Skip to main content Accueil Créateurs adu2021 skillxiv search-r2-refinement-rl
search-r2-refinement-rl Decompose search-integrated reasoning into actor (generates trajectories) and refiner (identifies and corrects flawed steps). Refiner performs surgical cut-and-regenerate correction preserving valid prefixes. Trained end-to-end with GRPO; adds minimal overhead (2%) to baseline.
Aller à l'installation Skills Marketplace Découvrez et explorez les compétences IA créées par la communauté.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Afficher les détails du prompt Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
npx skills add https://github.com/ADu2021/skillXiv --skill search-r2-refinement-rlLa commande reste sur une seule ligne. Faites défiler horizontalement pour la vérifier avant de la copier.
Vous préférez une copie locale ? Téléchargez les fichiers actuellement disponibles dans SkillsMP.
Télécharger Zip Téléchargement... Métiers associés SOC
Basé sur la classification professionnelle SOC
name search-r2-refinement-rl title Search-R2: Enhancing Search-Integrated Reasoning via Actor-Refiner Collaboration version 0.0.2 engine skillxiv-v0.0.2-claude-opus-4.6 license MIT url https://arxiv.org/abs/2602.03647 keywords ["Reasoning Refinement","Search Integration","RL Training","Trajectory Correction","GRPO Optimization"] description Decompose search-integrated reasoning into actor (generates trajectories) and refiner (identifies and corrects flawed steps). Refiner performs surgical cut-and-regenerate correction preserving valid prefixes. Trained end-to-end with GRPO; adds minimal overhead (2%) to baseline.
Search-R2: Actor-Refiner Architecture for Search-Integrated Reasoning
When solving complex problems with web search integration, agents often generate reasoning trajectories with flawed intermediate steps that invalidate later reasoning. Rather than discarding entire trajectories, Search-R2 uses a two-role system: the actor generates candidate reasoning paths with search queries, and the refiner identifies specific errors and regenerates only the problematic suffix. This targeted correction preserves valid reasoning while fixing errors efficiently.
The key insight is that not all trajectory errors require full re-reasoning. By pinpointing the exact error location and regenerating only that segment, the system maintains coherence while enabling effective learning via RL.
Core Concept
Search-R2 operates on two complementary roles:
Actor Policy : Generates search-integrated reasoning trajectories with search queriesMeta-Refiner : Analyzes trajectories, identifies flawed reasoning, and performs targeted correction
The system uses a discriminator to detect global coherence issues and a trimmer to pinpoint error locations, then regenerates optimal suffixes.
Architecture Overview
Actor Agent : Standard reasoning agent with search tool accessTrajectory Analyzer : Identifies steps with consistency issuesError Detector : Binary classifier identifying whether error existsError Locator : Pinpoints exact step where reasoning divergesSuffix Generator : Regenerates reasoning from error point forwardGRPO Trainer : Optimizes both actor and refiner end-to-end
Implementation
Step 1: Build Trajectory Analysis Module
Create tools to analyze reasoning trajectories for errors.
class TrajectoryAnalyzer :
def __init__ (self, model: str = "gpt-4-turbo" ):
"""Analyze search-integrated reasoning trajectories."""
self .model = model
def extract_reasoning_steps (self, trajectory: ) -> str
List
dict
"""Break trajectory into individual reasoning steps."""
"thought"
""
"action"
""
"observation"
""
"\n"
for
in
if
"Thought:"
if
"thought"
"thought"
""
"action"
""
"observation"
""
"thought"
"Thought:"
""
elif
"Action:"
"action"
"Action:"
""
elif
"Observation:"
"observation"
"Observation:"
""
if
"thought"
return
def
compute_trajectory_embedding
self, trajectory: str
"""Embed trajectory for consistency analysis."""
768
return
def
measure_consistency
self,
step_idx: int ,
trajectory_steps: List [dict ],
full_trajectory: str
float
"""
Measure consistency of reasoning at step.
Returns:
Score [0, 1]: 1 = fully consistent, 0 = contradictory
"""
if
0
return
1.0
"\n"
f"Step {i} : {s['thought' ]} "
for
in
enumerate
f"""
Given this reasoning context:
{context}
Is this next step logically consistent?
{current_step['thought' ]}
Rate consistency 0-10:"""
self
return
10.0
def
identify_error_location
self,
trajectory_steps: List [dict ],
full_trajectory: str
Optional
int
"""
Find the step index where reasoning first becomes inconsistent.
Returns:
Index of first inconsistent step, or None if trajectory is sound
"""
0.5
for
in
range
1
len
self
if
return
return
None
Step 2: Implement Error Detection and Localization Create discriminator that identifies presence and location of errors.
class ErrorDetector (torch.nn.Module):
def __init__ (self, hidden_dim: int = 768 ):
"""Binary classifier for trajectory errors."""
super ().__init__()
self .trajectory_encoder = torch.nn.GRU(
input_size=hidden_dim,
hidden_size=hidden_dim,
batch_first=True
)
self .error_classifier = torch.nn.Sequential(
torch.nn.Linear(hidden_dim, hidden_dim // 2 ),
torch.nn.ReLU(),
torch.nn.Linear(hidden_dim // 2 , 1 )
)
def forward (self, trajectory_embeddings: torch.Tensor ) -> torch.Tensor:
"""
Classify trajectory for errors.
Args:
trajectory_embeddings: [batch, num_steps, hidden_dim]
Returns:
logits: [batch] binary classification logits
"""
_, hidden = self .trajectory_encoder(trajectory_embeddings)
logits = self .error_classifier(hidden.squeeze(0 ))
return logits
class ErrorLocator (torch.nn.Module):
def __init__ (self, hidden_dim: int = 768 ):
"""Identify which step contains the error."""
super ().__init__()
self .step_scorer = torch.nn.Sequential(
torch.nn.Linear(hidden_dim * 2 , hidden_dim),
torch.nn.ReLU(),
torch.nn.Linear(hidden_dim, 1 )
)
def forward (
self,
trajectory_embeddings: torch.Tensor
) -> torch.Tensor:
"""
Score each step for error likelihood.
Args:
trajectory_embeddings: [batch, num_steps, hidden_dim]
Returns:
scores: [batch, num_steps] error likelihood per step
"""
batch_size, num_steps, hidden_dim = trajectory_embeddings.shape
scores = []
for step_idx in range (num_steps):
if step_idx == 0 :
context = torch.zeros(batch_size, hidden_dim,
device=trajectory_embeddings.device)
else :
context = trajectory_embeddings[:, :step_idx].mean(dim=1 )
current = trajectory_embeddings[:, step_idx]
combined = torch.cat([context, current], dim=-1 )
step_score = self .step_scorer(combined)
scores.append(step_score)
scores = torch.cat(scores, dim=-1 )
return scores
Step 3: Implement Cut-and-Regenerate Mechanism Create the refiner that performs surgical trajectory correction.
class TrajectoryRefiner :
def __init__ (self, model: str = "gpt-4-turbo" ):
"""Refine trajectories by targeted regeneration."""
self .model = model
self .analyzer = TrajectoryAnalyzer(model)
def refine_trajectory (
self,
original_trajectory: str ,
error_step_idx: int
) -> str :
"""
Cut trajectory at error and regenerate suffix.
Args:
original_trajectory: Original reasoning chain
error_step_idx: Index of first erroneous step
Returns:
Refined trajectory with corrected suffix
"""
steps = self .analyzer.extract_reasoning_steps(original_trajectory)
prefix = "\n" .join([
f"Thought: {s['thought' ]} \nAction: {s['action' ]} \nObservation: {s['observation' ]} "
for s in steps[:error_step_idx]
])
problem_context = prefix + "\n\n[Error detected above]"
regeneration_prompt = f"""
The following reasoning has an error:
{problem_context}
Generate a corrected continuation that fixes the error and reaches a conclusion.
Format: Thought: ... Action: ... Observation: ...
Corrected suffix:"""
corrected_suffix = self .model.generate(
regeneration_prompt,
max_tokens=512
)
refined_trajectory = prefix + "\n\n" + corrected_suffix
return refined_trajectory
def batch_refine (
self,
trajectories: List [str ],
error_indices: List [Optional [int ]]
) -> List [str ]:
"""Refine multiple trajectories in parallel."""
refined = []
for trajectory, error_idx in zip (trajectories, error_indices):
if error_idx is not None :
refined_traj = self .refine_trajectory(trajectory, error_idx)
refined.append(refined_traj)
else :
refined.append(trajectory)
return refined
Step 4: Implement GRPO Training with Refinement Train actor and refiner jointly using GRPO.
def train_search_r2 (
actor_model: nn.Module,
error_detector: ErrorDetector,
error_locator: ErrorLocator,
dataset: List [dict ],
num_epochs: int = 10 ,
group_size: int = 4
):
"""
Train Search-R2 with actor and refiner via GRPO.
Args:
actor_model: Reasoning agent
error_detector: Binary error classifier
error_locator: Error location identifier
dataset: Training tasks
num_epochs: Training epochs
group_size: GRPO group size
"""
optimizer = torch.optim.AdamW(
list (actor_model.parameters()) +
list (error_detector.parameters()) +
list (error_locator.parameters()),
lr=1e-5
)
refiner = TrajectoryRefiner()
for epoch in range (num_epochs):
for group_idx in range (0 , len (dataset), group_size):
group = dataset[group_idx:group_idx + group_size]
group_rewards = []
trajectories_refined = []
for task in group:
trajectory = actor_model.generate(task["description" ])
trajectory_emb = torch.randn(1 , 10 , 768 )
error_logits = error_detector(trajectory_emb)
has_error = error_logits > 0
if has_error:
error_scores = error_locator(trajectory_emb)
error_step_idx = torch.argmax(error_scores).item()
refined = refiner.refine_trajectory(
trajectory,
error_step_idx
)
else :
refined = trajectory
trajectories_refined.append(refined)
reward = evaluate_trajectory(refined, task["target" ])
group_rewards.append(reward)
group_tensor = torch.tensor(group_rewards, dtype=torch.float32)
group_mean = group_tensor.mean()
for traj, reward in zip (trajectories_refined, group_rewards):
advantage = reward - group_mean
if advantage != 0 :
loss = -advantage * compute_log_prob(actor_model, traj)
loss.backward()
optimizer.step()
optimizer.zero_grad()
print (f"Epoch {epoch} : avg_reward={group_tensor.mean():.4 f} " )
return actor_model, error_detector, error_locator
Step 5: Inference Pipeline Use trained models for deployment.
def search_r2_inference (
actor: nn.Module,
error_detector: ErrorDetector,
error_locator: ErrorLocator,
query: str ,
max_refinement_iterations: int = 1
) -> str :
"""
Generate response with optional refinement.
Args:
actor: Trained reasoning agent
error_detector: Error detector
error_locator: Error locator
query: User query
max_refinement_iterations: How many refinement passes
Returns:
Final refined response
"""
refiner = TrajectoryRefiner()
trajectory = actor.generate(query)
for iteration in range (max_refinement_iterations):
trajectory_emb = torch.randn(1 , 10 , 768 )
error_logits = error_detector(trajectory_emb)
if error_logits <= 0 :
break
error_scores = error_locator(trajectory_emb)
error_idx = torch.argmax(error_scores).item()
trajectory = refiner.refine_trajectory(trajectory, error_idx)
return trajectory
Practical Guidance
Complex reasoning tasks with web search integration
Scenarios where reasoning trajectories have logical errors
Problems where intermediate steps can be independently refined
RL-based training where full trajectory discarding is wasteful
Simple lookup tasks without complex reasoning
Real-time systems where refinement overhead matters
Tasks where any trajectory modification breaks coherence
Systems requiring deterministic behavior
Error detection false positives: Over-aggressive detection refines correct reasoning
Prefix contamination: Ensuring cut point doesn't bias suffix generation
Circular refinement: Error refiner introduces new errors; set max_iterations
Search query consistency: Regenerated suffix may require different searches
Hyperparameter Guidelines:
Parameter Range Tuning error_threshold 0.3-0.7 Higher = fewer refinements; lower = more aggressive max_refinement 1-3 iterations 1 typical; more for complex reasoning group_size 4-8 Standard GRPO setting
Reference Key results: Outperforms baseline Search-R1; minimal 2-9% overhead. Demonstrates surgical correction preserves reasoning quality. Code and training details released. Applicable to any search-integrated agent.
