inspect-ai 0.3.55__py3-none-any.whl → 0.3.56__py3-none-any.whl

inspect_ai/__init__.py

@@ -9,6 +9,7 @@ from inspect_ai._eval.registry import task
 from inspect_ai._eval.score import score, score_async
 from inspect_ai._eval.task import Epochs, Task, TaskInfo, Tasks
 from inspect_ai._util.constants import PKG_NAME
+from inspect_ai.solver._human_agent.agent import human_agent
 
 __version__ = importlib_version(PKG_NAME)
 

inspect_ai/_cli/common.py

@@ -10,7 +10,7 @@ from inspect_ai._util.constants import (
     DEFAULT_LOG_LEVEL,
     DEFAULT_LOG_LEVEL_TRANSCRIPT,
 )
-from inspect_ai._util.display import init_display_type
+from inspect_ai.util._display import init_display_type
 
 
 class CommonOptions(TypedDict):

inspect_ai/_cli/trace.py

@@ -4,7 +4,7 @@ import time
 from datetime import datetime
 from json import dumps
 from pathlib import Path
-from typing import Callable, cast
+from typing import Callable
 
 import click
 from pydantic_core import to_json
@@ -13,8 +13,12 @@ from rich.console import Console, RenderableType
 from rich.table import Column, Table
 
 from inspect_ai._util.error import PrerequisiteError
-from inspect_ai._util.logger import TRACE_FILE_NAME
-from inspect_ai._util.trace import ActionTraceRecord, inspect_trace_dir, read_trace_file
+from inspect_ai._util.trace import (
+    ActionTraceRecord,
+    inspect_trace_dir,
+    list_trace_files,
+    read_trace_file,
+)
 
 
 @click.group("trace")
@@ -36,32 +40,31 @@ def trace_command() -> None:
 )
 def list_command(json: bool) -> None:
     """List all trace files."""
-    trace_dir = inspect_trace_dir()
-    trace_files: list[dict[str, float | str]] = [
-        {"mtime": f.lstat().st_mtime, "file": f.absolute().as_posix()}
-        for f in trace_dir.iterdir()
-        if f.is_file()
-    ]
-    trace_files.sort(key=lambda f: cast(float, f["mtime"]), reverse=True)
+    trace_files = list_trace_files()
     if json:
-        print(dumps(trace_files, indent=2))
+        print(
+            dumps(
+                [dict(file=str(file.file), mtime=file.mtime) for file in trace_files],
+                indent=2,
+            )
+        )
     else:
         table = Table(box=None, show_header=True, pad_edge=False)
         table.add_column("Time")
         table.add_column("Trace File")
         for file in trace_files:
-            mtime = datetime.fromtimestamp(cast(float, file["mtime"])).astimezone()
+            mtime = datetime.fromtimestamp(file.mtime).astimezone()
             table.add_row(
-                mtime.strftime("%d-%b %H:%M:%S %Z"), shlex.quote(str(file["file"]))
+                mtime.strftime("%d-%b %H:%M:%S %Z"), shlex.quote(str(file.file))
             )
         r_print(table)
 
 
 @trace_command.command("dump")
-@click.argument("trace-file", type=str, required=False, default=TRACE_FILE_NAME)
-def read_command(trace_file: str) -> None:
+@click.argument("trace-file", type=str, required=False)
+def dump_command(trace_file: str | None) -> None:
     """Dump a trace file to stdout (as a JSON array of log records)."""
-    trace_file_path = resolve_trace_file_path(trace_file)
+    trace_file_path = _resolve_trace_file_path(trace_file)
 
     traces = read_trace_file(trace_file_path)
     print(
@@ -70,16 +73,16 @@ def read_command(trace_file: str) -> None:
 
 
 @trace_command.command("anomalies")
-@click.argument("trace-file", type=str, required=False, default=TRACE_FILE_NAME)
+@click.argument("trace-file", type=str, required=False)
 @click.option(
     "--all",
     is_flag=True,
     default=False,
     help="Show all anomolies including errors and timeouts (by default only still running and cancelled actions are shown).",
 )
-def anomolies_command(trace_file: str, all: bool) -> None:
+def anomolies_command(trace_file: str | None, all: bool) -> None:
     """Look for anomalies in a trace file (never completed or cancelled actions)."""
-    trace_file_path = resolve_trace_file_path(trace_file)
+    trace_file_path = _resolve_trace_file_path(trace_file)
     traces = read_trace_file(trace_file_path)
 
     # Track started actions
@@ -226,7 +229,17 @@ def _print_bucket(
         print_fn(table)
 
 
-def resolve_trace_file_path(trace_file: str) -> Path:
+def _resolve_trace_file(trace_file: str | None) -> str:
+    if trace_file is None:
+        trace_files = list_trace_files()
+        if len(trace_files) == 0:
+            raise PrerequisiteError("No trace files currently availalble.")
+        trace_file = str(trace_files[0].file)
+    return trace_file
+
+
+def _resolve_trace_file_path(trace_file: str | None) -> Path:
+    trace_file = _resolve_trace_file(trace_file)
     trace_file_path = Path(trace_file)
     if not trace_file_path.is_absolute():
         trace_file_path = inspect_trace_dir() / trace_file_path

inspect_ai/_display/core/active.py

@@ -3,7 +3,7 @@ from contextvars import ContextVar
 
 import rich
 
-from inspect_ai._util.display import display_type
+from inspect_ai.util._display import display_type
 from inspect_ai.util._trace import trace_enabled
 
 from ..rich.display import RichDisplay

inspect_ai/_display/core/display.py

@@ -99,7 +99,7 @@ class TaskScreen(contextlib.AbstractContextManager["TaskScreen"]):
     ) -> Iterator[Console]:
         yield rich.get_console()
 
-    async def input_panel(self, title: str, panel: type[TP]) -> TP:
+    async def input_panel(self, panel_type: type[TP]) -> TP:
         raise NotImplementedError("input_panel not implemented by current display")
 
 

inspect_ai/_display/core/footer.py

@@ -2,8 +2,8 @@ from rich.console import RenderableType
 from rich.text import Text
 
 from inspect_ai._util.logger import http_rate_limit_count
-from inspect_ai._util.throttle import throttle
 from inspect_ai.util._concurrency import concurrency_status
+from inspect_ai.util._throttle import throttle
 
 from .config import task_dict
 

inspect_ai/_display/core/progress.py

@@ -124,12 +124,6 @@ def progress_status_icon(result: TaskResult | None) -> str:
         return f"[{theme.meta}]⠿[{theme.meta}]"
 
 
-def progress_time(time: float) -> str:
-    minutes, seconds = divmod(time, 60)
-    hours, minutes = divmod(minutes, 60)
-    return f"{hours:2.0f}:{minutes:02.0f}:{seconds:02.0f}"
-
-
 def progress_count(complete: int, total: int, width: int | None = None) -> str:
     # Pad the display to keep it stable as the
     # complete metrics

inspect_ai/_display/core/rich.py

@@ -9,9 +9,9 @@ from rich.segment import Segment
 from rich.syntax import Syntax
 from typing_extensions import override
 
-from inspect_ai._util.display import display_type
 from inspect_ai._util.platform import is_running_in_jupyterlab, is_running_in_vscode
 from inspect_ai._util.transcript import transcript_code_theme
+from inspect_ai.util._display import display_type
 
 
 def is_vscode_notebook(console: Console) -> bool:

inspect_ai/_display/rich/display.py

@@ -12,9 +12,9 @@ from rich.table import Table
 from typing_extensions import override
 
 from inspect_ai._util.constants import CONSOLE_DISPLAY_WIDTH
-from inspect_ai._util.display import display_type
-from inspect_ai._util.throttle import throttle
 from inspect_ai.log._transcript import InputEvent, transcript
+from inspect_ai.util._display import display_type
+from inspect_ai.util._throttle import throttle
 from inspect_ai.util._trace import trace_enabled
 
 from ..core.config import task_config

inspect_ai/_display/textual/app.py

@@ -10,6 +10,7 @@ from textual.app import App, ComposeResult
 from textual.binding import Binding, BindingType
 from textual.css.query import NoMatches
 from textual.events import Print
+from textual.widget import Widget
 from textual.widgets import TabbedContent, TabPane
 from textual.widgets.tabbed_content import ContentTabs
 from textual.worker import Worker, WorkerState
@@ -345,13 +346,15 @@ class TaskScreenApp(App[TR]):
             self.update_title()
 
     # dynamic input panels
-    async def add_input_panel(self, title: str, panel: InputPanel) -> None:
+    async def add_input_panel(self, panel: InputPanel) -> None:
         tabs = self.query_one(TabbedContent)
-        await tabs.add_pane(TabPane(title, panel, id=as_input_panel_id(title)))
+        await tabs.add_pane(
+            TabPane(panel.title, panel, id=as_input_panel_id(type(panel)))
+        )
 
-    def get_input_panel(self, title: str) -> InputPanel | None:
+    def get_input_panel(self, panel_type: type) -> InputPanel | None:
         try:
-            tab_pane = self.query_one(f"#{as_input_panel_id(title)}")
+            tab_pane = self.query_one(f"#{as_input_panel_id(panel_type)}")
             if len(tab_pane.children) > 0:
                 return cast(InputPanel, tab_pane.children[0])
             else:
@@ -359,10 +362,6 @@ class TaskScreenApp(App[TR]):
         except NoMatches:
             return None
 
-    async def remove_input_panel(self, title: str) -> None:
-        tabs = self.query_one(TabbedContent)
-        await tabs.remove_pane(as_html_id(as_input_panel_id(title), title))
-
     class InputPanelHost(InputPanel.Host):
         def __init__(self, app: "TaskScreenApp[TR]", tab_id: str) -> None:
             self.app = app
@@ -383,7 +382,7 @@ class TaskScreenApp(App[TR]):
             # the tabs control so the user can switch back w/ the keyboard
             tab_pane = self.app.query_one(f"#{self.tab_id}")
             panel = cast(InputPanel, tab_pane.children[0])
-            for child in panel.children:
+            for child in panel.walk_children(Widget):
                 if child.focusable:
                     child.focus()
                     self.app.query_one(ContentTabs).focus()
@@ -455,19 +454,18 @@ class TextualTaskScreen(TaskScreen, Generic[TR]):
                     console.width = old_width
 
     @override
-    async def input_panel(self, title: str, panel: type[TP]) -> TP:
+    async def input_panel(self, panel_type: type[TP]) -> TP:
         async with self.lock:
-            panel_widget = self.app.get_input_panel(title)
+            panel_widget = self.app.get_input_panel(panel_type)
             if panel_widget is None:
-                panel_widget = panel(
-                    title,
+                panel_widget = panel_type(
                     TaskScreenApp[TR].InputPanelHost(
-                        self.app, as_input_panel_id(title)
+                        self.app, as_input_panel_id(panel_type)
                     ),
                 )
-                await self.app.add_input_panel(title, panel_widget)
+                await self.app.add_input_panel(panel_widget)
             return cast(TP, panel_widget)
 
 
-def as_input_panel_id(title: str) -> str:
-    return as_html_id("id-input-panel", title)
+def as_input_panel_id(panel_type: type) -> str:
+    return as_html_id("id-input-panel", panel_type.__name__)

inspect_ai/_display/textual/widgets/clock.py

@@ -4,7 +4,7 @@ from textual.reactive import reactive
 from textual.timer import Timer
 from textual.widgets import Static
 
-from inspect_ai._display.core.progress import progress_time
+from inspect_ai._util.format import format_progress_time
 
 
 class Clock(Static):
@@ -43,7 +43,7 @@ class Clock(Static):
         if start_time is not None:
             if self.timer is None:
                 self.timer = self.set_interval(self.interval, self.update_time)
-            self.update(progress_time(start_time))
+            self.update(format_progress_time(start_time))
         else:
             self.stop()
 
@@ -52,4 +52,4 @@ class Clock(Static):
             self.time = datetime.now().timestamp() - self.start_time
 
     def watch_time(self, time: float) -> None:
-        self.update(progress_time(time))
+        self.update(format_progress_time(time))

inspect_ai/_display/textual/widgets/samples.py

@@ -22,10 +22,10 @@ from textual.widgets import (
 )
 from textual.widgets.option_list import Option, Separator
 
+from inspect_ai._util.format import format_progress_time
 from inspect_ai._util.registry import registry_unqualified_name
 from inspect_ai.log._samples import ActiveSample
 
-from ...core.progress import progress_time
 from .clock import Clock
 from .transcript import TranscriptView
 
@@ -147,7 +147,9 @@ class SamplesList(OptionList):
             table.add_column(width=1)
             task_name = Text.from_markup(f"{registry_unqualified_name(sample.task)}")
             task_name.truncate(18, overflow="ellipsis", pad=True)
-            task_time = Text.from_markup(f"{progress_time(sample.execution_time)}")
+            task_time = Text.from_markup(
+                f"{format_progress_time(sample.execution_time)}"
+            )
             table.add_row(task_name, task_time, " ")
             sample_id = Text.from_markup(f"id: {sample.sample.id}")
             sample_id.truncate(18, overflow="ellipsis", pad=True)
@@ -308,12 +310,7 @@ class SandboxesView(Vertical):
         yield Vertical(id="sandboxes-list")
 
     async def sync_sample(self, sample: ActiveSample) -> None:
-        sandboxes = sample.sandboxes
-        show_sandboxes = (
-            len([sandbox for sandbox in sandboxes.values() if sandbox.container]) > 0
-        )
-
-        if show_sandboxes:
+        if len(sample.sandboxes) > 0:
             self.display = True
             sandboxes_caption = cast(Static, self.query_one("#sandboxes-caption"))
             sandboxes_caption.update("[bold]sandbox containers:[/bold]")
@@ -321,11 +318,7 @@ class SandboxesView(Vertical):
             sandboxes_list = self.query_one("#sandboxes-list")
             await sandboxes_list.remove_children()
             await sandboxes_list.mount_all(
-                [
-                    Static(sandbox.container)
-                    for sandbox in sandboxes.values()
-                    if sandbox.container
-                ]
+                [Static(sandbox.command) for sandbox in sample.sandboxes.values()]
             )
             sandboxes_list.mount(
                 Static(

inspect_ai/_eval/context.py

@@ -1,7 +1,9 @@
 from inspect_ai._util.dotenv import init_dotenv
 from inspect_ai._util.hooks import init_hooks
 from inspect_ai._util.logger import init_http_rate_limit_count, init_logger
+from inspect_ai.approval._apply import have_tool_approval, init_tool_approval
 from inspect_ai.approval._human.manager import init_human_approval_manager
+from inspect_ai.approval._policy import ApprovalPolicy
 from inspect_ai.log._samples import init_active_samples
 from inspect_ai.model import GenerateConfig, Model
 from inspect_ai.model._model import init_active_model, init_model_usage
@@ -24,6 +26,12 @@ def init_eval_context(
     init_human_approval_manager()
 
 
-def init_task_context(model: Model, config: GenerateConfig = GenerateConfig()) -> None:
+def init_task_context(
+    model: Model,
+    approval: list[ApprovalPolicy] | None = None,
+    config: GenerateConfig = GenerateConfig(),
+) -> None:
     init_active_model(model, config)
     init_model_usage()
+    if not have_tool_approval():
+        init_tool_approval(approval)

inspect_ai/_eval/score.py

@@ -11,7 +11,7 @@ from inspect_ai.log import (
     EvalMetric,
 )
 from inspect_ai.model import ModelName
-from inspect_ai.scorer import Metric, Score, Scorer, Target
+from inspect_ai.scorer import Metric, Scorer, Target
 from inspect_ai.scorer._metric import SampleScore
 from inspect_ai.scorer._reducer import (
     ScoreReducer,
@@ -108,7 +108,7 @@ async def score_async(
 
         # write them back (gather ensures that they come back in the same order)
         for index, score in enumerate(scores):
-            log.samples[index].scores = cast(dict[str, Score], score)
+            log.samples[index].scores = {k: v.score for k, v in score.items()}
 
         # collect metrics from EvalLog (they may overlap w/ the scorer metrics,
         # that will be taken care of in eval_results)
@@ -151,11 +151,8 @@ async def task_score(task: Task, log: EvalLog) -> EvalLog:
         sample_scores = [
             {
                 score_key: SampleScore(
+                    score=score,
                     sample_id=sample.id,
-                    value=score.value,
-                    answer=score.answer,
-                    explanation=score.explanation,
-                    metadata=score.metadata,
                 )
                 for score_key, score in sample.scores.items()
             }
@@ -185,11 +182,8 @@ async def run_score_task(
         scorer_name = unique_scorer_name(scorer, list(results.keys()))
 
         results[scorer_name] = SampleScore(
+            score=result,
             sample_id=state.sample_id,
-            value=result.value,
-            answer=result.answer,
-            explanation=result.explanation,
-            metadata=result.metadata,
         )
 
     progress()

inspect_ai/_eval/task/results.py

@@ -13,6 +13,7 @@ from inspect_ai._util.registry import (
 from inspect_ai.log import (
     EvalMetric,
     EvalResults,
+    EvalSampleScore,
     EvalScore,
 )
 from inspect_ai.log._log import EvalSampleReductions
@@ -345,7 +346,7 @@ def resolve_glob_metric_keys(
 
 def reduce_scores(
     scores: list[SampleScore], reducer: ScoreReducer
-) -> list[SampleScore]:
+) -> list[EvalSampleScore]:
     # Group the scores by sample_id
     grouped_scores: dict[str, list[SampleScore]] = defaultdict(list)
     for sample_score in scores:
@@ -353,11 +354,11 @@ def reduce_scores(
             grouped_scores[str(sample_score.sample_id)].append(sample_score)
 
     # reduce the scores
-    reduced_scores: list[SampleScore] = []
+    reduced_scores: list[EvalSampleScore] = []
     for scores in grouped_scores.values():
-        reduced = reducer(cast(list[Score], scores))
+        reduced = reducer([score.score for score in scores])
         reduced_scores.append(
-            SampleScore(
+            EvalSampleScore(
                 sample_id=scores[0].sample_id,
                 value=reduced.value,
                 answer=reduced.answer,

inspect_ai/_eval/task/run.py

@@ -6,7 +6,7 @@ from copy import deepcopy
 from dataclasses import dataclass, field
 from logging import getLogger
 from pathlib import PurePath
-from typing import Callable, Literal, cast
+from typing import Callable, Literal
 
 from typing_extensions import Unpack
 
@@ -62,7 +62,7 @@ from inspect_ai.model import (
 )
 from inspect_ai.model._model import init_sample_model_usage, sample_model_usage
 from inspect_ai.scorer import Scorer, Target
-from inspect_ai.scorer._metric import Metric, SampleScore, Score
+from inspect_ai.scorer._metric import Metric, SampleScore
 from inspect_ai.scorer._reducer.types import ScoreReducer
 from inspect_ai.scorer._score import init_scoring_context
 from inspect_ai.scorer._scorer import unique_scorer_name
@@ -136,7 +136,7 @@ async def task_run(options: TaskRunOptions) -> EvalLog:
     generate_config = task.config.merge(GenerateConfigArgs(**kwargs))
 
     # init task context
-    init_task_context(model, generate_config)
+    init_task_context(model, options.task.approval, generate_config)
 
     # establish run_dir for duration of execution
     with set_task_run_dir(task_run_dir(task)):
@@ -503,11 +503,8 @@ async def task_run_sample(
             sample_scores = (
                 {
                     key: SampleScore(
+                        score=score,
                         sample_id=previous_sample.id,
-                        value=score.value,
-                        answer=score.answer,
-                        explanation=score.explanation,
-                        metadata=score.metadata,
                     )
                     for key, score in previous_sample.scores.items()
                 }
@@ -652,11 +649,8 @@ async def task_run_sample(
                             )
                             if score_result is not None:
                                 sample_score = SampleScore(
+                                    score=score_result,
                                     sample_id=sample.id,
-                                    value=score_result.value,
-                                    answer=score_result.answer,
-                                    explanation=score_result.explanation,
-                                    metadata=score_result.metadata,
                                 )
                                 transcript()._event(
                                     ScoreEvent(score=score_result, target=sample.target)
@@ -759,7 +753,7 @@ async def log_sample(
         setup=sample.setup,
         messages=state.messages,
         output=state.output,
-        scores=cast(dict[str, Score], scores),
+        scores={k: v.score for k, v in scores.items()},
         store=dict(state.store.items()),
         events=list(transcript().events),
         model_usage=sample_model_usage(),

inspect_ai/_eval/task/task.py

@@ -7,6 +7,7 @@ from typing_extensions import TypedDict, Unpack
 
 from inspect_ai._util.logger import warn_once
 from inspect_ai._util.registry import is_registry_object, registry_info
+from inspect_ai.approval._policy import ApprovalPolicy, approval_policies_from_config
 from inspect_ai.dataset import Dataset, MemoryDataset, Sample
 from inspect_ai.log import EvalLog
 from inspect_ai.model import GenerateConfig
@@ -49,6 +50,9 @@ class Task:
         config (GenerateConfig): Model generation config.
         sandbox (SandboxEnvironmentType | None): Sandbox environment type
           (or optionally a str or tuple with a shorthand spec)
+        approval: (str | list[ApprovalPolicy] | None): Tool use approval policies.
+          Either a path to an approval policy config file or a list of approval policies.
+          Defaults to no approval policy.
         epochs (int | Epochs | None): Epochs to repeat samples for and optional score
            reducer function(s) used to combine sample scores (defaults to "mean")
         fail_on_error (bool | float | None): `True` to fail on first sample error
@@ -76,6 +80,7 @@ class Task:
         metrics: list[Metric] | dict[str, list[Metric]] | None = None,
         config: GenerateConfig = GenerateConfig(),
         sandbox: SandboxEnvironmentType | None = None,
+        approval: str | list[ApprovalPolicy] | None = None,
         epochs: int | Epochs | None = None,
         fail_on_error: bool | float | None = None,
         message_limit: int | None = None,
@@ -134,6 +139,11 @@ class Task:
         self.metrics = metrics
         self.config = config
         self.sandbox = resolve_sandbox_environment(sandbox)
+        self.approval = (
+            approval_policies_from_config(approval)
+            if isinstance(approval, str)
+            else approval
+        )
         self.epochs = epochs.epochs if epochs else None
         self.epochs_reducer = epochs.reducer if epochs else None
         self.fail_on_error = fail_on_error

inspect_ai/_util/ansi.py

@@ -0,0 +1,31 @@
+import os
+from typing import Any
+
+from rich.console import Console, RenderableType
+
+
+def render_text(
+    text: RenderableType | list[RenderableType], styles: bool = True, **options: Any
+) -> str:
+    """Render text from Rich renderables.
+
+    Args:
+      text (RenderableType | list[RenderableType]): Renderables.
+      styles (bool): If True, ansi escape codes will be included. False for plain text.
+        Defaults to True.
+      **options (Any): Additonal keyword arguments to pass to `Console` constructor.
+
+    Returns:
+       str: Rendered text (with ansi codes if `styles=True`)
+    """
+    # resolve to text
+    text = text if isinstance(text, list) else [text]
+
+    # print to console attached to /dev/null
+    with open(os.devnull, "w") as f:
+        console = Console(file=f, record=True, force_terminal=True, **options)
+        for t in text:
+            console.print(t)
+
+    # export (optionally w/ ansi styles)
+    return console.export_text(styles=styles).strip()

inspect_ai/_util/format.py

@@ -26,3 +26,10 @@ def format_value(value: object, width: int) -> str:
     elif isinstance(value, list | tuple | dict):
         return pprint.pformat(value, width=width)
     return str(value)
+
+
+def format_progress_time(time: float, pad_hours: bool = True) -> str:
+    minutes, seconds = divmod(time, 60)
+    hours, minutes = divmod(minutes, 60)
+    hours_fmt = f"{hours:2.0f}" if pad_hours else f"{hours:.0f}"
+    return f"{hours_fmt}:{minutes:02.0f}:{seconds:02.0f}"

inspect_ai/_util/logger.py

@@ -1,3 +1,4 @@
+import atexit
 import os
 from logging import (
     DEBUG,
@@ -30,7 +31,12 @@ from .constants import (
     TRACE_LOG_LEVEL,
 )
 from .error import PrerequisiteError
-from .trace import TraceFileHandler, TraceFormatter, inspect_trace_dir
+from .trace import (
+    TraceFormatter,
+    compress_trace_log,
+    inspect_trace_file,
+    rotate_trace_files,
+)
 
 TRACE_FILE_NAME = "trace.log"
 
@@ -56,19 +62,13 @@ class LogHandler(RichHandler):
         else:
             self.file_logger_level = 0
 
-        # add a trace handler
-        default_trace_file = inspect_trace_dir() / TRACE_FILE_NAME
-        have_existing_trace_file = default_trace_file.exists()
+        # add a trace file handler
+        rotate_trace_files()  # remove oldest if > 10 trace files
         env_trace_file = os.environ.get("INSPECT_TRACE_FILE", None)
-        trace_file = Path(env_trace_file) if env_trace_file else default_trace_file
-        trace_total_files = 10
-        self.trace_logger = TraceFileHandler(
-            trace_file.as_posix(),
-            backupCount=trace_total_files - 1,  # exclude the current file (10 total)
-        )
+        trace_file = Path(env_trace_file) if env_trace_file else inspect_trace_file()
+        self.trace_logger = FileHandler(trace_file)
         self.trace_logger.setFormatter(TraceFormatter())
-        if have_existing_trace_file:
-            self.trace_logger.doRollover()
+        atexit.register(compress_trace_log(self.trace_logger))
 
         # set trace level
         trace_level = os.environ.get("INSPECT_TRACE_LEVEL", TRACE_LOG_LEVEL)

inspect_ai/_util/throttle.py

@@ -3,7 +3,16 @@ from functools import wraps
 from typing import Any, Callable
 
 
-def throttle(seconds: int) -> Callable[..., Any]:
+def throttle(seconds: float) -> Callable[..., Any]:
+    """Throttle a function to ensure it is called no more than every n seconds.
+
+    Args:
+       seconds (float): Throttle time.
+
+    Returns:
+       Callable: Throttled function.
+    """
+
     def decorator(func: Callable[..., Any]) -> Callable[..., Any]:
         last_called: float = 0
         last_result: Any = None