inspect-ai 0.3.84__py3-none-any.whl → 0.3.85__py3-none-any.whl

inspect_ai/agent/_agent.py

@@ -225,16 +225,12 @@ def agent_with(
         name = name or info.name
         description = description or info.metadata.get(AGENT_DESCRIPTION, None)
 
-    # if the name is null then raise
-    if name is None:
-        raise ValueError("You must provide a name to agent_with")
-
     # now set registry info
     set_registry_info(
         agent,
         RegistryInfo(
             type="agent",
-            name=name,
+            name=name or "agent",
             metadata={AGENT_DESCRIPTION: description}
             if description is not None
             else {},

inspect_ai/model/_providers/providers.py

@@ -253,28 +253,6 @@ def none() -> type[ModelAPI]:
     return NoModel
 
 
-@modelapi("goodfire")
-def goodfire() -> type[ModelAPI]:
-    """Get the Goodfire API provider."""
-    FEATURE = "Goodfire API"
-    PACKAGE = "goodfire"
-    MIN_VERSION = "0.3.4"  # Support for newer Llama models and OpenAI compatibility
-
-    # verify we have the package
-    try:
-        import goodfire  # noqa: F401
-    except ImportError:
-        raise pip_dependency_error(FEATURE, [PACKAGE])
-
-    # verify version
-    verify_required_version(FEATURE, PACKAGE, MIN_VERSION)
-
-    # in the clear
-    from .goodfire import GoodfireAPI
-
-    return GoodfireAPI
-
-
 def validate_openai_client(feature: str) -> None:
     FEATURE = feature
     PACKAGE = "openai"

{inspect_ai-0.3.84.dist-info → inspect_ai-0.3.85.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inspect_ai
-Version: 0.3.84
+Version: 0.3.85
 Summary: Framework for large language model evaluations
 Author: UK AI Security Institute
 License: MIT License
@@ -56,7 +56,6 @@ Requires-Dist: aioboto3; extra == "dev"
 Requires-Dist: azure-ai-inference; extra == "dev"
 Requires-Dist: google-cloud-aiplatform; extra == "dev"
 Requires-Dist: google-genai; extra == "dev"
-Requires-Dist: goodfire; extra == "dev"
 Requires-Dist: griffe; extra == "dev"
 Requires-Dist: groq; extra == "dev"
 Requires-Dist: ipython; extra == "dev"

{inspect_ai-0.3.84.dist-info → inspect_ai-0.3.85.dist-info}/RECORD

@@ -450,7 +450,7 @@ inspect_ai/_view/www/src/workspace/tabs/SamplesTab.tsx,sha256=s6jt1-5_Hrgz3_ysT1
 inspect_ai/_view/www/src/workspace/tabs/grouping.ts,sha256=6lvFzReQKQ_43S20xN4kfBJN2F7Tfs2VWeSMIuHxUAI,6187
 inspect_ai/_view/www/src/workspace/tabs/types.ts,sha256=Wa1Y4tZwYO_QJr0Tg9-5xJFztmcMYCODSm6JvdzMpDw,471
 inspect_ai/agent/__init__.py,sha256=nzL9TPAARSJVZRPogWHxZ-qJriXBGmFUM9DV4NRi21o,749
-inspect_ai/agent/_agent.py,sha256=g0hw6sDTXg_4NRjs5Ohze404HzyyIyFNFDvlgGqL2Vw,7736
+inspect_ai/agent/_agent.py,sha256=5MXMrY5bsBQ4AI5y5rVaIYywp7JxQzUnhp-KqBIMF7I,7622
 inspect_ai/agent/_as_solver.py,sha256=_6H0L9JidC6JjpMaBRBAjIrgzE8GKEoygJjOC_JRoLQ,2340
 inspect_ai/agent/_as_tool.py,sha256=vT5hrcKfkyP90i4Ieuy_dx4cYsFKOMdPs-6x12cuqMk,4449
 inspect_ai/agent/_filter.py,sha256=qnT0HbT4edpDi0MwXY3Q3It2pzNRkTRXZDOqfCwMY6M,1234
@@ -547,7 +547,6 @@ inspect_ai/model/_providers/anthropic.py,sha256=PYxV0D_bt0Icp2wEWb6GMCpDb-uBFKYy
 inspect_ai/model/_providers/azureai.py,sha256=uXED_qmeyW1XAGBosbG7PJNk833RIeokKX3l_8O9gYA,14341
 inspect_ai/model/_providers/bedrock.py,sha256=rh8BvSUPWiFMh0TQwMYTlucfFrDKswtLhzozulrz7wE,24004
 inspect_ai/model/_providers/cloudflare.py,sha256=mWqBqc0zzf29UWz34biq8CxSu99a95YjpH_6A4na52g,4617
-inspect_ai/model/_providers/goodfire.py,sha256=J0nxGbF8lXBmc5YHBJCsZdF03mWT5SuWMb21d9ho3FM,8799
 inspect_ai/model/_providers/google.py,sha256=gcg8pvYAV5gYc4NXC5mLqFyuU7KuhyNrzdXIY57sYl8,28207
 inspect_ai/model/_providers/grok.py,sha256=dS88ueXiD-kHAFr0jCoTpTGLGa2VsUlB_TFP8L_2lBM,995
 inspect_ai/model/_providers/groq.py,sha256=mcRKu33e-mO5l06PGV6SjsildQd0XCti6QNXwwFWL7I,11246
@@ -561,7 +560,7 @@ inspect_ai/model/_providers/openai.py,sha256=NFdMpnI2vlmpI8h_vWnt8y4X_XaydaL9gH5
 inspect_ai/model/_providers/openai_o1.py,sha256=k-Xm_Wzn1KHKL6Z1KTHg4CTTr8ybgiHvXkLiLdjP7Os,12926
 inspect_ai/model/_providers/openai_responses.py,sha256=YPXt8KQfIEiiTpvtoQECBoNQLDLbwBW_KhBfM8vEhJk,6324
 inspect_ai/model/_providers/openrouter.py,sha256=pDimDmm_4FzS4GZx0n9z8z717mQf3IQlgEy30huzpc4,4730
-inspect_ai/model/_providers/providers.py,sha256=0WSi_FOWxW71sZ4GJ-OgJqbPS4tMIaPQqEG2hnxqfqc,6378
+inspect_ai/model/_providers/providers.py,sha256=Sd2D9OcWkukuBcl_-KDfdpxMaAShv1JZhL5KfAM87CE,5817
 inspect_ai/model/_providers/together.py,sha256=MoA3tyMKUnE0EekTqEIBBwvsaOp5c697kydLi1ZMYzE,9745
 inspect_ai/model/_providers/vertex.py,sha256=60W7kgoA83GtKdMeJgNU2IAw0N0wTscg4YCcMPu2bwo,17185
 inspect_ai/model/_providers/vllm.py,sha256=UYjCCXzw2hGJHVC3oPl-u2EI4iAm8ZncoIfYp1QJkbQ,14238
@@ -693,9 +692,9 @@ inspect_ai/util/_sandbox/docker/internal.py,sha256=c8X8TLrBPOvsfnq5TkMlb_bzTALyc
 inspect_ai/util/_sandbox/docker/prereqs.py,sha256=0j6_OauBBnVlpBleADcZavIAAQZy4WewVjbRn9c0stg,3355
 inspect_ai/util/_sandbox/docker/service.py,sha256=hhHIWH1VDFLwehdGd19aUBD_VKfDO3GCPxpw1HSwVQk,2437
 inspect_ai/util/_sandbox/docker/util.py,sha256=EeInihCNXgUWxaqZ4dNOJd719kXL2_jr63QCoXn68vA,3154
-inspect_ai-0.3.84.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
-inspect_ai-0.3.84.dist-info/METADATA,sha256=g-2UAMeNEN0cyQB6JUowoPVFebPlFpNsZuFiQwsxpVE,5005
-inspect_ai-0.3.84.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-inspect_ai-0.3.84.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
-inspect_ai-0.3.84.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
-inspect_ai-0.3.84.dist-info/RECORD,,
+inspect_ai-0.3.85.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
+inspect_ai-0.3.85.dist-info/METADATA,sha256=yNjYFfYdXBQnzqg-_uT3awl6p1mr9eOKlJrIOIPtCZs,4965
+inspect_ai-0.3.85.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+inspect_ai-0.3.85.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
+inspect_ai-0.3.85.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
+inspect_ai-0.3.85.dist-info/RECORD,,

inspect_ai/model/_providers/goodfire.py

@@ -1,253 +0,0 @@
-import os
-from typing import Any, List, Literal, get_args
-
-from goodfire import AsyncClient
-from goodfire.api.chat.interfaces import ChatMessage as GoodfireChatMessage
-from goodfire.api.exceptions import (
-    InvalidRequestException,
-    RateLimitException,
-    ServerErrorException,
-)
-from goodfire.variants.variants import SUPPORTED_MODELS, Variant
-from typing_extensions import override
-
-from inspect_ai.tool._tool_choice import ToolChoice
-from inspect_ai.tool._tool_info import ToolInfo
-
-from .._chat_message import (
-    ChatMessage,
-    ChatMessageAssistant,
-    ChatMessageSystem,
-    ChatMessageTool,
-    ChatMessageUser,
-)
-from .._generate_config import GenerateConfig
-from .._model import ModelAPI
-from .._model_call import ModelCall
-from .._model_output import (
-    ChatCompletionChoice,
-    ModelOutput,
-    ModelUsage,
-)
-from .util import environment_prerequisite_error, model_base_url
-
-# Constants
-GOODFIRE_API_KEY = "GOODFIRE_API_KEY"
-DEFAULT_BASE_URL = "https://api.goodfire.ai"
-DEFAULT_MAX_TOKENS = 4096
-DEFAULT_TEMPERATURE = 1.0  # Standard sampling temperature (baseline)
-DEFAULT_TOP_P = 1.0  # No nucleus sampling truncation (baseline)
-
-
-class GoodfireAPI(ModelAPI):
-    """Goodfire API provider.
-
-    This provider implements the Goodfire API for LLM inference. It supports:
-    - Chat completions with standard message formats
-    - Basic parameter controls (temperature, top_p, etc.)
-    - Usage statistics tracking
-    - Stop reason handling
-
-    Does not currently support:
-    - Tool calls
-    - Feature analysis
-    - Streaming responses
-
-    Known limitations:
-    - Limited role support (system/user/assistant only)
-    - Tool messages converted to user messages
-    """
-
-    client: AsyncClient
-    variant: Variant
-    model_args: dict[str, Any]
-
-    def __init__(
-        self,
-        model_name: str,
-        base_url: str | None = None,
-        api_key: str | None = None,
-        config: GenerateConfig = GenerateConfig(),
-        **model_args: Any,
-    ) -> None:
-        """Initialize the Goodfire API provider.
-
-        Args:
-            model_name: Name of the model to use
-            base_url: Optional custom API base URL
-            api_key: Optional API key (will check env vars if not provided)
-            config: Generation config options
-            **model_args: Additional arguments passed to the API
-        """
-        super().__init__(
-            model_name=model_name,
-            base_url=base_url,
-            api_key=api_key,
-            api_key_vars=[GOODFIRE_API_KEY],
-            config=config,
-        )
-
-        # resolve api_key
-        if not self.api_key:
-            self.api_key = os.environ.get(GOODFIRE_API_KEY)
-            if not self.api_key:
-                raise environment_prerequisite_error("Goodfire", GOODFIRE_API_KEY)
-
-        # Validate model name against supported models
-        supported_models = list(get_args(SUPPORTED_MODELS))
-        if self.model_name not in supported_models:
-            raise ValueError(
-                f"Model {self.model_name} not supported. Supported models: {supported_models}"
-            )
-
-        # Initialize client with minimal configuration
-        base_url_val = model_base_url(base_url, "GOODFIRE_BASE_URL")
-        assert isinstance(base_url_val, str) or base_url_val is None
-
-        # Store model args for use in generate
-        self.model_args = model_args
-
-        self.client = AsyncClient(
-            api_key=self.api_key,
-            base_url=base_url_val or DEFAULT_BASE_URL,
-        )
-
-        # Initialize variant directly with model name
-        self.variant = Variant(self.model_name)  # type: ignore
-
-    def _to_goodfire_message(self, message: ChatMessage) -> GoodfireChatMessage:
-        """Convert an Inspect message to a Goodfire message format.
-
-        Args:
-            message: The message to convert
-
-        Returns:
-            The converted message in Goodfire format
-
-        Raises:
-            ValueError: If the message type is unknown
-        """
-        role: Literal["system", "user", "assistant"] = "user"
-        if isinstance(message, ChatMessageSystem):
-            role = "system"
-        elif isinstance(message, ChatMessageUser):
-            role = "user"
-        elif isinstance(message, ChatMessageAssistant):
-            role = "assistant"
-        elif isinstance(message, ChatMessageTool):
-            role = "user"  # Convert tool messages to user messages
-        else:
-            raise ValueError(f"Unknown message type: {type(message)}")
-
-        content = str(message.content)
-        if isinstance(message, ChatMessageTool):
-            content = f"Tool {message.function}: {content}"
-
-        return GoodfireChatMessage(role=role, content=content)
-
-    def handle_error(self, ex: Exception) -> ModelOutput | Exception:
-        """Handle only errors that need special treatment for retry logic or model limits."""
-        # Handle token/context length errors
-        if isinstance(ex, InvalidRequestException):
-            error_msg = str(ex).lower()
-            if "context length" in error_msg or "max tokens" in error_msg:
-                return ModelOutput.from_content(
-                    model=self.model_name,
-                    content=str(ex),
-                    stop_reason="model_length",
-                    error=error_msg,
-                )
-
-        # Let all other errors propagate
-        return ex
-
-    @override
-    def should_retry(self, ex: Exception) -> bool:
-        """Check if exception is due to rate limiting."""
-        return isinstance(ex, RateLimitException | ServerErrorException)
-
-    @override
-    def connection_key(self) -> str:
-        """Return key for connection pooling."""
-        return f"goodfire:{self.api_key}"
-
-    @override
-    def max_tokens(self) -> int | None:
-        """Return maximum tokens supported by model."""
-        return DEFAULT_MAX_TOKENS  # Let Goodfire's Variant handle model-specific limits
-
-    async def generate(
-        self,
-        input: List[ChatMessage],
-        tools: List[ToolInfo],
-        tool_choice: ToolChoice,
-        config: GenerateConfig,
-        *,
-        cache: bool = True,
-    ) -> tuple[ModelOutput | Exception, ModelCall]:
-        """Generate output from the model."""
-        # Convert messages and prepare request params
-        messages = [self._to_goodfire_message(msg) for msg in input]
-        # Build request parameters with type hints
-        params: dict[str, Any] = {
-            "model": self.variant.base_model,  # Use base_model instead of stringifying the Variant
-            "messages": messages,
-            "max_completion_tokens": int(config.max_tokens)
-            if config.max_tokens
-            else DEFAULT_MAX_TOKENS,
-            "stream": False,
-        }
-
-        # Add generation parameters from config if not in model_args
-        if "temperature" not in self.model_args and config.temperature is not None:
-            params["temperature"] = float(config.temperature)
-        elif "temperature" not in self.model_args:
-            params["temperature"] = DEFAULT_TEMPERATURE
-
-        if "top_p" not in self.model_args and config.top_p is not None:
-            params["top_p"] = float(config.top_p)
-        elif "top_p" not in self.model_args:
-            params["top_p"] = DEFAULT_TOP_P
-
-        # Add any additional model args (highest priority)
-        api_params = {
-            k: v
-            for k, v in self.model_args.items()
-            if k not in ["api_key", "base_url", "model_args"]
-        }
-        params.update(api_params)
-
-        try:
-            # Use native async client
-            response = await self.client.chat.completions.create(**params)
-            response_dict = response.model_dump()
-
-            output = ModelOutput(
-                model=self.model_name,
-                choices=[
-                    ChatCompletionChoice(
-                        message=ChatMessageAssistant(
-                            content=response_dict["choices"][0]["message"]["content"],
-                            model=self.model_name,
-                        ),
-                        stop_reason="stop",
-                    )
-                ],
-                usage=ModelUsage(**response_dict["usage"])
-                if "usage" in response_dict
-                else None,
-            )
-            model_call = ModelCall.create(request=params, response=response_dict)
-            return (output, model_call)
-        except Exception as ex:
-            result = self.handle_error(ex)
-            model_call = ModelCall.create(
-                request=params,
-                response={},  # Empty response for error case
-            )
-            return (result, model_call)
-
-    @property
-    def name(self) -> str:
-        """Get provider name."""
-        return "goodfire"