themis-eval 0.1.0__py3-none-any.whl → 0.1.1__py3-none-any.whl

themis/utils/tracing.py

@@ -0,0 +1,320 @@
+"""Distributed tracing and observability utilities.
+
+This module provides lightweight span-based tracing for understanding
+experiment execution performance and behavior. Tracing is opt-in and
+disabled by default for minimal performance impact.
+
+Examples:
+    # Enable tracing
+    tracing.enable()
+
+    # Use context manager for automatic span management
+    with tracing.span("my_operation", task_id="123"):
+        # ... do work ...
+        with tracing.span("subprocess"):
+            # ... nested work ...
+            pass
+
+    # Get trace for analysis
+    trace = tracing.get_trace()
+    print(f"Total time: {trace.duration_ms()}ms")
+
+    # Export trace
+    tracing.export_json("trace.json")
+    tracing.disable()
+"""
+
+from __future__ import annotations
+
+import json
+import threading
+import time
+from contextlib import contextmanager
+from dataclasses import dataclass, field
+from typing import Any, Iterator
+
+
+@dataclass
+class Span:
+    """Represents a traced operation with timing and metadata.
+
+    Spans can be nested to create a tree of operations showing
+    how time is spent during experiment execution.
+    """
+
+    name: str
+    start_time: float
+    end_time: float | None = None
+    metadata: dict[str, Any] = field(default_factory=dict)
+    parent: Span | None = None
+    children: list[Span] = field(default_factory=list)
+    span_id: str = field(default_factory=lambda: str(time.time_ns()))
+
+    def duration_ms(self) -> float:
+        """Calculate span duration in milliseconds.
+
+        Returns:
+            Duration in milliseconds, or time elapsed so far if span not closed
+        """
+        if self.end_time is None:
+            return (time.perf_counter() - self.start_time) * 1000
+        return (self.end_time - self.start_time) * 1000
+
+    def is_complete(self) -> bool:
+        """Check if span has been closed."""
+        return self.end_time is not None
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert span to dictionary for serialization."""
+        return {
+            "name": self.name,
+            "span_id": self.span_id,
+            "start_time": self.start_time,
+            "end_time": self.end_time,
+            "duration_ms": self.duration_ms() if self.is_complete() else None,
+            "metadata": self.metadata,
+            "children": [child.to_dict() for child in self.children],
+        }
+
+    def find_spans(self, name: str) -> list[Span]:
+        """Find all spans with given name in this span and descendants.
+
+        Args:
+            name: Span name to search for
+
+        Returns:
+            List of matching spans
+        """
+        matches = []
+        if self.name == name:
+            matches.append(self)
+
+        for child in self.children:
+            matches.extend(child.find_spans(name))
+
+        return matches
+
+    def get_summary(self) -> dict[str, Any]:
+        """Get summary statistics for this span tree.
+
+        Returns:
+            Dictionary with timing statistics by span name
+        """
+        summary: dict[str, dict[str, Any]] = {}
+
+        def collect(span: Span):
+            if span.name not in summary:
+                summary[span.name] = {
+                    "count": 0,
+                    "total_ms": 0.0,
+                    "min_ms": float("inf"),
+                    "max_ms": 0.0,
+                }
+
+            stats = summary[span.name]
+            if span.is_complete():
+                duration = span.duration_ms()
+                stats["count"] += 1
+                stats["total_ms"] += duration
+                stats["min_ms"] = min(stats["min_ms"], duration)
+                stats["max_ms"] = max(stats["max_ms"], duration)
+
+            for child in span.children:
+                collect(child)
+
+        collect(self)
+
+        # Calculate averages
+        for stats in summary.values():
+            if stats["count"] > 0:
+                stats["avg_ms"] = stats["total_ms"] / stats["count"]
+
+        return summary
+
+
+class TracingContext:
+    """Thread-local tracing context.
+
+    This class manages the current span stack for each thread, allowing
+    nested spans and proper parent-child relationships.
+    """
+
+    def __init__(self):
+        self._local = threading.local()
+        self._enabled = False
+
+    def enable(self) -> None:
+        """Enable tracing."""
+        self._enabled = True
+
+    def disable(self) -> None:
+        """Disable tracing."""
+        self._enabled = False
+
+    def is_enabled(self) -> bool:
+        """Check if tracing is enabled."""
+        return self._enabled
+
+    def _get_state(self):
+        """Get thread-local state, initializing if needed."""
+        if not hasattr(self._local, "root"):
+            self._local.root = None
+            self._local.current = None
+        return self._local
+
+    def get_root(self) -> Span | None:
+        """Get the root span for current thread."""
+        return self._get_state().root
+
+    def get_current(self) -> Span | None:
+        """Get the current active span for current thread."""
+        return self._get_state().current
+
+    @contextmanager
+    def span(self, name: str, **metadata) -> Iterator[Span | None]:
+        """Create a traced span as a context manager.
+
+        Args:
+            name: Name of the span
+            **metadata: Additional metadata to attach to span
+
+        Yields:
+            Span object if tracing is enabled, None otherwise
+        """
+        if not self._enabled:
+            yield None
+            return
+
+        state = self._get_state()
+        span_obj = Span(
+            name=name,
+            start_time=time.perf_counter(),
+            metadata=metadata,
+            parent=state.current,
+        )
+
+        # Link to parent
+        if state.current is not None:
+            state.current.children.append(span_obj)
+        else:
+            # This is the root span
+            state.root = span_obj
+
+        # Make this the current span
+        prev_current = state.current
+        state.current = span_obj
+
+        try:
+            yield span_obj
+        finally:
+            # Close span
+            span_obj.end_time = time.perf_counter()
+
+            # Restore previous current
+            state.current = prev_current
+
+    def reset(self) -> None:
+        """Reset tracing state for current thread."""
+        state = self._get_state()
+        state.root = None
+        state.current = None
+
+
+# Global tracing context
+_global_context = TracingContext()
+
+
+# Public API functions
+def enable() -> None:
+    """Enable tracing globally."""
+    _global_context.enable()
+
+
+def disable() -> None:
+    """Disable tracing globally."""
+    _global_context.disable()
+
+
+def is_enabled() -> bool:
+    """Check if tracing is enabled."""
+    return _global_context.is_enabled()
+
+
+@contextmanager
+def span(name: str, **metadata) -> Iterator[Span | None]:
+    """Create a traced span.
+
+    This is the main API for creating spans. Use as a context manager:
+
+        with tracing.span("my_operation", task_id="123"):
+            # ... do work ...
+            pass
+
+    Args:
+        name: Name of the span
+        **metadata: Additional metadata to attach to span
+
+    Yields:
+        Span object if tracing is enabled, None otherwise
+    """
+    with _global_context.span(name, **metadata) as s:
+        yield s
+
+
+def get_trace() -> Span | None:
+    """Get the root span for the current thread's trace.
+
+    Returns:
+        Root span, or None if no trace exists
+    """
+    return _global_context.get_root()
+
+
+def reset() -> None:
+    """Reset tracing state for current thread."""
+    _global_context.reset()
+
+
+def export_json(filepath: str, indent: int = 2) -> None:
+    """Export current trace to JSON file.
+
+    Args:
+        filepath: Path to write JSON file
+        indent: JSON indentation level (default: 2)
+    """
+    trace = get_trace()
+    if trace is None:
+        raise ValueError("No trace to export")
+
+    with open(filepath, "w") as f:
+        json.dump(trace.to_dict(), f, indent=indent)
+
+
+def get_summary() -> dict[str, Any]:
+    """Get summary statistics for current trace.
+
+    Returns:
+        Dictionary with timing statistics by span name
+
+    Raises:
+        ValueError: If no trace exists
+    """
+    trace = get_trace()
+    if trace is None:
+        raise ValueError("No trace to summarize")
+
+    return trace.get_summary()
+
+
+__all__ = [
+    "Span",
+    "TracingContext",
+    "enable",
+    "disable",
+    "is_enabled",
+    "span",
+    "get_trace",
+    "reset",
+    "export_json",
+    "get_summary",
+]

{themis_eval-0.1.0.dist-info → themis_eval-0.1.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: themis-eval
-Version: 0.1.0
+Version: 0.1.1
 Summary: Lightweight evaluation platform for LLM experiments
 Author: Pittawat Taveekitworachai
 License: MIT

themis_eval-0.1.1.dist-info/RECORD

@@ -0,0 +1,134 @@
+themis/__init__.py,sha256=3bKi1PneI5PaTaDPXsArCVvfinkLFDRU91lvZIeg7V0,281
+themis/_version.py,sha256=tc4TJqWVv2dx4UzItLqneMPaG7vM8CQFDNW5pJgNoKg,345
+themis/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+themis/cli/__init__.py,sha256=An2DrMHRfmiee5BYJ6TGqvbG7sXWECjjyvEgcoGJ7cE,99
+themis/cli/__main__.py,sha256=df2pOghoSuq18hZmVVikmGhaFSaRe-jeDOnrsu-1QDM,135
+themis/cli/main.py,sha256=NSayD8zwH5NJjC_qaTwqhDR7_8PO3I2n1dPX-fBYa-A,3131
+themis/cli/new_project.py,sha256=D8asV4QbjgQNYvmXt_WhK4nPM-wKHe_K0VJiBdgtO_E,1121
+themis/cli/utils.py,sha256=NAPyFiXspfpx5vBxA8aEcOMmWEDyt-R8ywoHo_8Nr4A,1307
+themis/cli/commands/__init__.py,sha256=CTx7su3qTtq96qxLNclDsE6UM_86NhaS01M9-x9wFiw,287
+themis/cli/commands/benchmarks.py,sha256=HjCfmhu1FYUEGlaxUZGIZs0I-2gNG4tn_kF29NWPGcc,7885
+themis/cli/commands/comparison.py,sha256=Ki1_MMFFR4vBJkZTeIMWLh-_zdjbtJZurI3YyrEs4vw,12364
+themis/cli/commands/config_commands.py,sha256=eL6GtdIllOIHo8GbNN2jOqLn5VUPBuqhnro9ooPxDog,7387
+themis/cli/commands/cost.py,sha256=fFdF6hKIzsbPsyrJ1nt6-2m43PpVGUj8jx5T90tBTNo,7233
+themis/cli/commands/demo.py,sha256=akQqjG-hbUDfeB3bI8K4F5-S0ibJqhflGBFQ5nvdUgE,2135
+themis/cli/commands/info.py,sha256=9maOaw-TFiBpuVhaqlMKukGuZ_zgESetqbMQ1Qdvjxs,2515
+themis/cli/commands/leaderboard.py,sha256=AVvsYIwZAY18jn3sOq3QD45yNtfdHUEl7eixM4aMCKw,10615
+themis/cli/commands/math_benchmarks.py,sha256=nQ4TcPB7T9O3piAy4_TgrOQOQxh2Q8OyBreK_HoPCeQ,9946
+themis/cli/commands/mcq_benchmarks.py,sha256=Cls5W1jGd7TKizmw07CnZWY5N6ywR8VhJ6jKDnY_cRk,7026
+themis/cli/commands/sample_run.py,sha256=r3Ymg5dVHg4IAVJvzoP0ZWUWWUE4Dia1t0062Yhdk9Q,9445
+themis/cli/commands/visualize.py,sha256=ZECkB0NjIltuOeBE-Q1JnndZEMXVzc8KgcrbaP-GSXo,9740
+themis/config/__init__.py,sha256=YMdFG1iLvOQUnSPlc_ZJVn5zCCTbIozML64b4qUtGR8,476
+themis/config/loader.py,sha256=t_wcIDwekuy3EaLprQgWILSKH2h5lFkF7quvNfAHddM,746
+themis/config/registry.py,sha256=sSrL1mTjUG86s30o-dhuRInY4YeumEICtmxdjS-PiiQ,1055
+themis/config/runtime.py,sha256=hU69_oND7fJfAOIBJONENmsuf7Y8roO7n-w9OwxzoT8,7475
+themis/config/schema.py,sha256=SMR9QHp8OBkSnb1dHyOgg-IJWSqpXfyAqywnBeMy46M,3196
+themis/core/__init__.py,sha256=S8G1x-39sZ3_NQ5DJ6R1yBTWXp_gO0WxOtVjeB9sTwY,113
+themis/core/conversation.py,sha256=wwO8RS4t4plDR0Sf1KjYv_ejonlvKe0ZwAD-4sfGak8,10155
+themis/core/entities.py,sha256=S1Kw3qrx0vWWdOmskZnu3GuKGOGf0WDpHBazmwhEnmM,3770
+themis/core/serialization.py,sha256=cxfoSKwcZiNsnR8g_SAJAq1ZLrfLXM4S9_rVEDUT8qs,7071
+themis/core/tools.py,sha256=v0_ctsBCtinZGNC_I4C-h0GUPNM5ZeTi7z-U4iCtyp4,11035
+themis/core/types.py,sha256=I5rr9MMS0irX4lo-xlqGjosx-FjPgT64RzQAraM223A,3652
+themis/datasets/__init__.py,sha256=r1FobxkALtaXuphz1wU6LSbXhDezpmh6AUvoT-Hv1As,9013
+themis/datasets/base.py,sha256=HtgIIiCgaog9aRnjZ4kJV3Ta2Dhl-_wrD8PB2lvK810,8417
+themis/datasets/commonsense_qa.py,sha256=1eT_SaehbHC8F1HOJDpMi4RnzsKznGuJrdoL8RLuHu4,5842
+themis/datasets/competition_math.py,sha256=XWsKMAPICb66RMXZNA8AvtUJwWQav3HVDuvAby8QjZU,7346
+themis/datasets/coqa.py,sha256=bsh1AUYtagYIa7d-eM30JftGBQ7_0xUZq9dW9Pkz3Ks,4575
+themis/datasets/gpqa.py,sha256=VQEDbDywDGhRE69M-mXwY5AxUO5WQw2RF6ND7hOlv50,7110
+themis/datasets/gsm8k.py,sha256=wYbCACnF560RTAJTU-SMHqVvwzj8QT0bbzfQGTuRKl0,3887
+themis/datasets/gsm_symbolic.py,sha256=PeWRBJIt8zEcsWIIj4xw5iT2drBotv8jB3DrHkN-PDw,4055
+themis/datasets/math500.py,sha256=F1xBSadcCdnKWs-WKoFVHBeOt1tu6Rb6U2Yi4Ien578,3840
+themis/datasets/med_qa.py,sha256=Z6FET8UrDsQnsbuvPoVlO1thJ6jeMcmSnqUvnHY0GtI,6019
+themis/datasets/medmcqa.py,sha256=9MjkhjIqNaP_sslPnid4GHiqaGjXZkWGvwLJGaUeHzg,5532
+themis/datasets/mmlu_pro.py,sha256=51cZGLRe1BVuXhQ5vFzIk4TCq4p8iYUow9E9hSSNFAE,8384
+themis/datasets/piqa.py,sha256=YV-c56ZUrvIqS0wpBq4LdC683hDqEJaZCfhh4Bn5HwU,4420
+themis/datasets/registry.py,sha256=GCgfdr7dp0LpDiK0DLhOFg8tS18m1Rx6t7bOJlh10rU,5530
+themis/datasets/schema.py,sha256=9nfM0ygdtvh1PPNLLz1OF2P8Z4sjTVk7gT5GIffJsf0,7100
+themis/datasets/sciq.py,sha256=VXxO2_cDPhs08AnUBtT5NE6aKvnGMeNa8FXdQZqas2U,4677
+themis/datasets/social_i_qa.py,sha256=y2uT2mWwo7ArZwjEUSo5841oITDcFII77OCjNVTkwwk,4765
+themis/datasets/super_gpqa.py,sha256=Mr1ag_FyAk1haxg6_ONX5F84wQYtbSVjV-MlMNmaHlI,8452
+themis/evaluation/__init__.py,sha256=2Jl8tcVxYAsmHNAZev2mPS_mEwZcRzebqSM3QDc2cyY,36
+themis/evaluation/conditional.py,sha256=ayndI7FcwxdIMR8B4ddgcKZd5Jl5NQcBJUp7eXI6Djk,13881
+themis/evaluation/math_verify_utils.py,sha256=vXMvL11-IH16UHZ-mbi_r5hOFz7aUfR1J1laa6qmLMk,2213
+themis/evaluation/pipeline.py,sha256=OOowN59UdOMF2Hwy_G4ky5yzR4ajAnohil6xteWBHqE,1503
+themis/evaluation/reports.py,sha256=9om7jzZUtmlMH7EeteXp_98gfHct4x09AyTFy3FSAdQ,8715
+themis/evaluation/extractors/__init__.py,sha256=BanoC_8e0iam-VU7l7uhvhac_6w_JJZYoYE4xXPUrGk,566
+themis/evaluation/extractors/error_taxonomy_extractor.py,sha256=RrRx-23l3LwTdG89kvSQJng438cfYI-IdtOGUD6gEDw,2462
+themis/evaluation/extractors/exceptions.py,sha256=lI4HOU98FEUwqhBq2uEzY0ym0evK85O5V3t6ULoANtM,167
+themis/evaluation/extractors/identity_extractor.py,sha256=bBRgcry94AtUrchzBdHK-a9fnWAQT0eJDaPXjBmgBis,675
+themis/evaluation/extractors/json_field_extractor.py,sha256=5_ndU7NBw9CkfP8oXmeue3wkGZZauryvFt_GVjO7X44,1232
+themis/evaluation/extractors/math_verify_extractor.py,sha256=f8B_IeexS2QHCf-UUhqfk7xpqWcm8SgobLQ7GXYrM58,1155
+themis/evaluation/extractors/regex_extractor.py,sha256=xNo8YRy7JOnF9ohbK_xUHcbLGMUmE1qooUlS0ucrw44,1148
+themis/evaluation/metrics/__init__.py,sha256=HrrTgEwdCmXiib0ohUOArpUC0qHii2A_mgPX0m4LRoU,652
+themis/evaluation/metrics/composite_metric.py,sha256=nF4FXMWwEMt1Gfq4NhwEuNVD28d97cgV30kFgt1LdEg,1474
+themis/evaluation/metrics/consistency_metric.py,sha256=ITu4qFwgVmEiRD_tuVOhe5B9qtdmRZI3Pj-p0NlXizE,2404
+themis/evaluation/metrics/exact_match.py,sha256=bAH3QCddbD7s0Mp3-4VGvzwnOv7MB8nexwcFpO0yJuo,1596
+themis/evaluation/metrics/length_difference_tolerance.py,sha256=_YjDNGEMlodLogcw64RshG1-i0pKH27S3UiMeK6SdQw,966
+themis/evaluation/metrics/math_verify_accuracy.py,sha256=YhBhpONLmouLELfpcjNHiVSlwpxkE2wkStF6__du3_0,1329
+themis/evaluation/metrics/pairwise_judge_metric.py,sha256=DEYKwt3smzXiSUhDV4lWxDFXWoHz-JMg3z5bMjlLPKo,4890
+themis/evaluation/metrics/response_length.py,sha256=Xn2PQi4pMLhC_3bMmSbLEf-QVFOrMNm2ZJr0PiCDH-E,910
+themis/evaluation/metrics/rubric_judge_metric.py,sha256=KSSqwpMHaXCK6krbb_A93nppZ_0xk6Or30u7csnw7rM,4796
+themis/evaluation/pipelines/__init__.py,sha256=5YI1xaUULHisctFxrumN4XRpWYneoonX7nd9zBtsjvQ,384
+themis/evaluation/pipelines/composable_pipeline.py,sha256=nNP9MSvQQJvaSBw5_gO3FeyhGm9So2ZlGqh5qSvE8Ac,10905
+themis/evaluation/pipelines/standard_pipeline.py,sha256=5ub_7zfQVj9YbDQxYTdxNIVijlMktR0rrfmN0GSndnI,11610
+themis/evaluation/statistics/__init__.py,sha256=TTrScTLAW7EHNq0nbjuJs6iP3_HgDx1yy3EtYXx5JCk,1257
+themis/evaluation/statistics/bootstrap.py,sha256=JUQ8rtzFvW2e41I2pLJ7pqgSEjuJ1r6McyYLI42At9g,2409
+themis/evaluation/statistics/confidence_intervals.py,sha256=CN5EO2gWiSITQubuWuPryngnGXhGwczY9kO3mcG6JVc,3676
+themis/evaluation/statistics/distributions.py,sha256=h-sJgtpB-KnTwgN1B2TXrugjCJp8fl9DmGy-ZxZVEek,5538
+themis/evaluation/statistics/effect_sizes.py,sha256=EWFVDilczpR8rR3_YurWy7QcjYcNIEzGOvf931OYPww,3283
+themis/evaluation/statistics/hypothesis_tests.py,sha256=MVlVsY8wXifbBG5aSwauFShsQtIKqYREJApbriojS2o,10042
+themis/evaluation/statistics/types.py,sha256=hW0RYWs-G4C_njNl0ZGG9lJROgU2CfLWfnTQDWYmWuw,3685
+themis/evaluation/strategies/__init__.py,sha256=3f5LQkzlu3pRbN7dgDbdYOUNZTRexcn6f8D8I5-C724,439
+themis/evaluation/strategies/attempt_aware_evaluation_strategy.py,sha256=O3dlsQ2F0Ucv2Dhjz2Qf-jpPhwaVs3zrdQDRRu9du5w,1714
+themis/evaluation/strategies/default_evaluation_strategy.py,sha256=LShW-3Nxg_W4Ln-4qUvHJZqe5YMt64gHoK3uNJYLQNo,693
+themis/evaluation/strategies/evaluation_strategy.py,sha256=YFF-bXkz4Z52GuCw52FcklfEnf8dK8_z_I40DJRcmwE,669
+themis/evaluation/strategies/judge_evaluation_strategy.py,sha256=58pDB30y1VpM_1KPB6sGS0JImGZk5WTgnK9CKDF8N5k,2304
+themis/experiment/__init__.py,sha256=dGranqpESugmmfbQlTU9efwspazW6j3vcmAKEtAoWZk,182
+themis/experiment/builder.py,sha256=AEjCDeSOI2B0i0PBjkfY1GUDNrYGTGiqPvt0SxnDQFo,5618
+themis/experiment/cache_manager.py,sha256=KlSMPy5CICX3XmqnYwg6hP0TB4zzJu2nb-nLcZAFzy0,4012
+themis/experiment/comparison.py,sha256=Mr1L5Zj7i87xk9XUQ_UueLTsC-sDZH8YGwLwg_gG0VI,21562
+themis/experiment/cost.py,sha256=flhENfB5WKvyNWwPMDtygNZAv6y_yv4RoClsRz714Hc,10159
+themis/experiment/definitions.py,sha256=oOZBFfEQkSBiZd9CMutCQ5luH6oeUT9yAZFd7fpVjnw,2015
+themis/experiment/export.py,sha256=wsUICG9XG6HET2OBdQisHPXVv5pjF3HWCvcpv1A9igM,26012
+themis/experiment/export_csv.py,sha256=80w3gEGjeLjuiNq539rRP73k3MBtwrzJy90hgE91AKw,6030
+themis/experiment/integration_manager.py,sha256=wTVTjDGcUkzz4tfnwSxa5nK1A4e2FKCPazDYGcdzYS8,3325
+themis/experiment/math.py,sha256=P2E9F_UKI7pb-aXepSztGdr_g309WEMe83zqg1nWO7A,6973
+themis/experiment/mcq.py,sha256=DDB99FHQsU_5vMIRDRhSZ7pReYvVf57wLmmo3OU_An4,6276
+themis/experiment/orchestrator.py,sha256=D2ANvg2s4Dyo0ridZ3alDVl1dTW4kazLoMeG0knQ6-M,14244
+themis/experiment/pricing.py,sha256=fTM32yE3L8vahMP4sr1zr7dbp9zYCjiPN4D4VuZ8-q8,9346
+themis/experiment/storage.py,sha256=58tSwHn3J36UMMCWblkbs00ZjCyIqLkoR2Vib9c-zgE,9156
+themis/experiment/visualization.py,sha256=dJYHrp3mntl8CPc5HPI3iKqPztVsddQB3ogRkd_FCNc,18473
+themis/generation/__init__.py,sha256=6KVwCQYMpPIsXNuWDZOGuqHkUkA45lbSacIFn8ZbD4s,36
+themis/generation/agentic_runner.py,sha256=armBQBk7qZDBEwT8HqjIWomYDQm57NfrP5CZJzay2uA,13669
+themis/generation/batching.py,sha256=ddpgpn1pq_EwipvTg-K4WcoSs3c2rbW37jEA5Pa_spo,7557
+themis/generation/clients.py,sha256=6apXCp_VNQosnpnmohTHOhHGXw-VZgsUyLds8MwtYUE,4910
+themis/generation/conversation_runner.py,sha256=kSZHwEvfqzxZ-eQYxmg5OkNZcgEHggZExjad6nBOeTM,7980
+themis/generation/plan.py,sha256=RmPIdefXkQMHYv5EWiilpx91I9a-svw31imvG0wV3fE,15961
+themis/generation/router.py,sha256=jZc0KFL483f8TrYtt9yxzFKs-T9CG2CoE2kfOQdHMEc,1082
+themis/generation/runner.py,sha256=iHTE5vSMWMYRrv4PEWMaZflF939nv1wWccK8V0e092c,8009
+themis/generation/strategies.py,sha256=hjqaVkNycFxJWh_edJ7ilBl7HS6bL-8pYm24zTfoAvg,2975
+themis/generation/templates.py,sha256=ut_6akp8Y6Ey_9O3s64jDbwCB74pw62Zf8URlYcKHkA,2325
+themis/generation/turn_strategies.py,sha256=w33qhzpQbGTsfeOgOgMDovV0wEeXeNZUUBm5yZy1naw,10973
+themis/generation/types.py,sha256=MkJnZk6lMHmHzlJVEsuIC9ioRW8XhWcSk9AdDeb_aLE,338
+themis/generation/providers/litellm_provider.py,sha256=rlTuglIwhcvSakCo5G-ffgQtEHbCEX0ZeKk6M1MaWmU,8155
+themis/generation/providers/vllm_provider.py,sha256=0K4we6xDrRXlBXseC1ixLq2sJpRF4T8Ikv45dw-zNk4,4625
+themis/integrations/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+themis/integrations/huggingface.py,sha256=oUkLVyacmukDwHrMUVayHJ4fWnVCA77ZyuEkC7aogmY,2032
+themis/integrations/wandb.py,sha256=VVVReLM-qjw4DyLVJcoaSH1rSTMNf3ReOKMxOFwJKu4,2247
+themis/interfaces/__init__.py,sha256=OSHATlp1fNBJrVq0lNM85MGjWmSj0NKtnLnp3fmYn1k,2330
+themis/project/__init__.py,sha256=vgLv2nS62yz1XsFSFzFf7eIo6FyQJXpOY9OPRUcTQLQ,465
+themis/project/definitions.py,sha256=vHARw0IjFOWE4RL4mGRwvke36A6GWQGep6cQFIRcpJg,3329
+themis/project/patterns.py,sha256=2J51Q9Jq7X-2N57uexvR191gaZKwusef5vIuIVUQY-E,7743
+themis/providers/__init__.py,sha256=K5nG0DsK_YPY0cT9MBLk5BLcLbBo0wBP0vQvLjpAw_Y,189
+themis/providers/registry.py,sha256=Za5Kg3-A-35wS_jiGpPXV2q1k6he_dRIWVqt36dKN-4,1056
+themis/utils/api_generator.py,sha256=3oQ7mGZlFx2Dpm45pMg3rNIqNK2Smj05PjOMXp5RIkQ,10776
+themis/utils/cost_tracking.py,sha256=9_Z2iTfNaQse9G_bnqn4hme4T0fG2W-fxOLEDeF_3VI,11545
+themis/utils/dashboard.py,sha256=2yiIu9_oENglTde_J3G1d5cpQ5VtSnfbUvdliw5Og1E,13008
+themis/utils/logging_utils.py,sha256=YNSiDfO4LsciSzUhHF1aTVI5rkfnWiVbn1NcGjjmJuQ,1019
+themis/utils/progress.py,sha256=b3YwHKV5x3Cvr5rBukqifJimK3Si4CGY2fpN6a_ZySI,1434
+themis/utils/tracing.py,sha256=VTeiRjcW_B5fOOoSeAp37nrmlwP1DiqPcoe6OtIQ7dk,8468
+themis_eval-0.1.1.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
+themis_eval-0.1.1.dist-info/METADATA,sha256=7-oAglt5HH_AWi7yCzvTq_RUE07xTIVmZG02IcxHRjM,23516
+themis_eval-0.1.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+themis_eval-0.1.1.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
+themis_eval-0.1.1.dist-info/RECORD,,

themis_eval-0.1.0.dist-info/RECORD

@@ -1,8 +0,0 @@
-themis/__init__.py,sha256=3bKi1PneI5PaTaDPXsArCVvfinkLFDRU91lvZIeg7V0,281
-themis/_version.py,sha256=tc4TJqWVv2dx4UzItLqneMPaG7vM8CQFDNW5pJgNoKg,345
-themis/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-themis_eval-0.1.0.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
-themis_eval-0.1.0.dist-info/METADATA,sha256=bRdc6UhSKYmptIJVhp4cEK8K2-Vvc77rZnVfYav0uS4,23516
-themis_eval-0.1.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-themis_eval-0.1.0.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
-themis_eval-0.1.0.dist-info/RECORD,,