synth-ai 0.2.19__tar.gz → 0.2.23.dev2__tar.gz

{synth_ai-0.2.19/synth_ai.egg-info → synth_ai-0.2.23.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: synth-ai
-Version: 0.2.19
+Version: 0.2.23.dev2
 Summary: RL as a service SDK - Core AI functionality and tracing
 Author-email: Synth AI <josh@usesynth.ai>
 License-Expression: MIT
@@ -57,6 +57,7 @@ Requires-Dist: modal<2.0.0,>=1.1.4
 Requires-Dist: pyboy>=2.6.0
 Requires-Dist: setuptools>=80.9.0
 Requires-Dist: libsql-experimental>=0.0.55
+Requires-Dist: mcp>=1.21.0
 Provides-Extra: dev
 Requires-Dist: build>=1.2.2.post1; extra == "dev"
 Requires-Dist: twine>=4.0.0; extra == "dev"
@@ -88,9 +89,9 @@ Dynamic: license-file
 
 [![Python](https://img.shields.io/badge/python-3.11+-blue)](https://www.python.org/)
 [![License](https://img.shields.io/badge/license-MIT-green)](LICENSE)
-[![PyPI](https://img.shields.io/badge/PyPI-0.2.10-orange)](https://pypi.org/project/synth-ai/)
-![Coverage](https://img.shields.io/badge/coverage-9.09%25-red)
-![Tests](https://img.shields.io/badge/tests-37%2F38%20passing-brightgreen)
+[![PyPI](https://img.shields.io/badge/PyPI-0.2.22-orange)](https://pypi.org/project/synth-ai/)
+![Coverage](https://img.shields.io/badge/coverage-28.65%25-yellow)
+![Tests](https://img.shields.io/badge/tests-847%20passing-brightgreen)
 ![Blacksmith CI](https://img.shields.io/badge/CI-Blacksmith%20Worker-blue)
 
 > **Synth-AI** — Reinforcement Learning-as-a-Service for agents.  
@@ -98,7 +99,7 @@ Dynamic: license-file
 
 ---
 
-## 🚀 Install version 0.2.16
+## 🚀 Install version 0.2.22
 
 ```bash
 pip install synth-ai
@@ -279,9 +280,85 @@ When you run `uvx synth-ai setup` (or legacy `uvx synth-ai rl_demo setup`):
 
 ---
 
+## 🎯 Prompt Optimization
+
+Automatically optimize prompts for classification, reasoning, and instruction-following tasks using evolutionary algorithms. Synth supports two state-of-the-art algorithms: **GEPA** (Genetic Evolution of Prompt Architectures) and **MIPRO** (Meta-Instruction PROposer).
+
+**References:**
+- **GEPA**: Agrawal et al. (2025). "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning." [arXiv:2507.19457](https://arxiv.org/abs/2507.19457)
+- **MIPRO**: Opsahl-Ong et al. (2024). "Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs." [arXiv:2406.11695](https://arxiv.org/abs/2406.11695)
+
+### How It Works
+
+Prompt optimization uses an **interceptor pattern** that ensures optimized prompts never reach task apps. All prompt modifications happen in the backend via an inference interceptor that substitutes prompts before they reach the LLM.
+
+```
+✅ CORRECT FLOW:
+Backend → register_prompt → Interceptor → substitutes → LLM
+
+❌ WRONG FLOW:
+Backend → prompt_template in payload → Task App (NEVER DO THIS)
+```
+
+### Algorithms
+
+**GEPA (Genetic Evolution of Prompt Architectures)**
+- Population-based evolutionary search
+- LLM-guided mutations for intelligent prompt modifications
+- Pareto optimization balancing performance and prompt length
+- **Best for:** Broad exploration, diverse prompt variants, classification tasks
+- **Results:** Improves accuracy from 60-75% (baseline) to 85-90%+ over 15 generations
+
+**MIPRO (Meta-Instruction PROposer)**
+- Meta-LLM (e.g., GPT-4o-mini) generates instruction variants
+- TPE (Tree-structured Parzen Estimator) guides Bayesian search
+- Bootstrap phase collects few-shot examples from high-scoring seeds
+- **Best for:** Efficient optimization, task-specific improvements, faster convergence
+- **Results:** Achieves similar accuracy gains with fewer evaluations (~96 rollouts vs ~1000 for GEPA)
+
+### Quick Start
+
+1. **Build a prompt evaluation task app**
+   ```bash
+   # Task app evaluates prompt performance (classification accuracy, QA correctness, etc.)
+   ```
+
+2. **Create a prompt learning config**
+   ```toml
+   [prompt_learning]
+   algorithm = "gepa"  # or "mipro"
+   task_app_url = "https://my-task-app.modal.run"
+   
+   [prompt_learning.initial_prompt]
+   messages = [
+     { role = "system", content = "You are a banking assistant..." },
+     { role = "user", pattern = "Customer Query: {query}..." }
+   ]
+   
+   [prompt_learning.gepa]
+   initial_population_size = 20
+   num_generations = 15
+   ```
+
+3. **Launch optimization**
+   ```bash
+   uvx synth-ai train --type prompt_learning --config config.toml
+   ```
+
+4. **Query results**
+   ```python
+   from synth_ai.learning import get_prompt_text
+   best_prompt = get_prompt_text(job_id="pl_abc123", rank=1)
+   ```
+
+**Full documentation:** [Prompt Learning Guide →](https://docs.usesynth.ai/prompt-learning/overview)
+
+---
+
 ## 📚 Documentation
 
 - **SDK Docs:** [https://docs.usesynth.ai/sdk/get-started](https://docs.usesynth.ai/sdk/get-started)
+- **Prompt Learning:** [https://docs.usesynth.ai/prompt-learning/overview](https://docs.usesynth.ai/prompt-learning/overview)
 - **CLI Reference:** [https://docs.usesynth.ai/cli](https://docs.usesynth.ai/cli)
 - **API Reference:** [https://docs.usesynth.ai/api](https://docs.usesynth.ai/api)
 - **Changelog:** [https://docs.usesynth.ai/changelog](https://docs.usesynth.ai/changelog)

{synth_ai-0.2.19 → synth_ai-0.2.23.dev2}/README.md

@@ -2,9 +2,9 @@
 
 [![Python](https://img.shields.io/badge/python-3.11+-blue)](https://www.python.org/)
 [![License](https://img.shields.io/badge/license-MIT-green)](LICENSE)
-[![PyPI](https://img.shields.io/badge/PyPI-0.2.10-orange)](https://pypi.org/project/synth-ai/)
-![Coverage](https://img.shields.io/badge/coverage-9.09%25-red)
-![Tests](https://img.shields.io/badge/tests-37%2F38%20passing-brightgreen)
+[![PyPI](https://img.shields.io/badge/PyPI-0.2.22-orange)](https://pypi.org/project/synth-ai/)
+![Coverage](https://img.shields.io/badge/coverage-28.65%25-yellow)
+![Tests](https://img.shields.io/badge/tests-847%20passing-brightgreen)
 ![Blacksmith CI](https://img.shields.io/badge/CI-Blacksmith%20Worker-blue)
 
 > **Synth-AI** — Reinforcement Learning-as-a-Service for agents.  
@@ -12,7 +12,7 @@
 
 ---
 
-## 🚀 Install version 0.2.16
+## 🚀 Install version 0.2.22
 
 ```bash
 pip install synth-ai
@@ -193,9 +193,85 @@ When you run `uvx synth-ai setup` (or legacy `uvx synth-ai rl_demo setup`):
 
 ---
 
+## 🎯 Prompt Optimization
+
+Automatically optimize prompts for classification, reasoning, and instruction-following tasks using evolutionary algorithms. Synth supports two state-of-the-art algorithms: **GEPA** (Genetic Evolution of Prompt Architectures) and **MIPRO** (Meta-Instruction PROposer).
+
+**References:**
+- **GEPA**: Agrawal et al. (2025). "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning." [arXiv:2507.19457](https://arxiv.org/abs/2507.19457)
+- **MIPRO**: Opsahl-Ong et al. (2024). "Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs." [arXiv:2406.11695](https://arxiv.org/abs/2406.11695)
+
+### How It Works
+
+Prompt optimization uses an **interceptor pattern** that ensures optimized prompts never reach task apps. All prompt modifications happen in the backend via an inference interceptor that substitutes prompts before they reach the LLM.
+
+```
+✅ CORRECT FLOW:
+Backend → register_prompt → Interceptor → substitutes → LLM
+
+❌ WRONG FLOW:
+Backend → prompt_template in payload → Task App (NEVER DO THIS)
+```
+
+### Algorithms
+
+**GEPA (Genetic Evolution of Prompt Architectures)**
+- Population-based evolutionary search
+- LLM-guided mutations for intelligent prompt modifications
+- Pareto optimization balancing performance and prompt length
+- **Best for:** Broad exploration, diverse prompt variants, classification tasks
+- **Results:** Improves accuracy from 60-75% (baseline) to 85-90%+ over 15 generations
+
+**MIPRO (Meta-Instruction PROposer)**
+- Meta-LLM (e.g., GPT-4o-mini) generates instruction variants
+- TPE (Tree-structured Parzen Estimator) guides Bayesian search
+- Bootstrap phase collects few-shot examples from high-scoring seeds
+- **Best for:** Efficient optimization, task-specific improvements, faster convergence
+- **Results:** Achieves similar accuracy gains with fewer evaluations (~96 rollouts vs ~1000 for GEPA)
+
+### Quick Start
+
+1. **Build a prompt evaluation task app**
+   ```bash
+   # Task app evaluates prompt performance (classification accuracy, QA correctness, etc.)
+   ```
+
+2. **Create a prompt learning config**
+   ```toml
+   [prompt_learning]
+   algorithm = "gepa"  # or "mipro"
+   task_app_url = "https://my-task-app.modal.run"
+   
+   [prompt_learning.initial_prompt]
+   messages = [
+     { role = "system", content = "You are a banking assistant..." },
+     { role = "user", pattern = "Customer Query: {query}..." }
+   ]
+   
+   [prompt_learning.gepa]
+   initial_population_size = 20
+   num_generations = 15
+   ```
+
+3. **Launch optimization**
+   ```bash
+   uvx synth-ai train --type prompt_learning --config config.toml
+   ```
+
+4. **Query results**
+   ```python
+   from synth_ai.learning import get_prompt_text
+   best_prompt = get_prompt_text(job_id="pl_abc123", rank=1)
+   ```
+
+**Full documentation:** [Prompt Learning Guide →](https://docs.usesynth.ai/prompt-learning/overview)
+
+---
+
 ## 📚 Documentation
 
 - **SDK Docs:** [https://docs.usesynth.ai/sdk/get-started](https://docs.usesynth.ai/sdk/get-started)
+- **Prompt Learning:** [https://docs.usesynth.ai/prompt-learning/overview](https://docs.usesynth.ai/prompt-learning/overview)
 - **CLI Reference:** [https://docs.usesynth.ai/cli](https://docs.usesynth.ai/cli)
 - **API Reference:** [https://docs.usesynth.ai/api](https://docs.usesynth.ai/api)
 - **Changelog:** [https://docs.usesynth.ai/changelog](https://docs.usesynth.ai/changelog)

{synth_ai-0.2.19 → synth_ai-0.2.23.dev2}/examples/baseline/banking77_baseline.py

@@ -126,7 +126,46 @@ Valid intents: {', '.join(self.label_names)}"""
                     },
                     headers={"Authorization": f"Bearer {api_key}"} if api_key else {},
                 )
+                
+                # Check for API errors
+                if resp.status_code != 200:
+                    error_msg = f"API call failed with status {resp.status_code}: {resp.text}"
+                    return TaskResult(
+                        seed=seed,
+                        success=False,
+                        outcome_reward=0.0,
+                        total_steps=1,
+                        metadata={
+                            "query": example["text"],
+                            "expected": self.label_names[example["label"]],
+                            "predicted": "",
+                            "correct": False,
+                            "split": split,
+                            "error": error_msg,
+                        },
+                        error=error_msg,
+                    )
+                
                 response = resp.json()
+                
+                # Check for API error in response
+                if "error" in response:
+                    error_msg = f"API error: {response['error']}"
+                    return TaskResult(
+                        seed=seed,
+                        success=False,
+                        outcome_reward=0.0,
+                        total_steps=1,
+                        metadata={
+                            "query": example["text"],
+                            "expected": self.label_names[example["label"]],
+                            "predicted": "",
+                            "correct": False,
+                            "split": split,
+                            "error": error_msg,
+                        },
+                        error=error_msg,
+                    )
         
         # Extract prediction
         predicted_label = ""
@@ -187,7 +226,7 @@ banking77_baseline = BaselineConfig(
         ),
     },
     default_policy_config={
-        "model": "groq:llama-3.1-70b-versatile",
+        "model": "groq:llama-3.3-70b-versatile",  # Updated from decommissioned llama-3.1-70b-versatile
         "temperature": 0.0,
         "max_tokens": 128,
     },

synth_ai-0.2.23.dev2/examples/baseline/banking77_pipeline_baseline.py

@@ -0,0 +1,294 @@
+"""Baseline evaluation for the Banking77 two-step pipeline."""
+
+from __future__ import annotations
+
+import json
+import os
+from typing import Any, Dict, Iterable, Mapping
+
+import httpx
+from datasets import load_dataset
+
+from synth_ai.baseline import BaselineConfig, BaselineTaskRunner, DataSplit, TaskResult
+from synth_ai.inference.client import InferenceClient
+
+
+TOOL_NAME = "banking77_classify"
+_dataset = None
+_label_names: list[str] | None = None
+
+
+def _load_dataset():
+    global _dataset, _label_names
+    if _dataset is None:
+        try:
+            _dataset = load_dataset("PolyAI/banking77")
+        except Exception:
+            _dataset = load_dataset("banking77")
+        _label_names = _dataset["train"].features["label"].names
+    return _dataset, _label_names
+
+
+def _format_available_intents() -> str:
+    _, label_names = _load_dataset()
+    label_names = label_names or []
+    return "\n".join(f"{i + 1}. {label}" for i, label in enumerate(label_names))
+
+
+CLASSIFIER_SYSTEM_PROMPT = (
+    "You are an expert banking assistant. Classify the customer query into one of the "
+    "known Banking77 intents. Always respond using the `banking77_classify` tool."
+)
+
+CLASSIFIER_USER_TEMPLATE = (
+    "Customer Query: {query}\n\n"
+    "Classify this query into one of the banking intents using the tool call."
+)
+
+CALIBRATOR_SYSTEM_PROMPT = (
+    "You refine intent predictions from an upstream classifier. Review the suggested "
+    "intent alongside the original query. If the suggestion is valid, confirm it. "
+    "Otherwise, choose the closest Banking77 intent. Always respond via the "
+    "`banking77_classify` tool with the final label."
+)
+
+CALIBRATOR_USER_TEMPLATE = (
+    "Original Customer Query: {query}\n"
+    "Classifier Suggested Intent: {candidate_intent}\n\n"
+    "Return the best final intent via the tool call."
+)
+
+
+class Banking77PipelineTaskRunner(BaselineTaskRunner):
+    """Baseline runner that evaluates the two-step pipeline locally."""
+
+    def __init__(self, policy_config: Dict[str, Any], env_config: Dict[str, Any]):
+        super().__init__(policy_config, env_config)
+
+        self.dataset, self.label_names = _load_dataset()
+        self.model = policy_config["model"]
+        self.temperature = policy_config.get("temperature", 0.0)
+        self.max_tokens = policy_config.get("max_tokens", 128)
+        self.inference_url = policy_config.get("inference_url")
+
+        self.tool_schema = {
+            "type": "function",
+            "function": {
+                "name": TOOL_NAME,
+                "description": "Return the predicted banking77 intent label in the `intent` field.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "intent": {
+                            "type": "string",
+                            "description": "Predicted Banking77 intent label.",
+                        }
+                    },
+                    "required": ["intent"],
+                },
+            },
+        }
+
+    async def run_task(self, seed: int) -> TaskResult:
+        split = self.env_config.get("split", "train")
+        example = self.dataset[split][seed]
+
+        available_intents = _format_available_intents()
+        query = example["text"]
+
+        classifier_messages = self._build_messages(
+            module="classifier",
+            query=query,
+            available_intents=available_intents,
+            candidate_intent="",
+        )
+        classifier_response = await self._invoke_model(classifier_messages)
+        classifier_intent, classifier_record = self._extract_intent("classifier", classifier_response)
+
+        calibrator_messages = self._build_messages(
+            module="calibrator",
+            query=query,
+            available_intents=available_intents,
+            candidate_intent=classifier_intent,
+        )
+        calibrator_response = await self._invoke_model(calibrator_messages)
+        calibrator_intent, calibrator_record = self._extract_intent("calibrator", calibrator_response)
+
+        final_intent = calibrator_intent or classifier_intent
+        expected_intent = self.label_names[example["label"]]
+        correct = final_intent.lower().replace("_", " ") == expected_intent.lower().replace("_", " ")
+        reward = 1.0 if correct else 0.0
+
+        metadata = {
+            "query": query,
+            "expected_intent": expected_intent,
+            "classifier": classifier_record,
+            "calibrator": calibrator_record,
+            "final_intent": final_intent,
+            "correct": correct,
+            "split": split,
+        }
+
+        return TaskResult(
+            seed=seed,
+            success=True,
+            outcome_reward=reward,
+            total_steps=2,
+            metadata=metadata,
+        )
+
+    def _build_messages(
+        self,
+        *,
+        module: str,
+        query: str,
+        available_intents: str,
+        candidate_intent: str,
+    ) -> list[dict[str, str]]:
+        if module == "classifier":
+            system_prompt = CLASSIFIER_SYSTEM_PROMPT + "\n\nAvailable Intents:\n" + available_intents
+            user_prompt = CLASSIFIER_USER_TEMPLATE.format(query=query)
+        else:
+            system_prompt = CALIBRATOR_SYSTEM_PROMPT + "\n\nAvailable Intents:\n" + available_intents
+            user_prompt = CALIBRATOR_USER_TEMPLATE.format(
+                query=query,
+                candidate_intent=candidate_intent or "<none>",
+            )
+
+        return [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt},
+        ]
+
+    async def _invoke_model(self, messages: list[dict[str, str]]) -> dict[str, Any]:
+        tool_choice = {"type": "function", "function": {"name": TOOL_NAME}}
+
+        if self.inference_url and self.inference_url.startswith("http"):
+            api_key = (
+                os.getenv("SYNTH_API_KEY")
+                or os.getenv("OPENAI_API_KEY")
+                or os.getenv("GROQ_API_KEY")
+                or ""
+            )
+            base_url = self.inference_url.rstrip("/")
+            if not base_url.endswith("/api"):
+                base_url = f"{base_url}/api"
+            client = InferenceClient(base_url=base_url, api_key=api_key)
+            return await client.create_chat_completion(
+                model=self.model,
+                messages=messages,
+                tools=[self.tool_schema],
+                tool_choice=tool_choice,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens,
+            )
+
+        model_name = self.model
+        use_groq = False
+        if model_name.startswith("groq:"):
+            model_name = model_name.split(":", 1)[1]
+            use_groq = True
+
+        api_key = os.getenv("GROQ_API_KEY") if use_groq else os.getenv("OPENAI_API_KEY") or ""
+        base_url = "https://api.groq.com/openai/v1" if use_groq else "https://api.openai.com/v1"
+
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            response = await client.post(
+                f"{base_url}/chat/completions",
+                json={
+                    "model": model_name,
+                    "messages": messages,
+                    "tools": [self.tool_schema],
+                    "tool_choice": tool_choice,
+                    "temperature": self.temperature,
+                    "max_tokens": self.max_tokens,
+                },
+                headers={"Authorization": f"Bearer {api_key}"} if api_key else {},
+            )
+            response.raise_for_status()
+            return response.json()
+
+    def _extract_intent(
+        self,
+        module: str,
+        response: Mapping[str, Any],
+    ) -> tuple[str, Dict[str, Any]]:
+        choices = response.get("choices") or []
+        tool_calls: Iterable[Mapping[str, Any]] = []
+        content_text = ""
+        if choices:
+            message = (choices[0] or {}).get("message", {}) or {}
+            tool_calls = message.get("tool_calls", []) or []
+            content_text = str(message.get("content", ""))
+        intent = self._parse_intent(tool_calls, content_text)
+        if not intent:
+            raise RuntimeError(f"Module {module}: model response missing tool call intent")
+
+        record = {
+            "response": response,
+            "tool_calls": list(tool_calls),
+            "predicted_intent": intent,
+        }
+        return intent, record
+
+    @staticmethod
+    def _parse_intent(
+        tool_calls: Iterable[Mapping[str, Any]],
+        content_text: str,
+    ) -> str:
+        for call in tool_calls:
+            function = (call or {}).get("function", {}) or {}
+            if function.get("name") != TOOL_NAME:
+                continue
+            args_raw = function.get("arguments", "{}")
+            try:
+                args = json.loads(args_raw) if isinstance(args_raw, str) else args_raw
+            except json.JSONDecodeError:
+                continue
+            intent = str(args.get("intent", "")).strip()
+            if intent:
+                return intent
+        if content_text.strip():
+            return content_text.strip().split()[0]
+        return ""
+
+
+_load_dataset()
+banking77_pipeline_baseline = BaselineConfig(
+    baseline_id="banking77_pipeline",
+    name="Banking77 Two-Step Pipeline",
+    description="Two-stage Banking77 classification baseline with classifier and calibrator modules.",
+    task_runner=Banking77PipelineTaskRunner,
+    splits={
+        "train": DataSplit(
+            name="train",
+            seeds=list(range(min(10000, len(_dataset["train"])) if _dataset else 10000)),
+        ),
+        "val": DataSplit(
+            name="val",
+            seeds=list(range(min(1000, len(_dataset["test"])) if _dataset else 1000)),
+        ),
+        "test": DataSplit(
+            name="test",
+            seeds=list(range(min(3000, len(_dataset["test"])) if _dataset else 3000)),
+        ),
+    },
+    default_policy_config={
+        "model": "groq:llama-3.1-70b-versatile",
+        "temperature": 0.0,
+        "max_tokens": 256,
+    },
+    default_env_config={
+        "split": "train",
+    },
+    metadata={
+        "dataset": "PolyAI/banking77",
+        "num_classes": 77,
+        "task_type": "pipeline_classification",
+        "modules": ["classifier", "calibrator"],
+    },
+    tags=["classification", "pipeline", "multi-step"],
+)
+
+
+

{synth_ai-0.2.19 → synth_ai-0.2.23.dev2}/examples/blog_posts/gepa/configs/banking77_gepa_local.toml

@@ -10,23 +10,23 @@ 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."
+pattern = "You are an expert banking assistant. \n\n**Available Banking Intents:**\n{available_intents}\n\n**Task:**\nCall the `banking77_classify` tool with the `intent` parameter set to ONE of the intent labels listed above that best matches the customer query. The intent must be an exact match from the list."
 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."
+pattern = "Customer Query: {query}\n\nClassify this query by calling the tool with the correct intent label from the list above."
 order = 1
 
 [prompt_learning.initial_prompt.wildcards]
 query = "REQUIRED"  # Will be provided by task app at runtime
+available_intents = "OPTIONAL"  # Intent list (numbered 1-77) will be provided by task app
 
 # 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
@@ -42,7 +42,7 @@ proposer_type = "dspy"
 
 # Rollout configuration (mirrors RL [rollout] section)
 [prompt_learning.gepa.rollout]
-budget = 1000
+budget = 100
 max_concurrent = 20
 minibatch_size = 10
 

{synth_ai-0.2.19 → synth_ai-0.2.23.dev2}/examples/blog_posts/gepa/configs/banking77_gepa_test.toml

@@ -49,7 +49,6 @@ query = "REQUIRED"  # Will be provided by task app at runtime
 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
@@ -78,5 +77,4 @@ 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"
 

{synth_ai-0.2.19 → synth_ai-0.2.23.dev2}/examples/blog_posts/gepa/configs/banking77_mipro_local.toml

@@ -32,7 +32,6 @@ 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"
@@ -45,7 +44,6 @@ 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]