inspect-ai 0.3.51__py3-none-any.whl → 0.3.53__py3-none-any.whl

inspect_ai/_view/www/src/samples/SampleDisplay.mjs

@@ -350,7 +350,17 @@ const metadataViewsForSample = (id, sample) => {
   return sampleMetadatas;
 };
 
-const SampleSummary = ({ id, sample, style, sampleDescriptor }) => {
+/**
+ * Component to display a sample with relevant context and visibility control.
+ *
+ * @param {Object} props - The properties passed to the component.
+ * @param {string} props.parent_id - The id of the parent com
+ * @param {import("../types/log").EvalSample} [props.sample] - the sample
+ * @param {Object} [props.style] - Inline styles for the table element.
+ * @param {import("../samples/SamplesDescriptor.mjs").SamplesDescriptor} props.sampleDescriptor - the sample descriptor
+ * @returns {import("preact").JSX.Element} The TranscriptView component.
+ */
+const SampleSummary = ({ parent_id, sample, style, sampleDescriptor }) => {
   const input =
     sampleDescriptor?.messageShape.normalized.input > 0
       ? Math.max(0.15, sampleDescriptor.messageShape.normalized.input)
@@ -386,7 +396,7 @@ const SampleSummary = ({ id, sample, style, sampleDescriptor }) => {
   const columns = [];
   columns.push({
     label: "Id",
-    value: id,
+    value: sample.id,
     size: `${idSize}em`,
   });
 
@@ -412,7 +422,8 @@ const SampleSummary = ({ id, sample, style, sampleDescriptor }) => {
 
   const fullAnswer =
     sample && sampleDescriptor
-      ? sampleDescriptor.selectedScorer(sample).answer()
+      ? // @ts-ignore
+        sampleDescriptor.selectedScorer(sample).answer()
       : undefined;
   if (fullAnswer) {
     columns.push({
@@ -445,14 +456,16 @@ const SampleSummary = ({ id, sample, style, sampleDescriptor }) => {
           message=${sample.error.message}
           style=${{ marginTop: "0.4rem" }}
         />`
-      : sampleDescriptor?.selectedScore(sample).render(),
+      : // TODO: Cleanup once the PR lands which makes sample / sample summary share common interface
+        // @ts-ignore
+        sampleDescriptor?.selectedScore(sample).render(),
     size: "minmax(2em, auto)",
     center: true,
   });
 
   return html`
     <div
-      id=${`sample-heading-${id}`}
+      id=${`sample-heading-${parent_id}`}
       style=${{
         display: "grid",
         gridTemplateColumns: `${columns

inspect_ai/_view/www/src/samples/SampleList.mjs

@@ -145,7 +145,7 @@ export const SampleList = (props) => {
   );
 
   const listStyle = { ...style, flex: "1", overflowY: "auto", outline: "none" };
-  const { limit, answer } = gridColumns(sampleDescriptor);
+  const { limit, answer, target } = gridColumns(sampleDescriptor);
 
   const headerRow = html`<div
     style=${{
@@ -161,7 +161,7 @@ export const SampleList = (props) => {
   >
     <div>Id</div>
     <div>Input</div>
-    <div>Target</div>
+    <div>${target !== "0" ? "Target" : ""}</div>
     <div>${answer !== "0" ? "Answer" : ""}</div>
     <div>${limit !== "0" ? "Limit" : ""}</div>
     <div style=${{ justifySelf: "center" }}>Score</div>

inspect_ai/_view/www/src/samples/transcript/ToolEventView.mjs

@@ -29,10 +29,10 @@ export const ToolEventView = ({ id, event, style, depth }) => {
     return e.event === "approval";
   });
 
-  const title = `Tool: ${event.function}`;
+  const title = `Tool: ${event.view?.title || event.function}`;
   return html`
   <${EventPanel} id=${id} title="${title}" subTitle=${formatDateTime(new Date(event.timestamp))} icon=${ApplicationIcons.solvers.use_tools} style=${style}>  
-  <div name="Summary" style=${{ margin: "0.5em 0" }}>
+  <div name="Summary" style=${{ margin: "0.5em 0", width: "100%" }}>
     <${ToolCallView}
       functionCall=${functionCall}
       input=${input}

inspect_ai/log/_log.py

@@ -37,6 +37,9 @@ class EvalConfig(BaseModel):
     limit: int | tuple[int, int] | None = Field(default=None)
     """Sample limit (number of samples or range of samples)."""
 
+    sample_id: str | int | list[str | int] | None = Field(default=None)
+    """Evaluate specific sample(s)."""
+
     epochs: int | None = Field(default=None)
     """Number of epochs to run samples over."""
 
@@ -76,6 +79,9 @@ class EvalConfig(BaseModel):
     max_subprocesses: int | None = Field(default=None)
     """Maximum number of subprocesses to run concurrently."""
 
+    max_sandboxes: int | None = Field(default=None)
+    """Maximum number of sandboxes to run concurrently."""
+
     sandbox_cleanup: bool | None = Field(default=None)
     """Cleanup sandbox environments after task completes."""
 

inspect_ai/log/_recorders/eval.py

@@ -362,13 +362,14 @@ class ZipLogFile:
                     f"Error occurred during async write to {self._file}: {ex}. Falling back to sync write."
                 )
 
-            # write sync if we need to
-            if not written:
-                with file(self._file, "wb") as f:
-                    f.write(log_bytes)
-
-            # re-open zip file w/ self.temp_file pointer at end
-            self._open()
+            try:
+                # write sync if we need to
+                if not written:
+                    with file(self._file, "wb") as f:
+                        f.write(log_bytes)
+            finally:
+                # re-open zip file w/ self.temp_file pointer at end
+                self._open()
 
     async def close(self) -> EvalLog:
         async with self._lock:

inspect_ai/model/_call_tools.py

@@ -68,10 +68,6 @@ async def call_tools(
             # create a transript for this call
             init_transcript(Transcript(name=call.function))
 
-            # Amend the tool call with a custom view
-            view = tool_call_view(call, tdefs)
-            call.view = view
-
             result: Any = ""
             tool_error: ToolCallError | None = None
             try:
@@ -142,7 +138,7 @@ async def call_tools(
                 arguments=call.arguments,
                 result=content,
                 truncated=truncated,
-                view=view,
+                view=call.view,
                 error=tool_error,
                 events=list(transcript().events),
             )
@@ -163,7 +159,7 @@ async def call_tools(
                 id=call.id,
                 function=call.function,
                 arguments=call.arguments,
-                view=tool_call_view(call, tdefs),
+                view=call.view,
                 pending=True,
             )
             transcript()._event(event)

inspect_ai/model/_generate_config.py

@@ -72,6 +72,9 @@ class GenerateConfigArgs(TypedDict, total=False):
     cache_prompt: Literal["auto"] | bool | None
     """Whether to cache the prompt prefix. Defaults to "auto", which will enable caching for requests with tools. Anthropic only."""
 
+    reasoning_effort: Literal["low", "medium", "high"] | None
+    """Constrains effort on reasoning for reasoning models. Open AI o1 models only."""
+
 
 class GenerateConfig(BaseModel):
     """Base class for model generation configs."""
@@ -139,6 +142,9 @@ class GenerateConfig(BaseModel):
     cache_prompt: Literal["auto"] | bool | None = Field(default=None)
     """Whether to cache the prompt prefix. Defaults to "auto", which will enable caching for requests with tools. Anthropic only."""
 
+    reasoning_effort: Literal["low", "medium", "high"] | None = Field(default=None)
+    """Constrains effort on reasoning for reasoning models. Open AI o1 models only."""
+
     def merge(
         self, other: Union["GenerateConfig", GenerateConfigArgs]
     ) -> "GenerateConfig":

inspect_ai/model/_model.py

@@ -31,11 +31,11 @@ from inspect_ai._util.registry import (
 )
 from inspect_ai._util.retry import log_rate_limit_retry
 from inspect_ai.tool import Tool, ToolChoice, ToolFunction, ToolInfo
-from inspect_ai.tool._tool_def import ToolDef
+from inspect_ai.tool._tool_def import ToolDef, tool_defs
 from inspect_ai.util import concurrency
 
 from ._cache import CacheEntry, CachePolicy, cache_fetch, cache_store
-from ._call_tools import disable_parallel_tools, tools_info
+from ._call_tools import disable_parallel_tools, tool_call_view, tools_info
 from ._chat_message import (
     ChatMessage,
     ChatMessageAssistant,
@@ -248,7 +248,7 @@ class Model:
         async with self._connection_concurrency(config):
             return await self._generate(
                 input=input,
-                tools=tools_info(tools),
+                tools=tools,
                 tool_choice=tool_choice,
                 config=config,
                 cache=cache,
@@ -257,7 +257,10 @@ class Model:
     async def _generate(
         self,
         input: list[ChatMessage],
-        tools: list[ToolInfo],
+        tools: list[Tool]
+        | list[ToolDef]
+        | list[ToolInfo]
+        | list[Tool | ToolDef | ToolInfo],
         tool_choice: ToolChoice | None,
         config: GenerateConfig,
         cache: bool | CachePolicy = False,
@@ -265,6 +268,12 @@ class Model:
         # default to 'auto' for tool_choice (same as underlying model apis)
         tool_choice = tool_choice if tool_choice else "auto"
 
+        # extract tool defs if we can
+        tdefs = tool_defs([tool for tool in tools if not isinstance(tool, ToolInfo)])
+
+        # resolve all tools into tool_info
+        tools = tools_info(tools)
+
         # if we have a specific tool selected then filter out the others
         if isinstance(tool_choice, ToolFunction):
             tools = [tool for tool in tools if tool.name == tool_choice.name]
@@ -374,6 +383,11 @@ class Model:
             # update output with time elapsed
             output.time = time_elapsed
 
+            # add views to tool calls
+            for choice in output.choices:
+                for tool_call in choice.message.tool_calls or []:
+                    tool_call.view = tool_call_view(tool_call, tdefs)
+
             # complete the transcript event
             complete(output, call)
 

inspect_ai/model/_providers/azureai.py

@@ -89,6 +89,19 @@ class AzureAIAPI(ModelAPI):
             config=config,
         )
 
+        # collect known model_args (then delete them so we can pass the rest on)
+        def collect_model_arg(name: str) -> Any | None:
+            nonlocal model_args
+            value = model_args.get(name, None)
+            if value is not None:
+                model_args.pop(name)
+            return value
+
+        emulate_tools = collect_model_arg("emulate_tools")
+        self.emulate_tools = (
+            not not emulate_tools if emulate_tools is not None else None
+        )
+
         # resolve api_key
         if not self.api_key:
             self.api_key = os.environ.get(
@@ -118,8 +131,15 @@ class AzureAIAPI(ModelAPI):
         tool_choice: ToolChoice,
         config: GenerateConfig,
     ) -> ModelOutput | tuple[ModelOutput, ModelCall]:
-        # if its llama then do fake tool calls
-        handler: ChatAPIHandler | None = Llama31Handler() if self.is_llama() else None
+        # emulate tools (auto for llama, opt-in for others)
+        if self.emulate_tools is None and self.is_llama():
+            handler: ChatAPIHandler | None = Llama31Handler()
+        elif self.emulate_tools:
+            handler = Llama31Handler()
+        else:
+            handler = None
+
+        # resolve input
         if handler:
             input = handler.input_with_tools(input, tools)
 

inspect_ai/model/_providers/bedrock.py

@@ -236,15 +236,21 @@ class BedrockAPI(ModelAPI):
         self,
         model_name: str,
         base_url: str | None,
+        api_key: str | None = None,
         config: GenerateConfig = GenerateConfig(),
         **model_args: Any,
     ):
         super().__init__(
             model_name=model_name,
             base_url=model_base_url(base_url, "BEDROCK_BASE_URL"),
+            api_key=api_key,
+            api_key_vars=[],
             config=config,
         )
 
+        # save model_args
+        self.model_args = model_args
+
         # import aioboto3 on demand
         try:
             import aioboto3
@@ -263,6 +269,9 @@ class BedrockAPI(ModelAPI):
 
     @override
     def max_tokens(self) -> int | None:
+        if "llama3-70" in self.model_name or "llama3-8" in self.model_name:
+            return 2048
+
         if "llama3" in self.model_name or "claude3" in self.model_name:
             return 4096
 
@@ -316,6 +325,7 @@ class BedrockAPI(ModelAPI):
                     mode="adaptive",
                 ),
             ),
+            **self.model_args,
         ) as client:
             # Process the tools
             resolved_tools = converse_tools(tools)
@@ -658,6 +668,8 @@ def converse_image_type(type: str) -> ConverseImageFormat:
             return "png"
         case "image/webp":
             return "webp"
+        case "image/jpeg":
+            return "jpeg"
         case _:
             raise ValueError(
                 f"Image mime type {type} is not supported for Bedrock Converse models."
@@ -673,7 +685,11 @@ def converse_tools(tools: list[ToolInfo]) -> list[ConverseTool] | None:
         tool_spec = ConverseToolSpec(
             name=tool.name,
             description=tool.description,
-            inputSchema={"json": tool.parameters.model_dump(exclude_none=True)},
+            inputSchema={
+                "json": tool.parameters.model_dump(
+                    exclude_none=True, exclude={"additionalProperties"}
+                )
+            },
         )
         result.append(ConverseTool(toolSpec=tool_spec))
     return result

inspect_ai/model/_providers/hf.py

@@ -64,7 +64,7 @@ class HuggingFaceAPI(ModelAPI):
         def collect_model_arg(name: str) -> Any | None:
             nonlocal model_args
             value = model_args.get(name, None)
-            if value:
+            if value is not None:
                 model_args.pop(name)
             return value
 

inspect_ai/model/_providers/openai.py

@@ -18,6 +18,7 @@ from openai.types.chat import (
     ChatCompletionContentPartImageParam,
     ChatCompletionContentPartParam,
     ChatCompletionContentPartTextParam,
+    ChatCompletionDeveloperMessageParam,
     ChatCompletionMessage,
     ChatCompletionMessageParam,
     ChatCompletionMessageToolCallParam,
@@ -141,6 +142,18 @@ class OpenAIAPI(ModelAPI):
                 **model_args,
             )
 
+    def is_o1(self) -> bool:
+        return self.model_name.startswith("o1")
+
+    def is_o1_full(self) -> bool:
+        return self.is_o1() and not self.is_o1_mini() and not self.is_o1_preview()
+
+    def is_o1_mini(self) -> bool:
+        return self.model_name.startswith("o1-mini")
+
+    def is_o1_preview(self) -> bool:
+        return self.model_name.startswith("o1-preview")
+
     async def generate(
         self,
         input: list[ChatMessage],
@@ -148,8 +161,8 @@ class OpenAIAPI(ModelAPI):
         tool_choice: ToolChoice,
         config: GenerateConfig,
     ) -> ModelOutput | tuple[ModelOutput, ModelCall]:
-        # short-circuit to call o1- model
-        if self.model_name.startswith("o1-"):
+        # short-circuit to call o1- models that are text only
+        if self.is_o1_preview() or self.is_o1_mini():
             return await generate_o1(
                 client=self.client,
                 input=input,
@@ -179,7 +192,7 @@ class OpenAIAPI(ModelAPI):
 
         # prepare request (we do this so we can log the ModelCall)
         request = dict(
-            messages=await as_openai_chat_messages(input),
+            messages=await as_openai_chat_messages(input, self.is_o1_full()),
             tools=chat_tools(tools) if len(tools) > 0 else NOT_GIVEN,
             tool_choice=chat_tool_choice(tool_choice) if len(tools) > 0 else NOT_GIVEN,
             **self.completion_params(config, len(tools) > 0),
@@ -271,8 +284,10 @@ class OpenAIAPI(ModelAPI):
             params["logprobs"] = config.logprobs
         if config.top_logprobs is not None:
             params["top_logprobs"] = config.top_logprobs
-        if tools and config.parallel_tool_calls is not None:
+        if tools and config.parallel_tool_calls is not None and not self.is_o1():
             params["parallel_tool_calls"] = config.parallel_tool_calls
+        if config.reasoning_effort is not None and self.is_o1_full():
+            params["reasoning_effort"] = config.reasoning_effort
 
         return params
 
@@ -291,14 +306,23 @@ class OpenAIAPI(ModelAPI):
 
 
 async def as_openai_chat_messages(
-    messages: list[ChatMessage],
+    messages: list[ChatMessage], o1_full: bool
 ) -> list[ChatCompletionMessageParam]:
-    return [await openai_chat_message(message) for message in messages]
+    return [await openai_chat_message(message, o1_full) for message in messages]
 
 
-async def openai_chat_message(message: ChatMessage) -> ChatCompletionMessageParam:
+async def openai_chat_message(
+    message: ChatMessage, o1_full: bool
+) -> ChatCompletionMessageParam:
     if message.role == "system":
-        return ChatCompletionSystemMessageParam(role=message.role, content=message.text)
+        if o1_full:
+            return ChatCompletionDeveloperMessageParam(
+                role="developer", content=message.text
+            )
+        else:
+            return ChatCompletionSystemMessageParam(
+                role=message.role, content=message.text
+            )
     elif message.role == "user":
         return ChatCompletionUserMessageParam(
             role=message.role,

inspect_ai/model/_providers/providers.py

@@ -242,7 +242,7 @@ def mockllm() -> type[ModelAPI]:
 def validate_openai_client(feature: str) -> None:
     FEATURE = feature
     PACKAGE = "openai"
-    MIN_VERSION = "1.45.0"
+    MIN_VERSION = "1.58.1"
 
     # verify we have the package
     try:

inspect_ai/model/_providers/vllm.py

@@ -75,7 +75,7 @@ class VLLMAPI(ModelAPI):
         def collect_model_arg(name: str) -> Any | None:
             nonlocal model_args
             value = model_args.get(name, None)
-            if value:
+            if value is not None:
                 model_args.pop(name)
             return value
 

inspect_ai/model/_render.py

@@ -1,8 +1,7 @@
 from rich.console import RenderableType
 
-from inspect_ai._util.format import format_function_call
-from inspect_ai._util.transcript import transcript_markdown
 from inspect_ai.tool._tool_call import ToolCall
+from inspect_ai.tool._tool_transcript import transcript_tool_call
 
 from ._chat_message import ChatMessage, ChatMessageAssistant, ChatMessageTool
 
@@ -17,8 +16,10 @@ def messages_preceding_assistant(messages: list[ChatMessage]) -> list[ChatMessag
     return list(reversed(preceding))
 
 
-def render_tool_calls(tool_calls: list[ToolCall]) -> RenderableType:
-    formatted_calls: list[str] = []
+def render_tool_calls(tool_calls: list[ToolCall]) -> list[RenderableType]:
+    formatted_calls: list[RenderableType] = []
+
     for call in tool_calls:
-        formatted_calls.append(format_function_call(call.function, call.arguments))
-    return transcript_markdown("```python\n" + "\n\n".join(formatted_calls) + "\n```\n")
+        formatted_calls.extend(transcript_tool_call(call))
+
+    return formatted_calls

inspect_ai/model/_trace.py

@@ -42,7 +42,7 @@ def trace_assistant_message(
         # print tool calls
         if message.tool_calls:
             content.append(Text())
-            content.append(render_tool_calls(message.tool_calls))
+            content.extend(render_tool_calls(message.tool_calls))
 
         # print the assistant message
         trace_panel(title="Assistant", content=content)

inspect_ai/solver/_basic_agent.py

@@ -54,6 +54,7 @@ def basic_agent(
     max_attempts: int = 1,
     message_limit: int | None = None,
     token_limit: int | None = None,
+    max_tool_output: int | None = None,
     score_value: ValueToFloat | None = None,
     incorrect_message: str
     | Callable[[TaskState, list[Score]], str] = DEFAULT_INCORRECT_MESSAGE,
@@ -87,6 +88,8 @@ def basic_agent(
           If not specified, will use limit_messages defined for the task. If there is none
           defined for the task, 50 will be used as a default.
        token_limit (int | None): Limit on tokens used in sample before terminating agent.
+       max_tool_output (int | None): Maximum output length (in bytes).
+          Defaults to max_tool_output from active GenerateConfig.
        score_value (ValueToFloat): Function used to extract float from scores (defaults
          to standard value_to_float())
        incorrect_message (str | Callable[[TaskState, list[Score]], str]): User message reply for an
@@ -182,7 +185,9 @@ def basic_agent(
                 # resolve tools calls (if any)
                 if state.output.message.tool_calls:
                     # call tool functions
-                    tool_results = await call_tools(state.output.message, state.tools)
+                    tool_results = await call_tools(
+                        state.output.message, state.tools, max_output=max_tool_output
+                    )
                     state.messages.extend(tool_results)
 
                     # was an answer submitted?
@@ -194,11 +199,13 @@ def basic_agent(
                         # exit if we are at max_attempts
                         attempts += 1
                         if attempts >= max_attempts:
+                            state.completed = True
                             break
 
                         # exit if the submission is successful
                         answer_scores = await score(state)
                         if score_value_fn(answer_scores[0].value) == 1.0:
+                            state.completed = True
                             break
 
                         # otherwise notify the model that it was incorrect and continue

inspect_ai/tool/_tool_transcript.py

@@ -0,0 +1,28 @@
+from pydantic import JsonValue
+from rich.console import RenderableType
+from rich.text import Text
+from typing_extensions import Protocol
+
+from inspect_ai._util.transcript import transcript_function, transcript_markdown
+
+from ._tool_call import ToolCallContent
+
+
+class TranscriptToolCall(Protocol):
+    function: str
+    arguments: dict[str, JsonValue]
+    view: ToolCallContent | None
+
+
+def transcript_tool_call(call: TranscriptToolCall) -> list[RenderableType]:
+    content: list[RenderableType] = []
+    if call.view:
+        if call.view.title:
+            content.append(Text.from_markup(f"[bold]{call.view.title}[/bold]\n"))
+        if call.view.format == "markdown":
+            content.append(transcript_markdown(call.view.content))
+        else:
+            content.append(call.view.content)
+    else:
+        content.append(transcript_function(call.function, call.arguments))
+    return content

inspect_ai/util/_sandbox/context.py

@@ -109,7 +109,7 @@ def raise_no_sandbox() -> NoReturn:
 
 
 async def init_sandbox_environments_sample(
-    type: str,
+    sandboxenv_type: type[SandboxEnvironment],
     task_name: str,
     config: SandboxEnvironmentConfigType | None,
     files: dict[str, bytes],
@@ -117,7 +117,6 @@ async def init_sandbox_environments_sample(
     metadata: dict[str, Any],
 ) -> dict[str, SandboxEnvironment]:
     # get setup and cleanup functions
-    sandboxenv_type = registry_find_sandboxenv(type)
     sample_init = cast(SampleInit, getattr(sandboxenv_type, "sample_init"))
     sample_cleanup = cast(SampleCleanup, getattr(sandboxenv_type, "sample_cleanup"))
 

inspect_ai/util/_sandbox/docker/config.py

@@ -2,8 +2,6 @@ import os
 from logging import getLogger
 from pathlib import Path
 
-import aiofiles
-
 logger = getLogger(__name__)
 
 
@@ -17,7 +15,7 @@ CONFIG_FILES = [
 DOCKERFILE = "Dockerfile"
 
 
-async def resolve_compose_file(parent: str = "") -> str:
+def resolve_compose_file(parent: str = "") -> str:
     # existing compose file provides all the config we need
     compose = find_compose_file(parent)
     if compose is not None:
@@ -29,11 +27,11 @@ async def resolve_compose_file(parent: str = "") -> str:
 
     # dockerfile just needs a compose.yaml synthesized
     elif has_dockerfile(parent):
-        return await auto_compose_file(COMPOSE_DOCKERFILE_YAML, parent)
+        return auto_compose_file(COMPOSE_DOCKERFILE_YAML, parent)
 
     # otherwise provide a generic python container
     else:
-        return await auto_compose_file(COMPOSE_GENERIC_YAML, parent)
+        return auto_compose_file(COMPOSE_GENERIC_YAML, parent)
 
 
 def find_compose_file(parent: str = "") -> str | None:
@@ -59,9 +57,9 @@ def is_auto_compose_file(file: str) -> bool:
     return os.path.basename(file) == AUTO_COMPOSE_YAML
 
 
-async def ensure_auto_compose_file(file: str | None) -> None:
+def ensure_auto_compose_file(file: str | None) -> None:
     if file is not None and is_auto_compose_file(file) and not os.path.exists(file):
-        await resolve_compose_file(os.path.dirname(file))
+        resolve_compose_file(os.path.dirname(file))
 
 
 def safe_cleanup_auto_compose(file: str | None) -> None:
@@ -100,8 +98,8 @@ services:
 """
 
 
-async def auto_compose_file(contents: str, parent: str = "") -> str:
+def auto_compose_file(contents: str, parent: str = "") -> str:
     path = os.path.join(parent, AUTO_COMPOSE_YAML)
-    async with aiofiles.open(path, "w", encoding="utf-8") as f:
-        await f.write(contents)
+    with open(path, "w", encoding="utf-8") as f:
+        f.write(contents)
     return Path(path).resolve().as_posix()