stateset-agents 0.18.0

Production-ready RL framework for training multi-turn conversational AI agents using GRPO and GSPO

StateSet Agents

Reinforcement‑learning framework for multi‑turn conversational AI agents.

StateSet Agents is a production‑oriented RL stack for training and serving LLM‑backed agents that improve through multi‑turn interaction. The library provides:

• Async‑first agent APIs (MultiTurnAgent, ToolAgent) with Hugging Face and stub backends.
• Environments for conversational and task‑oriented episodes.
• Trajectories and value/advantage utilities tailored to dialogue.
• Composable reward functions (heuristic, domain, multi‑objective, neural).
• A family of group‑based policy‑optimization trainers (GRPO, GSPO, GEPO, DAPO, VAPO) plus PPO and RLAIF.
• Offline RL algorithms for learning from logged conversations (BCQ, BEAR, CQL, IQL, Decision Transformer).
• Sim‑to‑Real transfer for training in simulation and deploying to real users (domain randomization, system identification, progressive transfer).
• Continual learning + long‑term planning utilities (replay/LwF/EWC, plan context injection).
• Optional performance layers (vLLM generation, Rust acceleration, distributed training, HPO, FastAPI service).

If you want a framework that treats conversations as first‑class RL episodes (rather than single turns), this is it.

---

What's new

In v0.18.0 (latest release — live on PyPI):

• Bring your own agent's logs. stateset-agents ingest converts OpenAI chat-format and LangChain conversation dumps into framework trajectories — production logs from agents built anywhere plug straight into the grade → curate → retrain loop.
• The improvement loop in one command. stateset-agents improve run --transcripts DIR --reward NAME --output DIR grades every transcript, curates the best turns, and emits verified-runnable next-step training commands (a regression test executes the suggestions against the real CLI parsers).
• Flagship benchmark recipe. make flagship-benchmark-all — a reproducible 3-seed GSPO run on an 8B model with publish gates (benchmarks/FLAGSHIP.md).
• Nightly training guard operational. The nightly benchmark + CPU convergence pipeline is green end-to-end for the first time.

In v0.17.3:

• PyPI is current again. pip install stateset-agents installs the latest release (the 0.13.4-era gap is closed), wheels finally ship the runtime config presets, and stateset-rl-core 0.1.1 is on PyPI — pip install stateset-agents[rust] and [full] resolve for the first time.
• First fully green CI since v0.15.2 (10/10 jobs including Windows): pinned toolchain (mypy, numpy, pre-commit hooks), utf-8 file IO everywhere, Windows clock/codepage fixes, repaired strict security scan.

In v0.17.2:

• Packaging repair. Tag-triggered publishing with OIDC trusted-publishing support, Rust CI for both crates, dashboard/mobile env-based API config + API-key auth + Node pins + mobile CI.

In v0.17.1:

• Convergence proof in CI. tests/e2e/test_gspo_convergence_tiny.py trains real GSPO on a tiny model and asserts the target-token probability provably increases (verified against zero-signal and reversed-reward controls); runs in the nightly benchmark workflow.
• Honest demo labeling. dashboard/ and mobile/ are clearly marked as simulator-backed demos (not deployed); the mobile data hook now surfaces an isMockData flag instead of silently falling back.

In v0.17.0:

• Unified finetune driver. examples/finetune_gspo.py --model <preset> replaces the per-model script maze: a 12-model preset registry (examples/model_presets.py), shared flags (--use-lora, --use-4bit/--use-8bit, --use-vllm, --wandb, --export-merged, --starter-profile, --write-config), safe --dry-run default, and a real training path via --no-dry-run.
• Rate-limiter hardening. Bucket keys derive only from validated credentials (garbage keys share the IP bucket), the in-memory limiter is memory-capped, and the optional Redis backend retries after a cooldown with an atomic INCR+EXPIRE pipeline.
• grpo untangling. Shared rate_limiter moved to stateset_agents/api/rate_limiter.py; the api.grpo deprecation warning now fires only for the deprecated app surface.

In v0.16.0:

• RL-core correctness overhaul. All five trainers fixed and behaviorally tested: DAPO freezes rollout-time old log probs and honors µ inner updates; GEPO runs in log space; GSPO scores exactly the text it sampled (shared chat-template convention) and rescores vLLM rollouts; GSPO-token regained its gradient path; VAPO clips values against rollout predictions with terminal-token rewards and decoupled GAE; the GRPO loss path uses length-normalized ratios. A cross-trainer ratio-invariant suite guards all of it, and VAPO uses the Rust GAE kernel when installed.
• API hardening. The training-lab API is auth-gated behind API_ENABLE_TRAINING_LAB (off outside development) with bounded in-memory state; startup config validation fails closed in production; constant-time API-key comparison everywhere.
• Kimi-K3 and GLM 5.2 starter paths. First-class starters for moonshotai/Kimi-K3 (provisional presets until HF weights land) and zai-org/GLM-5.2 (754B MoE) — CLI commands, packaged modules, Helm/Kubernetes manifests, docs, and tests.

Earlier highlights: v0.15.3 shipped Rust accelerator wheels (abi3-py310) and model-level Prometheus inference metrics; v0.15.0 added the five-example getting-started path; v0.13.2 shipped the whitepaper §11.7 three-seed canonical benchmark (judge improvement +0.079, artifact).

Full breakdown in CHANGELOG.md.

---

Why group‑based optimization?

Traditional RLHF/PPO trains on one sampled response at a time. In long conversations this leads to high‑variance updates and brittle behavior.
StateSet Agents implements group‑relative methods:

• GRPO (Group Relative Policy Optimization): sample a group of trajectories per prompt, compute advantages relative to the group baseline, then apply clipped policy‑gradient updates.
• GSPO (Group Sequence Policy Optimization): a more stable sequence‑level variant (Alibaba Qwen team) that avoids token‑level collapse on long outputs and MoE models.

The result is steadier learning for dialogue tasks.

---

Core concepts

• Agent: wraps a causal LM and exposes initialize() and generate_response().
  • MultiTurnAgent handles conversation history and state.
  • ToolAgent adds function/tool calling.
• Environment: defines episode reset/step logic and optional reward hooks.
  • ConversationEnvironment ships with scenario‑driven multi‑turn conversations.
  • TaskEnvironment is for goal‑oriented tasks.
• Trajectory: a multi‑turn record of turns, rewards, and metadata (MultiTurnTrajectory).
• Rewards: RewardFunction subclasses and factories; combined via CompositeReward or multi‑objective reward models.
• Training: trainers in stateset_agents.training implement GRPO‑family updates, GAE/value heads, KL regularization, LoRA support, and optional distributed/vLLM execution.

---

Reward semantics

Reward functions can be evaluated per-step or only at episode end. Set reward_type on your RewardFunction to control how the environment applies it:

• RewardType.IMMEDIATE or RewardType.DENSE: compute per-step rewards only.
• RewardType.CUMULATIVE or RewardType.SPARSE: compute a final reward only.

If you pass a custom reward without reward_type, the environment assumes legacy behavior and may compute both step and final rewards. For new rewards, always set reward_type explicitly to avoid double counting.

---

Tool calling (ToolAgent)

ToolAgent lets a model request a tool via a JSON block, which the agent executes:

import asyncio
from stateset_agents.core.agent import AgentConfig, ToolAgent

def add(a: int, b: int) -> int:
    return a + b

async def main():
    agent = ToolAgent(
        AgentConfig(model_name="stub://tools", use_stub_model=True),
        tools=[
            {
                "name": "add",
                "description": "Add two integers",
                "parameters": {"a": "int", "b": "int"},
                "function": add,
            }
        ],
    )
    await agent.initialize()
    # The model should respond with a JSON tool call like:
    # {"tool": "add", "parameters": {"a": 1, "b": 2}}
    print(await agent.generate_response("Please calculate 1 + 2"))

asyncio.run(main())

---

Installation

Core (lightweight, stub‑ready)

pip install stateset-agents          # latest release (v0.18.0)

PyPI tracks the release tags. For unreleased work on master:

pip install "git+https://github.com/stateset/stateset-agents@master"

Training / real models

pip install "stateset-agents[training]"

Optional extras

pip install "stateset-agents[auto-research]" # Autonomous experiment loop + Optuna
pip install "stateset-agents[trl]"           # TRL GRPO integration + bitsandbytes
pip install "stateset-agents[vllm]"          # vLLM generation backend
pip install "stateset-agents[hpo]"           # Optuna/Ray Tune HPO
pip install "stateset-agents[api]"           # FastAPI service
pip install "stateset-agents[distributed]"   # DeepSpeed / multi‑GPU helpers
pip install "stateset-agents[rust]"          # Rust-accelerated GAE/advantage kernels (stateset-rl-core)
pip install "stateset-agents[full]"          # Most extras in one go

This repository also contains an internal, unpublished Rust crate at the repo root (a StateSet commerce daemon) that is unrelated to the stateset-rl-core accelerator behind [rust] above. See docs/RUST_CRATES.md for how the two Rust crates in this repo relate.

Qwen 3.5 starter path

If you want the fastest path to a first post-training run for Qwen/Qwen3.5-0.8B, use the dedicated CLI starter or the equivalent example script:

pip install "stateset-agents[training,trl]"
stateset-agents qwen3-5-0-8b --json-output
stateset-agents qwen3-5-0-8b --starter-profile memory --json-output
stateset-agents qwen3-5-0-8b --list-profiles --json-output
stateset-agents qwen3-5-0-8b --write-config ./qwen3_5_0_8b.json
stateset-agents qwen3-5-0-8b --config ./qwen3_5_0_8b.json --no-dry-run
python examples/finetune_qwen3_5_0_8b_gspo.py --dry-run

Use --list-profiles when you want to compare the built-in balanced, memory, and quality presets before saving or running one.

For the repo-specific walkthrough, see docs/QWEN3_FINETUNING_GUIDE.md.

Kimi-K2.6 starter path

If you want the fastest path to a first post-training run for moonshotai/Kimi-K2.6, use the dedicated CLI starter or the equivalent example script:

pip install "stateset-agents[training,trl]"
stateset-agents kimi-k2-6 --json-output
stateset-agents kimi-k2-6 --starter-profile memory --json-output
stateset-agents kimi-k2-6 --list-profiles --json-output
stateset-agents kimi-k2-6 --write-config ./kimi_k2_6.json
stateset-agents kimi-k2-6 --config ./kimi_k2_6.json --no-dry-run
python examples/finetune_kimi_k2_6_gspo.py --dry-run

Use --list-profiles when you want to compare the built-in balanced, memory, and quality presets before saving or running one.

Kimi-K3 starter path

The same starter flow ships for moonshotai/Kimi-K3. Note: Kimi K3 weights are not yet published on HuggingFace (as of 2026-07-16); the model ID and presets are provisional mirrors of the Kimi-K2.6 starter pending the official release. The kimi-k3 command lives on master (not yet in a PyPI release) — install from source as shown above.

stateset-agents kimi-k3 --json-output
stateset-agents kimi-k3 --starter-profile memory --json-output
stateset-agents kimi-k3 --list-profiles --json-output
stateset-agents kimi-k3 --write-config ./kimi_k3.json
stateset-agents kimi-k3 --config ./kimi_k3.json --no-dry-run
python examples/finetune_kimi_k3_gspo.py --dry-run

Gemma 4 31B starter path

If you want the fastest path to a first post-training run for google/gemma-4-31B-it, use the dedicated CLI starter or the equivalent example script:

pip install "stateset-agents[training,trl]"
stateset-agents gemma-4-31b --json-output
stateset-agents gemma-4-31b --starter-profile memory --json-output
stateset-agents gemma-4-31b --list-profiles --json-output
stateset-agents gemma-4-31b --write-config ./gemma4_31b.json
stateset-agents gemma-4-31b --config ./gemma4_31b.json --no-dry-run
python examples/finetune_gemma4_31b_gspo.py --dry-run

The memory profile uses 4-bit quantization and smaller context/group sizes for tighter GPU budgets.

GLM 5.1 starter path

zai-org/GLM-5.1 is a 754B-parameter MoE model (QLoRA-only, vLLM generation, multi-node or 8× H200/B200 serving). It ships as a starter module + example script rather than a CLI command:

pip install "stateset-agents[training,trl,vllm]"
python examples/finetune_glm5_1_gspo.py --dry-run
python examples/finetune_glm5_1_gspo.py --config ./glm5_1.json --no-dry-run

Import the helpers directly for programmatic use:

from stateset_agents.training.glm5_1_starter import (
    get_glm5_1_config,
    describe_glm5_1_starter_profiles,
    run_glm5_1_config,
)

See docs/GLM5_1_HOSTING_PLAN.md for the FP8 multi-node topology.

GLM 5.2 starter path

zai-org/GLM-5.2 is a 754B-parameter MoE model (QLoRA-only, vLLM generation, multi-node or 8× H200/B200 serving). It ships as a starter module + example script rather than a CLI command:

pip install "stateset-agents[training,trl,vllm]"
python examples/finetune_glm5_2_gspo.py --dry-run
python examples/finetune_glm5_2_gspo.py --config ./glm5_2.json --no-dry-run

Import the helpers directly for programmatic use:

from stateset_agents.training.glm5_2_starter import (
    get_glm5_2_config,
    describe_glm5_2_starter_profiles,
    run_glm5_2_config,
)

See docs/GLM5_2_HOSTING_PLAN.md for the FP8 multi-node topology.

Supported models

First-class starters ship for Qwen 3.5 0.8B, Gemma 4 31B IT, Kimi-K2.6, Kimi-K3 (provisional), GLM 5.1, and GLM 5.2. Reference examples and hosting plans cover Qwen 3.5 27B, Qwen 3, Qwen 2.5, Kimi-K2.5, Gemma 3 / Gemma 2 27B IT, Llama 3, Llama 2 7B, and Mistral 7B. Any HuggingFace causal LM compatible with AutoModelForCausalLM + TRL GRPO is supported through the generic flow.

See docs/SUPPORTED_MODELS.md for the full matrix, algorithm compatibility, and instructions for adding a new starter.

Dashboard and mobile app (demo, not deployed)

dashboard/ (React + Vite) and mobile/ (Expo) are working clients for a simulator-backed /api/lab/* "Training Lab" router. Both are real code you can run locally, but neither has a deployment path today — the router is gated behind auth and the API_ENABLE_TRAINING_LAB flag (off by default), and there is no hosted instance of either app. See dashboard/README.md and mobile/README.md for status and how to run them.

API serving (/v1/messages)

export INFERENCE_BACKEND=vllm
export INFERENCE_BACKEND_URL=http://localhost:8001
export INFERENCE_DEFAULT_MODEL=moonshotai/Kimi-K2.5
# Optional: ask the backend to include token usage in streaming chunks when supported.
export INFERENCE_STREAM_INCLUDE_USAGE=true

curl http://localhost:8000/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K2.5",
    "max_tokens": 128,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

OpenAI-compatible endpoint:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K2.5",
    "max_tokens": 128,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Helm deployment

helm upgrade --install stateset-agents deployment/helm/stateset-agents \
  --namespace stateset-agents

---

Quick start

1) Stub hello world (no downloads)

Runs without Torch/transformers and is ideal for CI or prototyping.

import asyncio
from stateset_agents import MultiTurnAgent
from stateset_agents.core.agent import AgentConfig

async def main():
    agent = MultiTurnAgent(AgentConfig(model_name="stub://demo"))
    await agent.initialize()
    reply = await agent.generate_response([{"role": "user", "content": "Hi!"}])
    print(reply)

asyncio.run(main())

2) Chat with a real model

import asyncio
from stateset_agents import MultiTurnAgent
from stateset_agents.core.agent import AgentConfig

async def main():
    agent = MultiTurnAgent(
        AgentConfig(
            model_name="your-real-model-id",
            max_new_tokens=128,
            temperature=0.7,
        )
    )
    await agent.initialize()
    messages = [{"role": "user", "content": "What is GRPO?"}]
    print(await agent.generate_response(messages))

asyncio.run(main())

For the zero-download onboarding path, run python examples/quick_start.py.

---

Train a multi‑turn agent with GRPO

The high‑level train(...) helper chooses single‑turn vs multi‑turn GRPO automatically.

import asyncio
from stateset_agents import (
    MultiTurnAgent,
    ConversationEnvironment,
    CompositeReward,
    HelpfulnessReward,
    SafetyReward,
    train,
)
from stateset_agents.core.agent import AgentConfig

async def main():
    # 1) Agent
    agent = MultiTurnAgent(
        AgentConfig(
            model_name="stub://quickstart",
            use_stub_model=True,
            system_prompt="You are a helpful customer support assistant.",
        )
    )
    await agent.initialize()

    # 2) Environment
    scenarios = [
        {
            "id": "refund",
            "topic": "refunds",
            "context": "User wants a refund for a delayed order.",
            "user_responses": [
                "My order is late.",
                "I'd like a refund.",
                "Thanks for your help.",
            ],
        }
    ]
    env = ConversationEnvironment(scenarios=scenarios, max_turns=6)

    # 3) Reward
    reward_fn = CompositeReward(
        [HelpfulnessReward(weight=0.7), SafetyReward(weight=0.3)]
    )

    # 4) Train
    trained_agent = await train(
        agent=agent,
        environment=env,
        reward_fn=reward_fn,
        num_episodes=4,
        profile="balanced",
        training_mode="single_turn",
        save_path="./outputs/refund_agent",
    )

    # 5) Try the trained model
    resp = await trained_agent.generate_response(
        [{"role": "user", "content": "My order was delayed, what can you do?"}]
    )
    print(resp)

asyncio.run(main())

More end‑to‑end scripts live in examples/complete_grpo_training.py and examples/production_ready_customer_service.py.

---

Continual learning + long‑term planning (optional)

Enable planning context and replay/LwF in the trainer with config overrides:

agent = MultiTurnAgent(
    AgentConfig(
        model_name="stub://quickstart",
        use_stub_model=True,
        enable_planning=True,
        planning_config={"max_steps": 4},
    )
)

trained_agent = await train(
    agent=agent,
    environment=env,
    reward_fn=reward_fn,
    num_episodes=4,
    training_mode="single_turn",
    # resume_from_checkpoint="./outputs/checkpoint-100",
    config_overrides={
        "continual_strategy": "replay_lwf",
        "continual_kl_beta": 0.1,
        "replay_buffer_size": 500,
        "replay_ratio": 0.3,
        "replay_sampling": "balanced",
        "task_id_key": "task_id",
        "task_schedule": ["task_a", "task_b"],
        "task_switch_steps": 25,
    },
)

context = {"conversation_id": "demo-trip", "goal": "Plan a 4-day trip to Kyoto"}
resp = await trained_agent.generate_response(
    [{"role": "user", "content": "Can you draft a plan?"}],
    context=context,
)

followup = await trained_agent.generate_response(
    [{"role": "user", "content": "Great. What should we do next?"}],
    context={"conversation_id": "demo-trip", "plan_update": {"action": "advance"}},
)

# To update the plan goal explicitly:
# context={"conversation_id": "demo-trip", "plan_goal": "Plan a 4-day trip to Osaka"}

---

Other training algorithms

All algorithms are available under stateset_agents.training when training deps are installed:

• GSPO: stable sequence‑level GRPO variant (GSPOTrainer, GSPOConfig, train_with_gspo)
• GEPO: expectation‑based group optimization for heterogeneous/distributed setups
• DAPO: decoupled clip + dynamic sampling for reasoning‑heavy tasks
• VAPO: value‑augmented group optimization (strong for math/reasoning)
• PPO baseline: standard PPO trainer for comparison
• RLAIF: RL from AI feedback via judge/reward models

Minimal GSPO sketch:

from stateset_agents.training import get_config_for_task, GSPOConfig, train_with_gspo
from stateset_agents.rewards.multi_objective_reward import create_customer_service_reward

base_cfg = get_config_for_task("customer_service", model_name="your-real-model-id")
gspo_cfg = GSPOConfig.from_training_config(base_cfg, num_outer_iterations=5)

trained_agent = await train_with_gspo(
    config=gspo_cfg,
    agent=agent,
    environment=env,
    reward_model=create_customer_service_reward(),
)

See docs/GSPO_GUIDE.md, docs/ADVANCED_RL_ALGORITHMS.md, and examples/train_with_gspo.py for full configs.

---

Scaffold a fine‑tuning project in 30 seconds

If you're building a fine‑tune for a client, start from a template instead of from scratch:

# See what's available
stateset-agents starter list

# Multi-turn customer support agent (the framework's differentiator)
stateset-agents starter customer-support ./my-client

# Single-turn math reasoner with verifiable rewards
stateset-agents starter gsm8k-math ./math-bench

# Agent that learns to invoke tools/APIs (weather, calculator, search stubs)
stateset-agents starter tool-calling-agent ./tool-agent

# Bare scaffold — edit everything
stateset-agents starter minimal ./hack

Each scaffold lands a runnable project: config.yaml, scenarios.jsonl (where applicable), reward.py, train.py, eval.py, serve.sh, plus a tailored README.md. From clone to running endpoint in three commands:

cd my-client
pip install -r requirements.txt
python train.py                          # trains on the bundled sample data
./serve.sh outputs/customer_support_v1   # serves via FastAPI gateway

Replace scenarios.jsonl with your client's data — same schema — and you're consulting.

---

Chat with your fine‑tune locally

# Interactive REPL — no API server needed, exits cleanly with /quit or Ctrl+D
stateset-agents chat --model Qwen/Qwen3.5-0.8B --checkpoint outputs/acme_v1

# With live reward grading — see scores after every assistant turn
stateset-agents chat --grade customer_support --history conversation.jsonl

The chat REPL is the fastest path from "did my fine-tune even load?" to "let me feel how it behaves on the queries that matter." The optional --history flag captures every turn to JSONL for later grading or replay; --grade shows live composite-reward scores so you can spot reward-function disagreements with your intuition in real time.

Curate good examples — build the next training set

After capturing many conversations, score them with the same reward function used during training, and curate the high-scoring ones as new training data:

# Grade every transcript in a directory + collect good examples into one JSONL
make grade-batch DIR=transcripts/ REWARD=customer_support \
                 CURATED=curated.jsonl THRESHOLD=0.7

# One-shot summary across all graded sessions
make grade-batch-summary GRADED_DIR=transcripts/graded

The curated file is idempotent across reruns — duplicate (prompt, response) pairs are skipped, so you can re-grade as your reward function evolves without polluting the curated set.

This closes the human-in-the-loop curation cycle: train → eval → chat → capture → grade → curate → train again.

Benchmark your fine‑tune

After training, you usually want a defensible number: did this actually improve over the base model, by how much, and is it reproducible? The framework ships a Phase‑0 benchmark pipeline that produces publication‑grade results across three tasks (GSM8K, the bundled customer‑support corpus, and the tool‑calling corpus).

Quick path: open one of the bundled Colab notebooks. The whitepaper §11.7 canonical result was produced by customer_support_3seed_judge.ipynb — judge improvement +0.079 with three-seed agreement on Qwen2.5-0.5B-Instruct (artifact).

Notebook	Task	Runtime on A100
notebooks/customer_support_3seed_judge.ipynb	Whitepaper §11.7 publication-gate notebook — 3 seeds × dual eval (rubric + LLM judge)	~25 min
notebooks/whitepaper_v1_comparative_trainers.ipynb	TRL GRPO vs GSPO vs DAPO head-to-head on §11.7 protocol	~45 min
notebooks/whitepaper_v1_gsm8k_benchmark.ipynb	GSM8K (single‑turn math) — binary reward	~45 min
notebooks/whitepaper_v1_gsm8k_benchmark_v2.ipynb	GSM8K — dense-reward A/B variant	~45 min
notebooks/customer_support_4h.ipynb	Multi‑turn customer support (single-seed)	~3 h
notebooks/vllm_speedup_benchmark.ipynb	HF generate vs vLLM throughput sweep for §6.4	~20 min

See notebooks/README.md for all ten core notebooks (the four above plus quickstart, tool-calling, curate, SFT-closure, and the standard GSM8K variant). Every notebook is JSON-validated and lint-checked in CI via scripts/lint_notebooks.py — pre-flighting against the eight foot-gun patterns from issue #16 (asyncio.run in Jupyter, abstract Agent base, flash-attn defaults, etc.).

CLI path (local A100 / H100):

# 6-second pipeline health check (no GPU)
make benchmark-smoke

# Run one configuration
make benchmark-phase0 TRAINER=gspo SEED=42

# Full matrix: 3 trainers × 3 seeds × 1 task = 9 runs
make benchmark-phase0-all

# Aggregate JSONs → markdown + CSV + PNG figures + gate report
make release-whitepaper-v1

The pipeline:

• Reproducibility. set_all_seeds() covers Python random, NumPy, PyTorch (CPU + CUDA), and Transformers in one call. Every result JSON carries the git commit hash.
• Schema. Each run produces a single JSON conforming to benchmark_results/SCHEMA.md. Every published number traces back to a file.
• Publication gates. 3 seeds, σ < 0.10, +0.03 improvement, single commit. Use make benchmark-aggregate-strict in CI to enforce.
• Figures. make benchmark-plot produces two whitepaper‑ready PNGs (pass@1 per trainer, improvement ranking) plus a matplotlib‑free text fallback.
• One‑shot release. make release-whitepaper-v1 aggregates → plots → generates the whitepaper §11.7 markdown snippet → copies figures into docs/figures/ → writes a release manifest. Six artifacts in one command.

See benchmark_results/README.md for the full pipeline reference.

---

Offline RL: Learn from logged conversations

Train agents from historical conversation logs without online interaction. Useful when:

• You have existing customer service transcripts
• Online training is expensive or risky
• You want to bootstrap before online fine‑tuning

Available Algorithms

Algorithm	Best For	Key Innovation
BCQ	Conservative learning	VAE‑constrained action space
BEAR	Distribution matching	MMD kernel regularization
CQL	Pessimistic Q‑values	Conservative Q‑function penalty
IQL	Expectile regression	Implicit value learning
Decision Transformer	Sequence modeling	Return‑conditioned generation

Quick Start

from stateset_agents.data import ConversationDataset, ConversationDatasetConfig
from stateset_agents.training import BCQTrainer, BCQConfig

# Load historical conversations
config = ConversationDatasetConfig(quality_threshold=0.7)
dataset = ConversationDataset.from_jsonl("conversations.jsonl", config)

# Train with BCQ
bcq_config = BCQConfig(
    hidden_dim=256,
    latent_dim=64,
    num_epochs=100,
)
trainer = BCQTrainer(bcq_config)
await trainer.train(dataset)

Hybrid Offline + Online Training

Combine offline pretraining with online GRPO fine‑tuning:

from stateset_agents.training import OfflineGRPOTrainer, OfflineGRPOConfig

config = OfflineGRPOConfig(
    offline_algorithm="cql",
    offline_pretrain_steps=1000,
    online_ratio=0.3,  # 30% online, 70% offline
)
trainer = OfflineGRPOTrainer(config)
trained = await trainer.train(agent, env, reward_fn, offline_dataset=dataset)

See docs/OFFLINE_RL_SIM_TO_REAL_GUIDE.md for complete documentation.

---

Sim‑to‑Real Transfer

Train in simulation, deploy to real users. The framework provides:

Domain Randomization

Generate diverse training scenarios with randomized user personas:

from stateset_agents.training import DomainRandomizer, DomainRandomizationConfig

config = DomainRandomizationConfig(
    persona_variation=0.3,
    topic_variation=0.2,
    style_variation=0.2,
)
randomizer = DomainRandomizer(config)

# Randomize during training
persona = randomizer.sample_persona()
scenario = randomizer.sample_scenario(topic="returns")

Conversation Simulator

Calibratable simulator with adjustable realism:

from stateset_agents.environments import ConversationSimulator, ConversationSimulatorConfig

simulator = ConversationSimulator(ConversationSimulatorConfig(
    base_model="gpt2",
    realism_level=0.8,
))

# Calibrate to real data
await simulator.calibrate(real_conversations)

# Measure sim‑to‑real gap
gap = simulator.compute_sim_real_gap(real_data, sim_data)

Progressive Transfer

Gradually transition from simulation to real interactions:

from stateset_agents.training import SimToRealTransfer, SimToRealConfig

transfer = SimToRealTransfer(SimToRealConfig(
    transfer_schedule="cosine",  # linear, exponential, step
    warmup_steps=100,
    total_steps=1000,
))

# Get current sim/real mixing ratio
sim_ratio = transfer.get_sim_ratio(current_step)

See docs/OFFLINE_RL_SIM_TO_REAL_GUIDE.md for complete documentation.

---

Hyperparameter optimization (HPO)

Install with stateset-agents[hpo], then:

from stateset_agents.training import TrainingConfig, TrainingProfile
from stateset_agents.training.hpo import quick_hpo

base_cfg = TrainingConfig.from_profile(
    TrainingProfile.BALANCED, num_episodes=100
)

summary = await quick_hpo(
    agent=agent,
    environment=env,
    reward_function=reward_fn,
    base_config=base_cfg,
    n_trials=30,
)
print(summary.best_params)

See docs/HPO_GUIDE.md and examples/hpo_training_example.py.

---

Custom rewards

Use the decorator for quick experiments:

from stateset_agents.core.reward import reward_function

@reward_function(weight=0.5)
async def politeness_reward(turns, context=None) -> float:
    return 1.0 if any("please" in t.content.lower() for t in turns) else 0.0

Combine with built‑ins via CompositeReward.

---

Custom environments

Subclass Environment for task‑specific dynamics:

from stateset_agents.core.environment import Environment, EnvironmentState
from stateset_agents.core.trajectory import ConversationTurn

class MyEnv(Environment):
    async def reset(self, scenario=None) -> EnvironmentState:
        ...

    async def step(
        self, state: EnvironmentState, action: ConversationTurn
    ):
        ...

---

Checkpoints

• train(..., save_path="...") saves an agent checkpoint.
• Load later:

from stateset_agents.core.agent import load_agent_from_checkpoint

agent = await load_agent_from_checkpoint("./outputs/refund_agent")

---

Auto‑Research

Run autonomous hyperparameter experiments overnight. The loop proposes configurations, trains with a time budget, evaluates on held‑out scenarios, and keeps only improvements.

# Quick test (no GPU)
stateset-agents auto-research --stub --max-experiments 5

# Real training with smart proposer
stateset-agents auto-research --proposer smart --improvement-patience 10

# From a config file
stateset-agents auto-research --config config.yaml

7 proposer strategies (perturbation, smart, adaptive, random, grid, bayesian, LLM), 5 search spaces, early abort on bad experiments, resume from checkpoint, W&B logging, and post‑run analysis with parameter importance.

# Load and analyze results after a run
from stateset_agents.training.auto_research import ExperimentTracker, compare_runs
tracker = ExperimentTracker.load("./auto_research_results")
tracker.print_summary()
print(compare_runs("./run_a", "./run_b"))

See docs/AUTO_RESEARCH_GUIDE.md for the full guide.

---

CLI

The CLI is a thin wrapper around the Python API:

stateset-agents version
stateset-agents doctor
stateset-agents train --stub
stateset-agents train --config ./config.yaml --dry-run false --save ./outputs/ckpt
stateset-agents evaluate --checkpoint ./outputs/ckpt --message "Hello"
stateset-agents serve --host 0.0.0.0 --port 8001
stateset-agents auto-research --proposer smart --max-experiments 50

For complex runs prefer the Python API and the examples folder.

---

Examples and docs

Start here:

• docs/WHITEPAPER.md — the v0.13.4 technical whitepaper. Anchored to a specific git commit; every claim is verifiable via Appendix C.
• docs/WHITEPAPER_ERRATA.md — corrections published after each whitepaper revision.
• docs/PLATFORM_TOUR.md — a guided walk from pip install to a published v1.0 whitepaper revision (linear, journey-style).
• docs/COOKBOOK.md — copy-paste recipes for 8 common workflows (look up what you need).
• notebooks/README.md — a map of the ten bundled Colab notebooks: which to open when.
• benchmark_results/whitepaper_v1/ — first-party result artifacts including the §11.7 canonical positive result.
• CHANGELOG.md — what changed in each release (latest release v0.18.0).

Other entry points:

• examples/getting_started/ — start here after pip install: five small examples (stub hello, custom reward, first GSPO fine-tune, LLM-judge eval, serve via FastAPI). All target the published PyPI version; the GPU-free three smoke-test the install end-to-end. Run make getting-started-smoke to verify all three at once.
• examples/finetune_gspo.py – unified finetune driver: --model <preset> over the 12-model registry (--list-models), safe --dry-run by default, --no-dry-run to train
• examples/hello_world.py – stub mode walkthrough
• examples/quick_start.py – stub-backed onboarding example with training + smoke test
• examples/complete_grpo_training.py – end‑to‑end GRPO training
• examples/train_with_gspo.py – GSPO + GSPO‑token training
• examples/train_with_trl_grpo.py – Hugging Face TRL GRPO integration
• examples/auto_research_quickstart.py – autonomous experiment loop

Key docs:

• docs/AUTO_RESEARCH_GUIDE.md
• docs/RL_FRAMEWORK_GUIDE.md — canonical usage guide
• docs/GSPO_GUIDE.md
• docs/OFFLINE_RL_SIM_TO_REAL_GUIDE.md
• docs/HPO_GUIDE.md
• docs/CLI_REFERENCE.md
• docs/ARCHITECTURE.md

---

Related Projects

• stateset-nsr - Neuro‑symbolic reasoning engine for explainable tools.
• stateset-api - Commerce/operations API that agents can drive.
• stateset-sync-server - Multi‑tenant orchestration and integrations.
• core - Cosmos SDK blockchain for on‑chain commerce.
• Public API docs: https://docs.stateset.com

---

Contributing

See CONTRIBUTING.md. Please run pytest -q and format with black/isort before opening a PR.

---

License

Business Source License 1.1. Non‑production use permitted until 2029‑09‑03, then transitions to Apache 2.0. See LICENSE.