup build <manifest.yaml> [--dry-run] [--output-dir <dir>]  dbt-for-SFT DAG: validate + plan + live materialise (v0.69.0; live v0.71.6)
 soup expect <data.jsonl> <suite.yaml>         Expectations suite: PII / token-length / refusal / judge (v0.69.0)
 soup data gen-magpie --base <m> --provider ollama|vllm --target N --output <jsonl> [--base-url <url>] [--quality-filter]  Magpie synthetic generator — live (v0.69.0; live v0.71.6)
+soup data best-of-n --base <m> --prompts <jsonl> --n 8 --judge <url> -o <sft.jsonl> [--emit-pairs <dpo.jsonl>]  Best-of-N rejection sampling: sample N locally, judge picks winner -> SFT (+ DPO) rows (v0.71.31)
+soup data evolve --input <seeds.jsonl> --provider ollama|vllm --model <m> --strategy depth|breadth --rounds N -o <jsonl>  Evol-Instruct (WizardLM) instruction evolution (v0.71.31)
 soup data persona-mix --prompts <jsonl> --n N --output <jsonl>  Persona-Hub diversity sampler (v0.69.0)
 soup data brain-rot <data.jsonl> [--strict]   Brain-rot detector — arXiv 2510.13928 (v0.69.0)
 soup iterative-dpo --base-model <m> --reward-model <rm> --prompts <p.jsonl> --output-dir <out> --rounds N --pairs-per-round N [--plan-only]  Iterative DPO loop driver — LIVE sample→score→pair→train (v0.70.0; live v0.71.11)

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/data.md

@@ -62,6 +62,15 @@ soup data persona-mix --prompts prompts.jsonl --n 500 --output mixed.jsonl
 
 # Brain-rot detector (arXiv 2510.13928) — refuses to train on excessive slop
 soup data brain-rot data.jsonl --strict --max-major-fraction 0.10
+
+# Best-of-N rejection sampling (v0.71.31) — sample N locally, a judge picks the winner
+soup data best-of-n --base HuggingFaceTB/SmolLM2-135M-Instruct \
+    --prompts prompts.jsonl --n 8 --judge ollama://llama3.1 \
+    -o best_of_n.jsonl --emit-pairs pairs.jsonl
+
+# Evol-Instruct (WizardLM depth/breadth, v0.71.31) — grow instruction diversity
+soup data evolve --input seeds.jsonl --provider ollama --model llama3.1 \
+    --strategy depth --rounds 2 -o evolved.jsonl
 ```
 
 Every command applies the project-wide TOCTOU policy (`os.lstat + S_ISLNK` symlink rejection before any open) and cwd containment via the shared `paths.enforce_under_cwd_and_no_symlink` helper. All five are LIVE: `soup build` materialises with five built-in transforms (`identity` / `drop_empty` / `lowercase` / `strip` / `dedup_exact`) and SQLite-tracked incremental re-transform (v0.71.6); `soup data gen-magpie` harvests via raw completion against `--provider ollama|vllm` (loopback-only; `anthropic` rejected — no raw-completion endpoint, v0.71.6).

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/evaluation.md

@@ -330,6 +330,11 @@ soup ship --base HuggingFaceTB/SmolLM2-135M-Instruct --adapter ./out \
 soup ship --base <m> --adapter ./out --task-eval tasks.jsonl \
   --task-mode judge_score --judge-model ollama://llama3.1
 
+# Leg-1 via a true pairwise judge win-rate (v0.71.31 #284) — the judge picks
+# base vs tuned per prompt (swap-debiased); base = 0.5 coin-flip, won <=> winrate > 0.5
+soup ship --base <m> --adapter ./out --task-eval tasks.jsonl \
+  --task-mode pairwise --judge-model ollama://llama3.1
+
 # Leg-2 via lm-eval benchmarks, base scores supplied by --baseline
 soup ship --base <m> --tuned ./out --task-eval tasks.jsonl \
   --general-suite mmlu,hellaswag --baseline registry://abc123

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/serving-and-export.md

@@ -418,7 +418,7 @@ soup ui
 
 **Pages:**
 - **Dashboard** — view all experiment runs, loss charts, system info, multi-run comparison
-- **New Training** — create configs from templates or 134 ready-made recipes, validate, start training with live SSE log streaming and progress bar
+- **New Training** — create configs from templates or 138 ready-made recipes, validate, start training with live SSE log streaming and progress bar
 - **Data Explorer** — browse and inspect datasets (JSONL, JSON, CSV, Parquet)
 - **Model Chat** — chat with streaming responses, configurable temperature/top_p/max_tokens, system prompt, adapter selection, markdown rendering, chat export
 
@@ -427,7 +427,7 @@ soup ui
 - **Enhanced Metrics** — 2x2 chart grid (loss, LR, grad_norm, throughput) + GPU memory chart, eval results table
 - **Multi-Run Compare** — overlay loss curves from up to 5 runs side-by-side
 - **Chat Upgrade** — SSE streaming via proxy, typing indicator, cancel button, markdown renderer (bold, italic, code blocks), chat export as JSON
-- **Config Builder** — recipe dropdown (134 recipes), config schema API for dynamic form generation
+- **Config Builder** — recipe dropdown (138 recipes), config schema API for dynamic form generation
 
 **Security:** The Web UI generates a random auth token at startup (printed to console). All mutating endpoints (start/stop training, delete runs, inspect data, validate config) require `Authorization: Bearer <token>` header. CORS is restricted to the served origin. Data inspection is sandboxed to the working directory.
 

{soup_cli-0.71.29 → soup_cli-0.71.31}/docs/training.md

@@ -15,6 +15,7 @@
 - [EBFT / GDPO Loss Variants](#ebft--gdpo-loss-variants)
 - [GRPO Objective Variants](#grpo-objective-variants)
 - [Process Reward Model (PRM)](#process-reward-model-prm)
+- [PRM-guided GRPO (process-supervised RL)](#prm-guided-grpo-process-supervised-rl)
 - [Weighted Multi-Objective Preference Loss](#weighted-multi-objective-preference-loss)
 - [MoE Model Support](#moe-model-support)
 - [Vision / Multimodal Fine-tuning](#vision--multimodal-fine-tuning)
@@ -343,7 +344,89 @@ training:
 
 The trainer loads `AutoModelForCausalLM`, attaches an `nn.Linear(hidden, 1)`
 reward head, and computes MSE between predicted scalars at step-boundary tokens
-and the per-step labels.
+and the per-step labels. The reward head is saved inside the model checkpoint
+(`reward_head.*` in `model.safetensors`) and the tokenizer is saved alongside it,
+so the resulting directory is loadable standalone.
+
+
+## PRM-guided GRPO (process-supervised RL)
+
+Use a trained PRM as the **per-step reward** inside GRPO — the o1-era
+process-supervision signal. Set `training.prm_reward` to a PRM directory (a
+`task=prm` checkpoint) or HF id; the PRM splits each generated completion into
+reasoning steps (newline heuristic), scores every step with its reward head, and
+folds the per-step scores into one scalar reward that GRPO optimises. It
+**replaces** `reward_fn` and rides the existing reward-shaping +
+reward-hack-mitigation seam, so the v0.71.26 controller still observes it (TRL
+logs it as `rewards/prm_reward`).
+
+```yaml
+task: grpo
+backend: transformers        # required (the PRM reward runs a transformers forward)
+modality: text               # required
+data:
+  format: chatml
+  train: ./grpo_prompts.jsonl
+training:
+  prm_reward: ./my-prm       # a `soup train task=prm` checkpoint dir (or HF id)
+  prm_aggregate: min         # weakest-link (default) | prod | last
+  num_generations: 4
+  grpo_beta: 0.04
+```
+
+`prm_aggregate='min'` (weakest-link, the standard PRM aggregation) is the safe
+default; `prod` assumes calibrated `[0,1]` step scores (Soup's PRM head is
+trained with unconstrained MSE, so `prod` can blow up on uncalibrated labels).
+
+**Bundled rollout environments.** Three deterministic pure-Python toy
+environments seed the openenv rollout path out-of-the-box — pair any of them
+with `rollout_backend=openenv`:
+
+```yaml
+training:
+  rollout_backend: openenv
+  rollout_func: soup_cli.envs.calculator:rollout   # or retrieval_qa / guess_number
+  reward_fn: verifiable
+  verifiable_domain: math
+```
+
+Ready-made recipes: `grpo-env-calculator`, `grpo-env-retrieval-qa`,
+`grpo-env-guess-number`. The environments are deterministic single-shot
+prompt/answer *seeders* (the live openenv contract passes only the seed prompts,
+not the model) — not interactive multi-turn episodes.
+
+**Scope:** proof-of-mechanism only — validated on SmolLM2-135M with a tiny
+synthetic PRM (the PRM reward scores good completions above bad and drives GRPO's
+advantages). Not a production reward-model claim; scale validation is help-wanted
+(#286).
+
+
+## Online DPO (`task='online_dpo'`) — judge in the loop (v0.71.31)
+
+Unlike offline DPO (static `prompt/chosen/rejected` rows), **Online DPO** generates
+two completions per prompt **on-policy** each step and asks a *judge* — or a
+*reward model* — which is better; the winner becomes `chosen`, the loser
+`rejected`. The judge closes the loop. Wraps TRL `OnlineDPOTrainer`; data is
+prompt-only (like GRPO). Transformers + text only.
+
+```yaml
+base: HuggingFaceTB/SmolLM2-135M-Instruct
+task: online_dpo
+data:
+  train: ./data/prompts.jsonl      # prompt-only (or any format — prompts are extracted)
+training:
+  online_dpo_judge: "ollama://llama3.1"   # a pairwise judge (ollama://|https://|http://localhost)
+  # OR: reward_model: ./my-reward-model   # exactly one of judge / reward_model
+  online_dpo_loss_type: sigmoid           # sigmoid | ipo
+  online_dpo_max_new_tokens: 64
+  dpo_beta: 0.1
+  lora: { r: 8, alpha: 16, target_modules: auto }
+```
+
+The judge is Soup's own OpenAI-compatible `JudgeEvaluator` adapted to TRL's
+`BasePairwiseJudge` (swap-debiased: a winner is only recorded when both A,B and
+B,A orders agree). Recipe: `online-dpo-smollm2-135m`. Proof-of-mechanism was
+validated on SmolLM2-135M with a synthetic judge (not a production RLHF claim; #286).
 
 
 ## Weighted Multi-Objective Preference Loss

{soup_cli-0.71.29 → soup_cli-0.71.31}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "soup-cli"
-version = "0.71.29"
+version = "0.71.31"
 description = "Fine-tune and post-train LLMs in one command. No SSH, no config hell."
 readme = "README.md"
 license = "Apache-2.0"

{soup_cli-0.71.29 → soup_cli-0.71.31}/src/soup_cli/__init__.py

@@ -1,3 +1,3 @@
 """Soup CLI — Fine-tune and post-train LLMs in one command."""
 
-__version__ = "0.71.29"
+__version__ = "0.71.31"