菜单
Train multi-agent reasoning systems with decoupled reward signals and pipeline parallelism—enable specialized Solver/Verifier/Corrector agents to iteratively refine solutions without waiting for full trajectories, handling extended reasoning up to 320K tokens.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ADu2021/skillXiv --skill mars-rl-multi-agent-reasoning命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | mars-rl-multi-agent-reasoning |
| title | MarsRL: Multi-Agent Reasoning System via RL with Agentic Pipeline Parallelism |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2511.11373 |
| keywords | ["Multi-Agent RL","Reasoning","Pipeline Parallelism","Credit Assignment","Extended Reasoning"] |
| description | Train multi-agent reasoning systems with decoupled reward signals and pipeline parallelism—enable specialized Solver/Verifier/Corrector agents to iteratively refine solutions without waiting for full trajectories, handling extended reasoning up to 320K tokens. |
Extended reasoning requires multiple agents: a Solver generates candidate solutions, a Verifier checks correctness, a Corrector refines errors. Traditional RL training waits for complete trajectories before updating, creating bottlenecks when trajectories span hundreds of thousands of tokens. MarsRL solves this via two innovations:
This enables efficient training of specialized agents working in concert, where each agent specializes (Solver on generation, Verifier on discrimination, Corrector on refinement) without reward contamination.
Multi-agent reasoning systems decompose complex problems into specialized subtasks. However, monolithic reward signals (single score for entire trajectory) fail to credit individual agents for their contributions—a Solver's poor generation gets blamed on the Verifier, and vice versa.
MarsRL decouples credit assignment: the Solver is rewarded only for solution correctness, the Verifier for discrimination accuracy, the Corrector for successful refinement. Additionally, rather than accumulating complete trajectories (potentially 320k tokens) before training, MarsRL begins training immediately after each 16k-token segment completes. This "segment rollout" approach combined with pipeline parallelism reduces training latency while maintaining RL signal quality.
Step 1: Define Agent-Specific Rewards. Assign rewards based on individual agent responsibility, not shared outcome.
:
():
.reference_answers = reference_answers
():
.is_correct(solution, reference_answer):
:
-
():
ground_truth_correct = .is_correct(solution, .reference_answers[problem])
verifier_judgment == ground_truth_correct:
:
-
():
original_correct = .is_correct(original_solution, reference_answer)
corrected_correct = .is_correct(corrected_solution, reference_answer)
original_correct corrected_correct:
corrected_correct:
:
-
():
solution.strip() == reference.strip()
Step 2: Implement Segment-Based Rollout. Decode agents in 16k-token chunks and queue for training immediately.
class SegmentRollout:
def __init__(self, segment_size=16000, agent_budget=64000):
self.segment_size = segment_size
self.agent_budget = agent_budget
self.training_queue = []
def generate_solver_trajectory(self, problem, solver_model):
"""
Solver generates solution in segments.
Each segment enters training queue upon completion.
"""
solution_tokens = []
solver_model.reset_state()
num_segments = self.agent_budget // self.segment_size
for seg_idx in range(num_segments):
# Decode one segment
segment = solver_model.generate(
problem,
max_tokens=self.segment_size,
state=solver_model.get_state()
)
solution_tokens.extend(segment)
# Immediately queue for training (don't wait for full solution)
self.training_queue.append({
'agent': 'solver',
'problem': problem,
'partial_solution': ''.join(solution_tokens),
'segment_idx': seg_idx,
'trajectory': solver_model.get_trajectory() # logits, actions, etc.
})
# Check for early stopping
if solver_model.is_complete(solution_tokens):
break
return ''.join(solution_tokens)
def process_training_queue(self, reward_fn):
"""
Process queued segments for RL training.
Training begins immediately; doesn't wait for downstream agents.
"""
while self.training_queue:
item = self.training_queue.pop(0)
if item['agent'] == 'solver':
# Compute reward for solver
reward = reward_fn.reward_solver(
item['problem'],
item['partial_solution'],
# Use final reference for this problem
)
# Update solver model via RL (e.g., PPO, GRPO)
rl_loss = compute_rl_loss(item['trajectory'], reward)
solver_model.backward(rl_loss)
# Similar for Verifier, Corrector
Step 3: Grouped Agentic Rollouts. Each problem generates multiple diverse trajectories per agent.
def grouped_agentic_rollouts(problem, num_solutions=8):
"""
Generate k solver outputs per problem;
subsequent agents sample from these for diverse supervision.
"""
solver_outputs = []
for i in range(num_solutions):
solution = solver_model.generate(problem)
solver_outputs.append(solution)
# Verifier evaluates all solver outputs
verification_results = []
for solution in solver_outputs:
judgment = verifier_model.judge(problem, solution)
verification_results.append({
'solution': solution,
'is_correct': judgment['is_correct'],
'confidence': judgment['confidence']
})
# Corrector focuses on errors
for result in verification_results:
if not result['is_correct']:
# Corrector attempts to refine error
corrected = corrector_model.correct(
problem,
result['solution'],
result['confidence']
)
# Add to training queue for corrector reward
training_queue.append({
'agent': 'corrector',
'problem': problem,
'original': result['solution'],
'corrected': corrected
})
return {
'solver_outputs': solver_outputs,
'verification_results': verification_results
}
When to Use: Complex reasoning tasks (math, coding, multi-step proofs) where extended thinking helps, and where you can decompose reasoning into specialized subtasks (generation, verification, refinement).
Architecture Choices:
Pitfalls:
When NOT to Use: Single-agent tasks (classification, simple QA) where decomposition adds overhead; problems not benefiting from verification or refinement.
Integration: Combine with outcome verification (automated test suites for code) for clearer reward signals; pairs well with expert demonstrations for initialization.
Reference: https://arxiv.org/abs/2511.11373