synth-ai 0.2.16__py3-none-any.whl → 0.2.19__py3-none-any.whl

examples/blog_posts/gepa/README.md

@@ -0,0 +1,355 @@
+# GEPA: Genetic Evolution for Prompt Optimization
+
+This directory contains examples and configurations for using GEPA (Genetic Evolution for Prompt Optimization) to optimize prompts for various classification and reasoning tasks.
+
+## Overview
+
+**GEPA** is an evolutionary algorithm that optimizes prompts through genetic operations (mutation, crossover, selection) across multiple generations. It's particularly effective for:
+- Intent classification (Banking77)
+- Multi-hop QA (HotpotQA)
+- Instruction following (IFBench)
+- Claim verification (HoVer)
+- Privacy-aware delegation (PUPA)
+
+## Supported Tasks
+
+Configuration files live under `configs/`:
+
+| Task | Description | Config Files |
+|------|-------------|--------------|
+| **Banking77** | Intent classification (77 banking intents) | `banking77_gepa_local.toml`, `banking77_mipro_local.toml` |
+| **HotpotQA** | Multi-hop question answering | `hotpotqa_gepa_local.toml`, `hotpotqa_mipro_local.toml` |
+| **IFBench** | Instruction following benchmark | `ifbench_gepa_local.toml`, `ifbench_mipro_local.toml` |
+| **HoVer** | Claim verification against Wikipedia | `hover_gepa_local.toml`, `hover_mipro_local.toml` |
+| **PUPA** | Privacy-aware task delegation | `pupa_gepa_local.toml`, `pupa_mipro_local.toml` |
+
+Each template targets a different default port (8110–8113) so you can run multiple task apps side-by-side.
+
+---
+
+## Quick Start (Banking77 Example)
+
+### Prerequisites
+
+```bash
+# 1. Install dependencies
+uv pip install -e .
+
+# 2. Set environment variables
+export SYNTH_API_KEY="your-backend-api-key"
+export GROQ_API_KEY="gsk_your_groq_key"
+export ENVIRONMENT_API_KEY="$(python -c 'import secrets; print(secrets.token_urlsafe(32))')"
+```
+
+**Where to get API keys:**
+- **GROQ_API_KEY**: Get from https://console.groq.com/keys
+- **SYNTH_API_KEY**: Get from your backend admin or `.env.dev` file
+- **ENVIRONMENT_API_KEY**: Generate a random secure token (command above)
+
+### Step 1: Start the Backend
+
+```bash
+# Make sure your backend is running
+curl http://localhost:8000/api/health
+# Should return: {"status":"ok"}
+```
+
+### Step 2: Deploy Task App
+
+**Option A: Using helper script (recommended)**
+```bash
+# Terminal 1
+./examples/blog_posts/gepa/deploy_banking77_task_app.sh
+```
+
+**Option B: Using CLI**
+```bash
+uvx synth-ai deploy banking77 --runtime uvicorn --port 8102
+```
+
+**Option C: Deploy to Modal**
+```bash
+uvx synth-ai deploy banking77 --runtime modal --name banking77-gepa --env-file .env
+```
+
+### Step 3: Run GEPA Optimization
+
+**Option A: Using helper script (recommended)**
+```bash
+# Terminal 2
+./examples/blog_posts/gepa/run_gepa_banking77.sh
+```
+
+**Option B: Using CLI directly**
+```bash
+uvx synth-ai train \
+  --config examples/blog_posts/gepa/configs/banking77_gepa_local.toml \
+  --backend http://localhost:8000 \
+  --poll
+```
+
+### Step 4: Monitor Progress
+
+You'll see real-time output like:
+```
+🧬 Running GEPA on Banking77
+=============================
+✅ Backend URL: http://localhost:8000
+✅ Task app is healthy
+
+🚀 Starting GEPA training...
+
+proposal[0] train_accuracy=0.65 len=120 tool_rate=0.95 N=30
+  🔄 TRANSFORMATION:
+    [SYSTEM]: Classify customer banking queries into intents...
+
+Generation 1/15: Best reward=0.75 (75% accuracy)
+Generation 2/15: Best reward=0.82 (82% accuracy)
+...
+✅ GEPA training complete!
+```
+
+Results are automatically saved to `configs/results/gepa_results_<job_id>_<timestamp>.txt`.
+
+---
+
+## Configuration
+
+### Example: Banking77 GEPA Configuration
+
+```toml
+[prompt_learning]
+algorithm = "gepa"
+task_app_url = "http://127.0.0.1:8102"
+task_app_id = "banking77"
+
+# Training seeds (30 seeds from train pool)
+evaluation_seeds = [50, 51, 52, ..., 79]
+
+# Validation seeds (50 seeds from validation pool - not in training)
+validation_seeds = [0, 1, 2, ..., 49]
+
+[prompt_learning.gepa]
+initial_population_size = 20    # Starting population of prompts
+num_generations = 15            # Number of evolutionary cycles
+mutation_rate = 0.3             # Probability of mutation
+crossover_rate = 0.5            # Probability of crossover
+rollout_budget = 1000           # Total rollouts across all generations
+max_concurrent_rollouts = 20    # Parallel rollout limit
+pareto_set_size = 20           # Size of Pareto front
+```
+
+### Key Parameters
+
+| Parameter | Description | Typical Range |
+|-----------|-------------|---------------|
+| `initial_population_size` | Starting number of prompt variants | 10-50 |
+| `num_generations` | Evolutionary cycles to run | 5-30 |
+| `mutation_rate` | Probability of mutating a prompt | 0.1-0.5 |
+| `crossover_rate` | Probability of combining two prompts | 0.3-0.7 |
+| `rollout_budget` | Total task evaluations allowed | 200-2000 |
+| `max_concurrent_rollouts` | Parallel rollout limit | 10-50 |
+| `pareto_set_size` | Multi-objective optimization frontier size | 10-30 |
+
+---
+
+## Querying Results
+
+After GEPA completes, you can query job results programmatically:
+
+### Python API
+
+```python
+from synth_ai.learning import get_prompts, get_prompt_text, get_scoring_summary
+
+# Get all results
+results = get_prompts(
+    job_id="pl_abc123",
+    base_url="http://localhost:8000",
+    api_key="sk_..."
+)
+
+# Access best prompt
+best_prompt = results["best_prompt"]
+best_score = results["best_score"]
+print(f"Best Score: {best_score:.3f}")
+
+# Get top-K prompts
+for prompt_info in results["top_prompts"]:
+    print(f"Rank {prompt_info['rank']}: {prompt_info['train_accuracy']:.3f}")
+    print(prompt_info["full_text"])
+
+# Quick access to best prompt text only
+best_text = get_prompt_text(
+    job_id="pl_abc123",
+    base_url="http://localhost:8000",
+    api_key="sk_...",
+    rank=1  # 1 = best, 2 = second best, etc.
+)
+
+# Get scoring statistics
+summary = get_scoring_summary(
+    job_id="pl_abc123",
+    base_url="http://localhost:8000",
+    api_key="sk_..."
+)
+print(f"Best: {summary['best_train_accuracy']:.3f}")
+print(f"Mean: {summary['mean_train_accuracy']:.3f}")
+print(f"Tried: {summary['num_candidates_tried']}")
+```
+
+### Command Line
+
+```bash
+# Set environment variables
+export BACKEND_BASE_URL="http://localhost:8000"
+export SYNTH_API_KEY="sk_..."
+
+# Run the example script
+python examples/blog_posts/gepa/query_prompts_example.py pl_abc123
+```
+
+### REST API
+
+```bash
+# Get job status
+curl -H "Authorization: Bearer $SYNTH_API_KEY" \
+  http://localhost:8000/api/prompt-learning/online/jobs/JOB_ID
+
+# Stream events
+curl -H "Authorization: Bearer $SYNTH_API_KEY" \
+  http://localhost:8000/api/prompt-learning/online/jobs/JOB_ID/events/stream
+
+# Get metrics
+curl -H "Authorization: Bearer $SYNTH_API_KEY" \
+  http://localhost:8000/api/prompt-learning/online/jobs/JOB_ID/metrics
+```
+
+---
+
+## Expected Results
+
+GEPA typically improves accuracy over generations:
+
+| Generation | Typical Accuracy | Notes |
+|------------|------------------|-------|
+| 1 (baseline) | 60-75% | Initial random/baseline prompts |
+| 5 | 75-80% | Early optimization gains |
+| 10 | 80-85% | Convergence begins |
+| 15 (final) | 85-90%+ | Optimized prompts on Pareto front |
+
+The Pareto front contains multiple prompt variants balancing:
+- **Accuracy** (primary objective)
+- **Token count** (efficiency objective)
+- **Tool call rate** (task-specific objective)
+
+---
+
+## Helper Scripts
+
+| Script | Purpose |
+|--------|---------|
+| `deploy_banking77_task_app.sh` | Start Banking77 task app locally |
+| `run_gepa_banking77.sh` | Run GEPA optimization with validation checks |
+| `test_gepa_local.sh` | Quick test script for local setup |
+| `verify_banking77_setup.sh` | Comprehensive setup verification |
+| `query_prompts_example.py` | Example script for querying results |
+
+---
+
+## Troubleshooting
+
+### ❌ "Banking77 task app is not running"
+
+**Solution:** Start the task app first
+```bash
+./examples/blog_posts/gepa/deploy_banking77_task_app.sh
+```
+
+### ❌ "Cannot connect to backend"
+
+**Solution:** Verify backend is running
+```bash
+curl http://localhost:8000/api/health
+```
+
+If not running, start your backend service.
+
+### ❌ "GROQ_API_KEY environment variable is required"
+
+**Solution:** Export your Groq API key
+```bash
+export GROQ_API_KEY="gsk_your_key_here"
+```
+
+### ❌ "Failed to download dataset"
+
+**Solution:** Check internet connection. The task app downloads from Hugging Face.
+
+If you have the dataset locally:
+```bash
+export BANKING77_DATASET_NAME="/path/to/local/banking77"
+```
+
+### ❌ Pattern validation failed
+
+**Solution:** Ensure your config's `initial_prompt.messages` uses the `{query}` wildcard:
+```toml
+[[prompt_learning.initial_prompt.messages]]
+role = "user"
+pattern = "Customer Query: {query}\n\nClassify this query."
+```
+
+### ⚠️ Metrics not streaming
+
+**Solution:** 
+1. Verify backend `/metrics` endpoint exists
+2. Check SDK `StreamConfig` enables `StreamType.METRICS`
+3. Restart local backend to pick up latest code
+
+---
+
+## Files in This Directory
+
+```
+examples/blog_posts/gepa/
+├── README.md                         # This file - comprehensive guide
+├── configs/                          # Configuration files
+│   ├── banking77_gepa_local.toml    # Banking77 GEPA config
+│   ├── banking77_mipro_local.toml   # Banking77 MIPRO config
+│   ├── hotpotqa_gepa_local.toml     # HotpotQA configs
+│   ├── ifbench_gepa_local.toml      # IFBench configs
+│   ├── hover_gepa_local.toml        # HoVer configs
+│   └── pupa_gepa_local.toml         # PUPA configs
+├── deploy_banking77_task_app.sh     # Helper: Start task app
+├── run_gepa_banking77.sh            # Helper: Run GEPA
+├── test_gepa_local.sh               # Helper: Quick test
+├── verify_banking77_setup.sh        # Helper: Verify setup
+├── (baseline: examples/baseline/banking77_baseline.py)
+├── query_prompts_example.py         # Query results example
+└── task_apps.py                     # Task app registry
+```
+
+---
+
+## Next Steps
+
+1. **Evaluate optimized prompts**: Test best prompts on held-out validation split
+2. **Compare with baseline**: Run `uvx synth-ai baseline banking77` to measure improvement
+3. **Experiment with parameters**: Adjust mutation/crossover rates, population size
+4. **Try MIPRO**: Compare GEPA with MIPROv2 optimization
+5. **Benchmark across tasks**: Test on HotpotQA, IFBench, HoVer, PUPA
+
+---
+
+## Support
+
+For issues or questions:
+
+1. Verify all API keys are set correctly
+2. Check task app: `curl -H "X-API-Key: $ENVIRONMENT_API_KEY" http://127.0.0.1:8102/health`
+3. Check backend: `curl http://localhost:8000/api/health`
+4. Review logs in both terminals for error messages
+5. Run verification script: `./verify_banking77_setup.sh`
+
+Happy optimizing! 🧬🚀

examples/blog_posts/gepa/configs/banking77_gepa_local.toml

@@ -0,0 +1,95 @@
+[prompt_learning]
+algorithm = "gepa"
+task_app_url = "https://synth-laboratories-dev--synth-banking77-web-web.modal.run"
+task_app_id = "banking77"
+
+# Initial prompt pattern (pattern-based mode)
+[prompt_learning.initial_prompt]
+id = "banking77_pattern"
+name = "Banking77 Classification Pattern"
+
+[[prompt_learning.initial_prompt.messages]]
+role = "system"
+pattern = "You are an expert banking assistant that classifies customer queries into banking intents. Given a customer message, respond with exactly one intent label from the provided list using the `banking77_classify` tool."
+order = 0
+
+[[prompt_learning.initial_prompt.messages]]
+role = "user"
+pattern = "Customer Query: {query}\n\nClassify this query into one of the banking intents using the tool call."
+order = 1
+
+[prompt_learning.initial_prompt.wildcards]
+query = "REQUIRED"  # Will be provided by task app at runtime
+
+# Policy configuration (model, provider, etc.)
+[prompt_learning.policy]
+inference_mode = "synth_hosted"
+model = "openai/gpt-oss-20b"
+provider = "groq"
+inference_url = "https://api.groq.com/openai/v1"
+temperature = 0.0
+max_completion_tokens = 512
+policy_name = "banking77-classifier"  # Required for Banking77 task app
+
+# Training split config
+[prompt_learning.env_config]
+pool = "train"
+
+# GEPA-specific configuration with nested subsections (mirrors RL structure)
+[prompt_learning.gepa]
+env_name = "banking77"
+proposer_type = "dspy"
+
+# Rollout configuration (mirrors RL [rollout] section)
+[prompt_learning.gepa.rollout]
+budget = 1000
+max_concurrent = 20
+minibatch_size = 10
+
+# Evaluation configuration (mirrors RL [evaluation] section)
+[prompt_learning.gepa.evaluation]
+seeds = [
+  50, 51, 52, 53, 54, 55, 56, 57, 58, 59,
+  60, 61, 62, 63, 64, 65, 66, 67, 68, 69,
+  70, 71, 72, 73, 74, 75, 76, 77, 78, 79
+]  # Training seeds (30 seeds from train pool)
+validation_seeds = [
+  0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
+  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+  20, 21, 22, 23, 24, 25, 26, 27, 28, 29,
+  30, 31, 32, 33, 34, 35, 36, 37, 38, 39,
+  40, 41, 42, 43, 44, 45, 46, 47, 48, 49
+]  # Held-out validation seeds (50 seeds from validation pool - not in training)
+validation_pool = "validation"
+validation_top_k = 3
+test_pool = [2, 3]  # Test pool for final evaluation (small held-out set)
+
+# Mutation configuration (LLM-guided mutation settings)
+[prompt_learning.gepa.mutation]
+rate = 0.3
+llm_model = "openai/gpt-oss-120b"
+llm_provider = "groq"
+llm_inference_url = "https://api.groq.com/openai/v1"
+
+# Population configuration (evolution parameters)
+[prompt_learning.gepa.population]
+initial_size = 10
+num_generations = 3
+children_per_generation = 12
+crossover_rate = 0.5
+selection_pressure = 1.0
+patience_generations = 3
+
+# Archive configuration (Pareto archive settings)
+[prompt_learning.gepa.archive]
+size = 40
+pareto_set_size = 32
+pareto_eps = 1e-6
+feedback_fraction = 0.5
+
+# Token and budget configuration
+[prompt_learning.gepa.token]
+# max_limit = 1000  # Uncomment to set a token limit
+counting_model = "gpt-4"
+enforce_pattern_limit = true
+# max_spend_usd = 100.0  # Uncomment to set a budget cap

examples/blog_posts/gepa/configs/banking77_gepa_test.toml

@@ -0,0 +1,82 @@
+# GEPA Prompt Learning for Banking77
+# Local backend configuration (localhost:8000)
+
+[prompt_learning]
+algorithm = "gepa"
+task_app_url = "https://synth-laboratories-dev--synth-banking77-web-web.modal.run"
+task_app_id = "banking77"
+evaluation_seeds = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]
+
+# Held-out validation config
+validation_seeds = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
+validation_pool = "validation"
+validation_top_k = 3
+
+# Training split config
+[prompt_learning.env_config]
+pool = "train"
+
+# Seeds for evaluation (increase to score prompts with more rollouts)
+evaluation_seeds = [
+  0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
+  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+  20, 21, 22, 23, 24, 25, 26, 27, 28, 29
+]
+
+# Test pool for final evaluation (held-out episodes)
+test_pool = [10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
+
+# Initial prompt pattern (pattern-based mode)
+[prompt_learning.initial_prompt]
+id = "banking77_pattern"
+name = "Banking77 Classification Pattern"
+
+[[prompt_learning.initial_prompt.messages]]
+role = "system"
+pattern = "You are an expert banking assistant that classifies customer queries into banking intents. Given a customer message, respond with exactly one intent label from the provided list using the `banking77_classify` tool."
+order = 0
+
+[[prompt_learning.initial_prompt.messages]]
+role = "user"
+pattern = "Customer Query: {query}\n\nClassify this query into one of the banking intents using the tool call."
+order = 1
+
+[prompt_learning.initial_prompt.wildcards]
+query = "REQUIRED"  # Will be provided by task app at runtime
+
+# Policy configuration (model, provider, etc.)
+[prompt_learning.policy]
+inference_mode = "synth_hosted"
+model = "openai/gpt-oss-120b"
+provider = "groq"
+inference_url = "https://api.groq.com/openai/v1"
+temperature = 0.0
+max_completion_tokens = 512
+policy_name = "banking77-classifier"  # Required for Banking77 task app
+
+# GEPA-specific configuration
+[prompt_learning.gepa]
+env_name = "banking77"
+initial_population_size = 40
+num_generations = 10
+mutation_rate = 0.3
+crossover_rate = 0.5
+selection_pressure = 1.0
+minibatch_size = 12
+pareto_set_size = 40
+feedback_fraction = 0.5
+children_per_generation = 16
+patience_generations = 5
+rollout_budget = 1500
+archive_size = 30
+pareto_eps = 1e-6
+max_concurrent_rollouts = 20  # Maximum concurrent rollouts across all transformations
+
+# Instruction proposer selection
+proposer_type = "dspy"
+
+# LLM-guided mutation configuration
+mutation_llm_model = "openai/gpt-oss-20b"
+mutation_llm_provider = "groq"
+mutation_llm_inference_url = "https://api.groq.com/openai/v1"
+

examples/blog_posts/gepa/configs/banking77_mipro_local.toml

@@ -0,0 +1,52 @@
+# MIPROv2 Prompt Learning for Banking77
+# Local backend configuration targeting the Banking77 intent classification task app.
+
+[prompt_learning]
+algorithm = "mipro"
+task_app_url = "http://127.0.0.1:8102"
+task_app_id = "banking77"
+
+# Seeds evaluated during optimisation
+evaluation_seeds = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+
+# Held-out seeds for final scoring
+test_pool = [10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
+
+[prompt_learning.initial_prompt]
+id = "banking77_pattern"
+name = "Banking77 Classification Pattern"
+
+[[prompt_learning.initial_prompt.messages]]
+role = "system"
+pattern = "You are an expert banking assistant that classifies customer queries into banking intents. Return only the intent label using the `banking77_classify` tool."
+order = 0
+
+[[prompt_learning.initial_prompt.messages]]
+role = "user"
+pattern = "Customer Query: {query}\n\nClassify this query into one of the banking intents using the tool call."
+order = 1
+
+[prompt_learning.initial_prompt.wildcards]
+query = "REQUIRED"
+
+[prompt_learning.policy]
+model = "openai/gpt-oss-20b"
+provider = "groq"
+inference_url = "https://api.groq.com/openai/v1"
+temperature = 0.0
+max_completion_tokens = 128
+policy_name = "banking77-mipro"
+
+[prompt_learning.mipro]
+env_name = "banking77"
+num_iterations = 16
+num_evaluations_per_iteration = 6
+batch_size = 6
+max_concurrent = 16
+meta_model = "gpt-4.1-mini"
+meta_model_provider = "openai"
+meta_model_inference_url = "https://api.openai.com/v1"
+few_shot_score_threshold = 0.85
+test_pool = [20, 21, 22, 23, 24]
+bootstrap_train_seeds = [0, 1, 2, 3, 4]
+online_pool = [5, 6, 7, 8, 9]

examples/blog_posts/gepa/configs/hotpotqa_gepa_local.toml

@@ -0,0 +1,59 @@
+# GEPA Prompt Learning for HotpotQA
+# Local backend configuration targeting the HotpotQA task app.
+
+[prompt_learning]
+algorithm = "gepa"
+task_app_url = "http://127.0.0.1:8110"
+task_app_id = "hotpotqa"
+
+# Seeds for online evaluation (episode IDs)
+evaluation_seeds = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+
+# Held-out pool used for final evaluation
+test_pool = [10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
+
+[prompt_learning.initial_prompt]
+id = "hotpotqa_chain"
+name = "HotpotQA Multi-Hop Reasoning"
+
+[[prompt_learning.initial_prompt.messages]]
+role = "system"
+pattern = "You are a research assistant that answers multi-hop questions. Use the provided supporting passages to reason out the final answer. Reply with the format:\nAnswer: <short answer>\nSupport: <brief justification referencing the passages>."
+order = 0
+
+[[prompt_learning.initial_prompt.messages]]
+role = "user"
+pattern = "Question: {question}\n\nPassages:\n{context}\n\nProvide the final answer and cite the relevant supporting facts."
+order = 1
+
+[prompt_learning.initial_prompt.wildcards]
+question = "REQUIRED"
+context = "REQUIRED"
+
+[prompt_learning.policy]
+model = "openai/gpt-oss-20b"
+provider = "groq"
+inference_url = "https://api.groq.com/openai/v1"
+temperature = 0.0
+max_completion_tokens = 512
+policy_name = "hotpotqa-gepa"
+
+[prompt_learning.gepa]
+env_name = "hotpotqa"
+initial_population_size = 24
+num_generations = 15
+mutation_rate = 0.35
+crossover_rate = 0.55
+selection_pressure = 1.0
+minibatch_size = 8
+pareto_set_size = 24
+feedback_fraction = 0.5
+children_per_generation = 12
+patience_generations = 5
+rollout_budget = 600
+archive_size = 36
+pareto_eps = 1e-6
+max_concurrent_rollouts = 24
+mutation_llm_model = "openai/gpt-oss-20b"
+mutation_llm_provider = "groq"
+mutation_llm_inference_url = "https://api.groq.com/openai/v1"

examples/blog_posts/gepa/configs/hotpotqa_gepa_qwen.toml

@@ -0,0 +1,36 @@
+[prompt_learning]
+algorithm = "gepa"
+task_app_url = "https://synth-laboratories-dev--synth-banking77-web-web.modal.run" # TODO: replace with HotpotQA task app URL
+task_app_id = "hotpotqa"
+
+# Seeds
+evaluation_seeds = [0,1,2,3,4,5,6,7,8,9]
+
+# Held-out validation
+validation_seeds = [10,11,12,13,14,15,16,17,18,19]
+validation_pool = "validation"
+validation_top_k = 3
+
+# Train split configuration
+[prompt_learning.env_config]
+pool = "train"
+
+# Policy model (synth Qwen via backend inference proxy)
+[prompt_learning.policy]
+provider = "synth"
+model = "Qwen/Qwen3-8B"
+# inference_url will be mapped to backend /api/inference/v1 by the optimizer
+
+# GEPA parameters (tune as needed)
+[prompt_learning.gepa]
+env_name = "hotpotqa"
+initial_population_size = 24
+num_generations = 6
+children_per_generation = 12
+minibatch_size = 10
+pareto_set_size = 32
+rollout_budget = 600
+max_concurrent_rollouts = 16
+mutation_llm_model = "openai/gpt-oss-120b"
+mutation_llm_provider = "groq"
+proposer_type = "dspy"