Merge pull request #188 from codelion/feat-add-deep-think-plugin

init

Author: codelion

optillm/__init__.py

@@ -2,7 +2,7 @@
 import os
 
 # Version information
-__version__ = "0.1.12"
+__version__ = "0.1.13"
 
 # Get the path to the root optillm.py
 spec = util.spec_from_file_location(

optillm/plugins/deepthink/README.md

@@ -0,0 +1,136 @@
+# Deep Think Plugin
+
+## Overview
+
+The Deep Think plugin combines two powerful approaches for enhanced reasoning in large language models:
+
+1. **SELF-DISCOVER Framework**: A method where LLMs self-discover task-intrinsic reasoning structures by selecting, adapting, and implementing atomic reasoning modules into a coherent reasoning plan.
+
+2. **Uncertainty-Routed Chain-of-Thought**: An approach that generates multiple chain-of-thought samples, evaluates confidence through consistency, and routes to either majority voting (high confidence) or greedy decoding (low confidence).
+
+## Key Features
+
+- **Adaptive Reasoning Structure**: Automatically discovers the best reasoning approach for each specific task
+- **Confidence-Based Routing**: Uses uncertainty estimation to decide between multiple samples or single greedy output
+- **Reasoning Model Support**: Designed for models that produce structured thinking in `<think></think>` tags
+- **Multiple Sampling**: Generates multiple reasoning paths and selects the most reliable one
+
+## How It Works
+
+### Stage 1: SELF-DISCOVER Reasoning Structure
+
+1. **SELECT**: From 39 atomic reasoning modules, select those most relevant for the task
+2. **ADAPT**: Rephrase selected modules to be task-specific
+3. **IMPLEMENT**: Create a structured JSON reasoning plan
+
+### Stage 2: Uncertainty-Routed Generation
+
+1. **Multiple Sampling**: Generate n samples (default: 3) using the discovered structure
+2. **Confidence Evaluation**: Assess consistency across samples
+3. **Route Decision**: 
+   - High confidence → Use majority vote
+   - Low confidence → Use greedy sample (temperature=0)
+
+## Usage
+
+```python
+# Via optillm model prefix
+model = "deepthink-your-model-name"
+
+# Via optillm_approach in request
+{
+    "model": "your-model-name",
+    "optillm_approach": "deepthink",
+    "messages": [...],
+    "deepthink_samples": 3,           # Number of samples for uncertainty routing
+    "confidence_threshold": 0.7,      # Threshold for majority vs greedy routing
+    "max_tokens": 16382,             # Extended context for reasoning
+    "temperature": 0.7,              # Default temperature for sampling
+    "top_p": 0.95                    # Default top_p for sampling
+}
+```
+
+## Configuration Parameters
+
+- `deepthink_samples` (int, default=3): Number of reasoning samples to generate
+- `confidence_threshold` (float, default=0.7): Confidence threshold for routing decision
+- `max_tokens` (int, default=16382): Maximum tokens for generation
+- `temperature` (float, default=0.7): Sampling temperature
+- `top_p` (float, default=0.95): Top-p sampling parameter
+- `enable_self_discover` (bool, default=True): Whether to use SELF-DISCOVER structure
+- `reasoning_modules_limit` (int, default=5): Max reasoning modules to select
+
+## Atomic Reasoning Modules
+
+The plugin includes 39 reasoning modules covering:
+- Critical thinking and analysis
+- Creative and innovative approaches  
+- Systems thinking and holistic analysis
+- Risk assessment and evaluation
+- Step-by-step decomposition
+- Collaborative and perspective-taking approaches
+- Reflective and meta-cognitive strategies
+
+## Examples
+
+### Mathematical Problem Solving
+Input: "Solve: If a train travels 120 miles in 2 hours, how long will it take to travel 300 miles?"
+
+The plugin will:
+1. Discover a reasoning structure focused on rate calculations
+2. Generate multiple solution paths
+3. Evaluate consistency and select the most reliable answer
+
+### Complex Reasoning Task
+Input: "Analyze the potential long-term economic impacts of remote work adoption"
+
+The plugin will:
+1. Select reasoning modules like systems thinking, risk analysis, and critical thinking
+2. Create a structured analysis plan
+3. Generate multiple perspectives and synthesize the most coherent analysis
+
+## Implementation Details
+
+- **Reasoning Extraction**: Automatically extracts content from `<think></think>` tags
+- **Consistency Scoring**: Uses multiple metrics including answer similarity and reasoning coherence
+- **Adaptive Thresholds**: Can be fine-tuned based on model performance
+- **Token Efficiency**: Optimized to minimize redundant computation while maximizing reasoning quality
+
+## Performance
+
+The Deep Think approach has shown significant improvements on complex reasoning tasks, with particularly strong results on mathematical competition problems.
+
+### AIME 2025 Results
+
+| Model | Approach | Accuracy | Improvement |
+|-------|----------|----------|-------------|
+| qwen-3-32b | Baseline | 43.33% | - |
+| qwen-3-32b | Deep Think | **63.33%** | **+20.00pp** |
+
+*Experimental settings: max_completion_tokens=16382, temperature=0.7, top_p=0.95*
+
+**Key Findings:**
+- **46% relative improvement** over baseline on mathematical reasoning
+- **Cerebras inference** was crucial for enabling high inference-time compute without latency penalty
+- The combination of SELF-DISCOVER structure discovery and uncertainty-routed sampling proved particularly effective for competition mathematics
+- Enhanced accuracy on multi-step problems requiring systematic reasoning
+
+### Other Improvements
+
+The Deep Think approach has also demonstrated:
+- Enhanced accuracy on multi-step problems
+- Better handling of ambiguous or open-ended questions
+- Improved consistency across different problem types
+- Reduced hallucination through confidence-based routing
+
+## Limitations
+
+- Increased computational cost due to multiple sampling
+- Longer response times for complex reasoning tasks
+- Requires models capable of structured thinking output
+- May over-engineer solutions for simple problems
+
+## References
+
+- Zhou, P. et al. "SELF-DISCOVER: Large Language Models Self-Compose Reasoning Structures" (2024)
+- Uncertainty-routed chain-of-thought approaches in advanced reasoning systems

optillm/plugins/deepthink/__init__.py

@@ -0,0 +1,6 @@
+"""
+Deep Think Plugin for OptILM
+
+A plugin that combines SELF-DISCOVER framework with uncertainty-routed 
+chain-of-thought for enhanced reasoning capabilities.
+"""

optillm/plugins/deepthink/reasoning_modules.py

@@ -0,0 +1,234 @@
+"""
+Atomic Reasoning Modules for SELF-DISCOVER Framework
+
+This module contains the 39 reasoning modules as described in the SELF-DISCOVER paper.
+These modules represent high-level cognitive heuristics for problem-solving.
+"""
+
+# 39 Atomic Reasoning Modules from SELF-DISCOVER paper
+REASONING_MODULES = [
+    {
+        "id": 1,
+        "name": "experimental_design",
+        "description": "How could I devise an experiment to help solve that problem?"
+    },
+    {
+        "id": 2,
+        "name": "iterative_problem_solving",
+        "description": "Make a list of ideas for solving this problem, and apply them one by one to the problem to see if any progress can be made."
+    },
+    {
+        "id": 3,
+        "name": "progress_measurement",
+        "description": "How could I measure progress on this problem?"
+    },
+    {
+        "id": 4,
+        "name": "problem_simplification",
+        "description": "How can I simplify the problem so that it is easier to solve?"
+    },
+    {
+        "id": 5,
+        "name": "assumption_analysis",
+        "description": "What are the key assumptions underlying this problem?"
+    },
+    {
+        "id": 6,
+        "name": "risk_assessment",
+        "description": "What are the potential risks and drawbacks of each solution?"
+    },
+    {
+        "id": 7,
+        "name": "perspective_analysis",
+        "description": "What are the alternative perspectives or viewpoints on this problem?"
+    },
+    {
+        "id": 8,
+        "name": "long_term_implications",
+        "description": "What are the long-term implications of this problem and its solutions?"
+    },
+    {
+        "id": 9,
+        "name": "problem_decomposition",
+        "description": "How can I break down this problem into smaller, more manageable parts?"
+    },
+    {
+        "id": 10,
+        "name": "critical_thinking",
+        "description": "Critical Thinking: This style involves analyzing the problem from different perspectives, questioning assumptions, and evaluating the evidence or information available. It focuses on logical reasoning, evidence-based decision-making, and identifying potential biases or flaws in thinking."
+    },
+    {
+        "id": 11,
+        "name": "creative_thinking",
+        "description": "Try creative thinking, generate innovative and out-of-the-box ideas to solve the problem. Explore unconventional solutions, thinking beyond traditional boundaries, and encouraging imagination and originality."
+    },
+    {
+        "id": 12,
+        "name": "collaborative_thinking",
+        "description": "Seek input and collaboration from others to solve the problem. Emphasize teamwork, open communication, and leveraging the diverse perspectives and expertise of a group to come up with effective solutions."
+    },
+    {
+        "id": 13,
+        "name": "systems_thinking",
+        "description": "Use systems thinking: Consider the problem as part of a larger system and understanding the interconnectedness of various elements. Focus on identifying the underlying causes, feedback loops, and interdependencies that influence the problem, and developing holistic solutions that address the system as a whole."
+    },
+    {
+        "id": 14,
+        "name": "risk_analysis",
+        "description": "Use Risk Analysis: Evaluate potential risks, uncertainties, and tradeoffs associated with different solutions or approaches to a problem. Emphasize assessing the potential consequences and likelihood of success or failure, and making informed decisions based on a balanced analysis of risks and benefits."
+    },
+    {
+        "id": 15,
+        "name": "reflective_thinking",
+        "description": "Use Reflective Thinking: Step back from the problem, take the time for introspection and self-reflection. Examine personal biases, assumptions, and mental models that may influence problem-solving, and being open to learning from past experiences to improve future approaches."
+    },
+    {
+        "id": 16,
+        "name": "core_issue_identification",
+        "description": "What is the core issue or problem that needs to be addressed?"
+    },
+    {
+        "id": 17,
+        "name": "causal_analysis",
+        "description": "What are the underlying causes or factors contributing to the problem?"
+    },
+    {
+        "id": 18,
+        "name": "historical_analysis",
+        "description": "Are there any potential solutions or strategies that have been tried before? If yes, what were the outcomes and lessons learned?"
+    },
+    {
+        "id": 19,
+        "name": "obstacle_identification",
+        "description": "What are the potential obstacles or challenges that might arise in solving this problem?"
+    },
+    {
+        "id": 20,
+        "name": "data_analysis",
+        "description": "Are there any relevant data or information that can provide insights into the problem? If yes, what data sources are available, and how can they be analyzed?"
+    },
+    {
+        "id": 21,
+        "name": "stakeholder_analysis",
+        "description": "Are there any stakeholders or individuals who are directly affected by the problem? What are their perspectives and needs?"
+    },
+    {
+        "id": 22,
+        "name": "resource_analysis",
+        "description": "What resources (financial, human, technological, etc.) are needed to tackle the problem effectively?"
+    },
+    {
+        "id": 23,
+        "name": "success_metrics",
+        "description": "How can progress or success in solving the problem be measured or evaluated?"
+    },
+    {
+        "id": 24,
+        "name": "metric_identification",
+        "description": "What indicators or metrics can be used?"
+    },
+    {
+        "id": 25,
+        "name": "problem_type_technical",
+        "description": "Is the problem a technical or practical one that requires a specific expertise or skill set? Or is it more of a conceptual or theoretical problem?"
+    },
+    {
+        "id": 26,
+        "name": "physical_constraints",
+        "description": "Does the problem involve a physical constraint, such as limited resources, infrastructure, or space?"
+    },
+    {
+        "id": 27,
+        "name": "behavioral_aspects",
+        "description": "Is the problem related to human behavior, such as a social, cultural, or psychological issue?"
+    },
+    {
+        "id": 28,
+        "name": "decision_making",
+        "description": "Does the problem involve decision-making or planning, where choices need to be made under uncertainty or with competing objectives?"
+    },
+    {
+        "id": 29,
+        "name": "analytical_problem",
+        "description": "Is the problem an analytical one that requires data analysis, modeling, or optimization techniques?"
+    },
+    {
+        "id": 30,
+        "name": "design_challenge",
+        "description": "Is the problem a design challenge that requires creative solutions and innovation?"
+    },
+    {
+        "id": 31,
+        "name": "systemic_issues",
+        "description": "Does the problem require addressing systemic or structural issues rather than just individual instances?"
+    },
+    {
+        "id": 32,
+        "name": "time_sensitivity",
+        "description": "Is the problem time-sensitive or urgent, requiring immediate attention and action?"
+    },
+    {
+        "id": 33,
+        "name": "typical_solutions",
+        "description": "What kinds of solution typically are produced for this kind of problem specification?"
+    },
+    {
+        "id": 34,
+        "name": "alternative_solutions",
+        "description": "Given the problem specification and the current best solution, have a guess about other possible solutions."
+    },
+    {
+        "id": 35,
+        "name": "radical_rethinking",
+        "description": "Let's imagine the current best solution is totally wrong, what other ways are there to think about the problem specification?"
+    },
+    {
+        "id": 36,
+        "name": "solution_modification",
+        "description": "What is the best way to modify this current best solution, given what you know about these kinds of problem specification?"
+    },
+    {
+        "id": 37,
+        "name": "novel_solution",
+        "description": "Ignoring the current best solution, create an entirely new solution to the problem."
+    },
+    {
+        "id": 38,
+        "name": "step_by_step",
+        "description": "Let's think step by step."
+    },
+    {
+        "id": 39,
+        "name": "step_by_step_plan",
+        "description": "Let's make a step by step plan and implement it with good notion and explanation."
+    }
+]
+
+def get_all_modules():
+    """Return all 39 reasoning modules."""
+    return REASONING_MODULES
+
+def get_modules_by_category():
+    """Categorize modules by their primary focus."""
+    categories = {
+        "analytical": [1, 3, 5, 10, 14, 17, 20, 23, 24, 25, 29],
+        "creative": [2, 4, 11, 30, 34, 35, 37],
+        "systematic": [9, 13, 16, 18, 22, 31, 33, 36, 38, 39],
+        "collaborative": [7, 12, 15, 21],
+        "risk_oriented": [6, 8, 14, 19],
+        "behavioral": [27, 28],
+        "constraint_focused": [26, 32]
+    }
+    
+    return {
+        category: [REASONING_MODULES[i-1] for i in indices]
+        for category, indices in categories.items()
+    }
+
+def get_modules_by_ids(module_ids):
+    """Get specific modules by their IDs."""
+    return [module for module in REASONING_MODULES if module["id"] in module_ids]
+
+def get_module_descriptions():
+    """Get just the descriptions for prompting."""
+    return [f"{module['name']}: {module['description']}" for module in REASONING_MODULES]