Merge cc03093705 into 40fdce79b3

llama_stack/apis/datatypes.py

@@ -35,6 +35,8 @@ class Api(Enum):
     tool_groups = "tool_groups"
     files = "files"
 
+    synthetic_data_generation = "synthetic_data_generation"
+
     # built-in API
     inspect = "inspect"
 

llama_stack/apis/synthetic_data_generation/providers/__init__.py

@@ -0,0 +1,19 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# SPDX-License-Identifier: MIT
+
+from typing import cast
+
+from llama_stack.apis.synthetic_data_generation import SyntheticDataGeneration
+from llama_stack.providers.utils.resolver import get_provider_impl as _get_provider_impl
+
+
+def get_provider_impl() -> SyntheticDataGeneration:
+    return cast(SyntheticDataGeneration, _get_provider_impl(SyntheticDataGeneration))

llama_stack/providers/inline/synthetic_data_generation/synthetic_data_kit/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.

llama_stack/providers/inline/synthetic_data_generation/synthetic_data_kit/config.py

@@ -0,0 +1,100 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# SPDX-License-Identifier: MIT
+
+import requests
+from typing import Any, Dict, Optional
+from pydantic import BaseModel, Field
+
+from llama_stack.apis.inference import Message
+from llama_stack.apis.synthetic_data_generation import (
+    FilteringFunction,
+    SyntheticDataGeneration,
+    SyntheticDataGenerationResponse,
+)
+
+
+class SyntheticDataKitConfig(BaseModel):
+    """Configuration for the Synthetic Data Kit provider"""
+    llm: Dict[str, Any] = Field(
+        default_factory=lambda: {
+            "provider": "vllm",
+            "model": "meta-llama/Llama-3.2-3B-Instruct",
+        }
+    )
+    vllm: Dict[str, Any] = Field(
+        default_factory=lambda: {
+            "api_base": "http://localhost:8000/v1",
+        }
+    )
+    generation: Dict[str, Any] = Field(
+        default_factory=lambda: {
+            "temperature": 0.7,
+            "chunk_size": 4000,
+            "num_pairs": 25,
+        }
+    )
+    curate: Dict[str, Any] = Field(
+        default_factory=lambda: {
+            "threshold": 7.0,
+            "batch_size": 8,
+        }
+    )
+
+    @classmethod
+    def sample_run_config(cls) -> "SyntheticDataKitConfig":
+        """Create a sample configuration for testing"""
+        return cls()
+
+
+class SyntheticDataKitProvider(SyntheticDataGeneration):
+    def __init__(self, config: SyntheticDataKitConfig):
+        self.config = config
+        self._validate_connection()
+
+    def _validate_connection(self) -> None:
+        """Validate connection to vLLM server"""
+        try:
+            response = requests.get(f"http://localhost:{self.config.vllm['port']}/health")
+            response.raise_for_status()
+        except Exception as e:
+            raise RuntimeError(f"Failed to connect to vLLM server: {e}") from e
+
+    def synthetic_data_generate(
+        self,
+        dialogs: list[Message],
+        filtering_function: FilteringFunction = FilteringFunction.none,
+        model: str | None = None,
+    ) -> SyntheticDataGenerationResponse:
+        # Convert dialogs to SDK format
+        formatted_dialogs = [{"role": dialog.role, "content": dialog.content} for dialog in dialogs]
+
+        payload = {
+            "dialogs": formatted_dialogs,
+            "filtering_function": filtering_function.value,
+            "model": model or self.config.llm["model"],
+            "generation": self.config.generation,
+            "curate": self.config.curate if filtering_function != FilteringFunction.none else None,
+        }
+
+        try:
+            response = requests.post(
+                f"http://localhost:{self.config.vllm['port']}/v1/synthetic-data-generation/generate",
+                json=payload,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+            return SyntheticDataGenerationResponse(
+                synthetic_data=result.get("synthetic_data", []),
+                statistics=result.get("statistics"),
+            )
+        except requests.exceptions.RequestException as e:
+            raise RuntimeError(f"Synthetic data generation failed: {e}") from e

llama_stack/providers/inline/synthetic_data_generation/synthetic_data_kit/synthetic_data_kit.py

@@ -0,0 +1,55 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# SPDX-License-Identifier: MIT
+
+from typing import Optional
+import synthetic_data_kit as sdk
+
+from llama_stack.apis.inference import Message
+from llama_stack.apis.synthetic_data_generation import (
+    FilteringFunction,
+    SyntheticDataGeneration,
+    SyntheticDataGenerationResponse,
+)
+
+from .config import SyntheticDataKitConfig
+
+
+class SyntheticDataKitProvider(SyntheticDataGeneration):
+    def __init__(self, config: SyntheticDataKitConfig):
+        self.config = config
+        self.sdk = sdk.SyntheticDataKit(
+            llm=self.config.llm,
+            vllm=self.config.vllm,
+            generation=self.config.generation,
+            curate=self.config.curate,
+        )
+
+    async def synthetic_data_generate(
+        self,
+        dialogs: list[Message],
+        filtering_function: FilteringFunction = FilteringFunction.none,
+        model: Optional[str] = None,
+    ) -> SyntheticDataGenerationResponse:
+        # Convert dialogs to text format
+        text_content = "\n".join(d.content for d in dialogs)
+        
+        # Generate synthetic data
+        if filtering_function == FilteringFunction.none:
+            result = await self.sdk.create(text_content, type="qa")
+        else:
+            # Generate and then curate
+            generated = await self.sdk.create(text_content, type="qa")
+            result = await self.sdk.curate(generated)
+
+        return SyntheticDataGenerationResponse(
+            synthetic_data=result.get("synthetic_data", []),
+            statistics=result.get("statistics"),
+        )

llama_stack/providers/registry/synthetic_data_generation.py

@@ -0,0 +1,28 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# SPDX-License-Identifier: MIT
+
+from llama_stack.providers.datatypes import Api, InlineProviderSpec, ProviderSpec
+
+
+def available_providers() -> list[ProviderSpec]:
+    return [
+        InlineProviderSpec(
+            api=Api.synthetic_data_generation,
+            provider_type="inline::synthetic_data_kit",
+            pip_packages=[
+                "synthetic-data-kit",
+                "vllm",
+                "pydantic",
+            ],
+            module="llama_stack.providers.inline.synthetic_data_generation.synthetic_data_kit_inline",
+            config_class="llama_stack.providers.inline.synthetic_data_generation.config.SyntheticDataKitConfig",
+        ),
+    ]

tests/integration/providers/inline/synthetic_data_generation/test_synthetic_data_kit_integration.py

@@ -0,0 +1,101 @@
+import os
+import pytest
+from typing import cast
+
+from llama_stack.apis.inference import Message
+from llama_stack.apis.synthetic_data_generation import (
+    SyntheticDataGeneration,
+    FilteringFunction,
+)
+from llama_stack.apis.synthetic_data_generation.providers import get_provider_impl
+from llama_stack.distribution.client import LlamaStackAsLibraryClient
+
+
+@pytest.fixture
+async def client():
+    # Use LlamaStackAsLibraryClient for inline testing
+    return LlamaStackAsLibraryClient()
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_kit_provider_integration(client: LlamaStackAsLibraryClient):
+    provider = await get_provider_impl()
+    assert isinstance(provider, SyntheticDataGeneration)
+
+    # Test single message generation
+    dialogs = [
+        Message(role="user", content="What is artificial intelligence?"),
+    ]
+    
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.none,
+    )
+    
+    assert response.synthetic_data is not None
+    assert len(response.synthetic_data) > 0
+    assert all(isinstance(item, dict) for item in response.synthetic_data)
+    assert all("question" in item and "answer" in item for item in response.synthetic_data)
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_kit_provider_with_filtering(client: LlamaStackAsLibraryClient):
+    provider = await get_provider_impl()
+    
+    # Test generation with filtering
+    dialogs = [
+        Message(role="user", content="Explain quantum computing."),
+        Message(role="assistant", content="Quantum computing uses quantum mechanics..."),
+    ]
+    
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.top_k,
+    )
+    
+    assert response.synthetic_data is not None
+    assert len(response.synthetic_data) > 0
+    assert response.statistics is not None
+    assert "threshold" in response.statistics
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_kit_provider_error_handling(client: LlamaStackAsLibraryClient):
+    provider = await get_provider_impl()
+    
+    # Test with empty dialogs
+    with pytest.raises(ValueError):
+        await provider.synthetic_data_generate(
+            dialogs=[],
+            filtering_function=FilteringFunction.none,
+        )
+    
+    # Test with invalid model
+    with pytest.raises(RuntimeError):
+        await provider.synthetic_data_generate(
+            dialogs=[Message(role="user", content="Test")],
+            filtering_function=FilteringFunction.none,
+            model="invalid-model",
+        )
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_kit_provider_with_env_config(client: LlamaStackAsLibraryClient):
+    # Set environment variables for testing
+    os.environ["SYNTHETIC_DATA_KIT_MODEL"] = "meta-llama/Llama-3.2-7B-Instruct"
+    
+    provider = await get_provider_impl()
+    dialogs = [
+        Message(role="user", content="What is deep learning?"),
+        Message(role="assistant", content="Deep learning is a subset of machine learning..."),
+    ]
+    
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.none,
+    )
+    
+    assert response.synthetic_data is not None
+    assert len(response.synthetic_data) > 0
+    # Clean up environment
+    del os.environ["SYNTHETIC_DATA_KIT_MODEL"] 

tests/unit/providers/inline/synthetic_data_generation/test_synthetic_data_kit.py

@@ -0,0 +1,132 @@
+import os
+import pytest
+from typing import cast
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from llama_stack.apis.inference import Message
+from llama_stack.apis.synthetic_data_generation import (
+    SyntheticDataGeneration,
+    SyntheticDataGenerationResponse,
+    FilteringFunction,
+)
+from llama_stack.providers.inline.synthetic_data_generation.synthetic_data_kit.config import (
+    SyntheticDataKitConfig,
+)
+from llama_stack.providers.inline.synthetic_data_generation.synthetic_data_kit.synthetic_data_kit import (
+    SyntheticDataKitProvider,
+)
+
+
+def test_config_defaults():
+    """Test default configuration values"""
+    config = SyntheticDataKitConfig()
+    assert config.llm["provider"] == "vllm"
+    assert config.llm["model"] == "meta-llama/Llama-3.2-3B-Instruct"
+    assert config.vllm["api_base"] == "http://localhost:8000/v1"
+    assert config.generation["temperature"] == 0.7
+    assert config.generation["chunk_size"] == 4000
+    assert config.curate["threshold"] == 7.0
+
+
+def test_sample_run_config():
+    """Test sample configuration with environment variables"""
+    # Test default configuration
+    config = SyntheticDataKitConfig.sample_run_config()
+    assert isinstance(config, SyntheticDataKitConfig)
+    assert config.llm["model"] == "meta-llama/Llama-3.2-3B-Instruct"
+    
+    # Test environment variable override
+    os.environ["SYNTHETIC_DATA_KIT_MODEL"] = "meta-llama/Llama-3.2-7B-Instruct"
+    config = SyntheticDataKitConfig.sample_run_config()
+    assert config.llm["model"] == "meta-llama/Llama-3.2-7B-Instruct"
+
+
+@pytest.fixture
+def mock_sdk():
+    """Create a mock SDK instance"""
+    with patch("synthetic_data_kit.SyntheticDataKit") as mock:
+        sdk_instance = MagicMock()
+        sdk_instance.create = AsyncMock()
+        sdk_instance.curate = AsyncMock()
+        mock.return_value = sdk_instance
+        yield sdk_instance
+
+
+@pytest.fixture
+def config():
+    return SyntheticDataKitConfig()
+
+
+@pytest.fixture
+def provider(config: SyntheticDataKitConfig, mock_sdk):
+    return SyntheticDataKitProvider(config)
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_generate_basic(provider: SyntheticDataGeneration, mock_sdk):
+    # Setup mock response
+    mock_sdk.create.return_value = {
+        "synthetic_data": [{"question": "What is ML?", "answer": "Machine learning..."}],
+        "statistics": {"count": 1}
+    }
+    
+    dialogs = [Message(role="user", content="What is machine learning?")]
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.none,
+    )
+    
+    # Verify SDK was called correctly
+    mock_sdk.create.assert_called_once_with("What is machine learning?", type="qa")
+    assert isinstance(response, SyntheticDataGenerationResponse)
+    assert len(response.synthetic_data) == 1
+    assert response.statistics == {"count": 1}
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_generate_with_filtering(provider: SyntheticDataGeneration, mock_sdk):
+    # Setup mock responses
+    mock_sdk.create.return_value = {
+        "synthetic_data": [{"question": "What is quantum?", "answer": "Quantum..."}],
+    }
+    mock_sdk.curate.return_value = {
+        "synthetic_data": [{"question": "What is quantum?", "answer": "Quantum..."}],
+        "statistics": {"threshold": 7.5}
+    }
+    
+    dialogs = [Message(role="user", content="Explain quantum computing.")]
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.top_k,
+    )
+    
+    # Verify both create and curate were called
+    mock_sdk.create.assert_called_once_with("Explain quantum computing.", type="qa")
+    mock_sdk.curate.assert_called_once()
+    assert isinstance(response, SyntheticDataGenerationResponse)
+    assert response.statistics["threshold"] == 7.5
+
+
+@pytest.mark.asyncio
+async def test_synthetic_data_generate_multiple_messages(provider: SyntheticDataGeneration, mock_sdk):
+    mock_sdk.create.return_value = {
+        "synthetic_data": [{"question": "What is deep learning?", "answer": "Deep..."}],
+        "statistics": {"count": 1}
+    }
+    
+    dialogs = [
+        Message(role="user", content="What is deep learning?"),
+        Message(role="assistant", content="Deep learning is..."),
+        Message(role="user", content="Can you explain more?")
+    ]
+    
+    response = await provider.synthetic_data_generate(
+        dialogs=dialogs,
+        filtering_function=FilteringFunction.none,
+    )
+    
+    # Verify content was joined correctly
+    expected_content = "What is deep learning?\nDeep learning is...\nCan you explain more?"
+    mock_sdk.create.assert_called_once_with(expected_content, type="qa")
+    assert isinstance(response, SyntheticDataGenerationResponse)
+    assert response.synthetic_data is not None