benchkit-cli 0.1.0__py3-none-any.whl

benchkit/tui/theme.py

@@ -0,0 +1,135 @@
+"""Dogi themes for BenchKit.
+
+Ported from the Dogi editor theme by Doğukan Ürker
+(https://github.com/DogukanUrker/DogiZed). Dogi is a flat, high-contrast
+palette: pure black or pure white everywhere, no tinted panels or borders, and
+colour reserved for state. The TUI gets its separation from translucent
+overlays instead of solid panel colours, which is what `bk-raise` and
+`bk-border` carry.
+"""
+
+from __future__ import annotations
+
+from textual.theme import Theme
+
+# Dogi Dark
+DARK_BACKGROUND = "#000000"
+DARK_FOREGROUND = "#FFFFFF"
+DARK_BLUE = "#60A5FA"
+DARK_PINK = "#F472B6"
+DARK_CYAN = "#22D3EE"
+DARK_GREEN = "#34D399"
+DARK_YELLOW = "#FACC15"
+DARK_RED = "#EF4444"
+DARK_MUTED = "#6B7280"
+DARK_PLACEHOLDER = "#4A4A4A"
+
+# Dogi Light
+LIGHT_BACKGROUND = "#FFFFFF"
+LIGHT_FOREGROUND = "#0B0B0B"
+LIGHT_BLUE = "#2563EB"
+LIGHT_PINK = "#DB2777"
+LIGHT_CYAN = "#0891B2"
+LIGHT_GREEN = "#059669"
+LIGHT_YELLOW = "#D97706"
+LIGHT_RED = "#DC2626"
+LIGHT_MUTED = "#6B7280"
+LIGHT_PLACEHOLDER = "#9CA3AF"
+
+DOGI_DARK = Theme(
+    name="dogi",
+    primary=DARK_BLUE,
+    secondary=DARK_PINK,
+    accent=DARK_CYAN,
+    success=DARK_GREEN,
+    warning=DARK_YELLOW,
+    error=DARK_RED,
+    foreground=DARK_FOREGROUND,
+    background=DARK_BACKGROUND,
+    surface=DARK_BACKGROUND,
+    panel=DARK_BACKGROUND,
+    # Solid stand-ins for Dogi's translucent white overlay (#FFFFFF08).
+    boost="#141414",
+    dark=True,
+    variables={
+        # Dogi draws no hard borders; these translucent whites stand in.
+        "bk-border": "#FFFFFF 14%",
+        "bk-raise": "#FFFFFF 6%",
+        "bk-selection": "#FFFFFF 20%",
+        "text-muted": DARK_MUTED,
+        "text-disabled": DARK_PLACEHOLDER,
+        "input-cursor-background": DARK_BLUE,
+        "input-cursor-foreground": DARK_BACKGROUND,
+        "input-selection-background": f"{DARK_BLUE} 35%",
+        "block-cursor-background": DARK_BLUE,
+        "block-cursor-foreground": DARK_BACKGROUND,
+        "block-cursor-text-style": "bold",
+        "footer-key-foreground": DARK_BLUE,
+        "footer-description-foreground": DARK_FOREGROUND,
+        "scrollbar": "#FFFFFF 12%",
+        "scrollbar-hover": "#FFFFFF 20%",
+        "scrollbar-active": DARK_BLUE,
+        "scrollbar-background": DARK_BACKGROUND,
+        "link-color-hover": DARK_BLUE,
+        "markdown-table-border": "#FFFFFF 14%",
+    },
+)
+
+DOGI_LIGHT = Theme(
+    name="dogi-light",
+    primary=LIGHT_BLUE,
+    secondary=LIGHT_PINK,
+    accent=LIGHT_CYAN,
+    success=LIGHT_GREEN,
+    warning=LIGHT_YELLOW,
+    error=LIGHT_RED,
+    foreground=LIGHT_FOREGROUND,
+    background=LIGHT_BACKGROUND,
+    surface=LIGHT_BACKGROUND,
+    panel=LIGHT_BACKGROUND,
+    # Dogi's #00000010 hover overlay, flattened.
+    boost="#F0F0F0",
+    dark=False,
+    variables={
+        "bk-border": "#000000 14%",
+        "bk-raise": "#000000 5%",
+        "bk-selection": "#000000 18%",
+        "text-muted": LIGHT_MUTED,
+        "text-disabled": LIGHT_PLACEHOLDER,
+        "input-cursor-background": LIGHT_BLUE,
+        "input-cursor-foreground": LIGHT_BACKGROUND,
+        "input-selection-background": f"{LIGHT_BLUE} 25%",
+        "block-cursor-background": LIGHT_BLUE,
+        "block-cursor-foreground": LIGHT_BACKGROUND,
+        "block-cursor-text-style": "bold",
+        "footer-key-foreground": LIGHT_BLUE,
+        "footer-description-foreground": LIGHT_FOREGROUND,
+        "scrollbar": "#000000 15%",
+        "scrollbar-hover": "#000000 25%",
+        "scrollbar-active": LIGHT_BLUE,
+        "scrollbar-background": LIGHT_BACKGROUND,
+        "link-color-hover": LIGHT_BLUE,
+        "markdown-table-border": "#000000 14%",
+    },
+)
+
+THEMES = [DOGI_DARK, DOGI_LIGHT]
+
+DEFAULT_THEME = DOGI_DARK.name
+ALTERNATE_THEME = DOGI_LIGHT.name
+
+# Fallbacks so the stylesheet still parses under Textual's built-in themes,
+# which know nothing about the Dogi variables. Variable values are substituted
+# literally, so these have to be real colours rather than other variables.
+VARIABLE_DEFAULTS = {
+    "bk-border": "#808080 40%",
+    "bk-raise": "#808080 15%",
+    "bk-selection": "#808080 30%",
+}
+
+
+def score_palette(dark: bool) -> dict[str, str]:
+    """Pass/warn/fail colours that stay readable on the active background."""
+    if dark:
+        return {"high": DARK_GREEN, "mid": DARK_YELLOW, "low": DARK_RED}
+    return {"high": LIGHT_GREEN, "mid": LIGHT_YELLOW, "low": LIGHT_RED}

benchkit/tui/widgets.py

@@ -0,0 +1,120 @@
+"""Reusable presentational widgets for the BenchKit TUI."""
+
+from __future__ import annotations
+
+from typing import Any
+
+from textual.app import ComposeResult
+from textual.content import Content
+from textual.reactive import reactive
+from textual.screen import Screen
+from textual.widget import Widget
+from textual.widgets import Static
+
+
+class StatCard(Widget):
+    """A label above a large value, used for the live run statistics."""
+
+    DEFAULT_CSS = """
+    StatCard {
+        width: 1fr;
+        height: 4;
+        padding: 0 1;
+        border-left: outer $bk-border;
+        background: $surface;
+    }
+    StatCard .stat-label {
+        color: $text-muted;
+        text-style: none;
+        text-wrap: nowrap;
+        text-overflow: ellipsis;
+    }
+    StatCard .stat-value {
+        text-style: bold;
+        text-wrap: nowrap;
+        text-overflow: ellipsis;
+    }
+    StatCard .stat-hint {
+        color: $text-muted;
+        text-wrap: nowrap;
+        text-overflow: ellipsis;
+    }
+    """
+
+    value: reactive[str] = reactive("--")
+    hint: reactive[str] = reactive("")
+
+    def __init__(
+        self,
+        label: str,
+        value: str = "--",
+        hint: str = "",
+        *,
+        id: str | None = None,
+        classes: str | None = None,
+    ) -> None:
+        super().__init__(id=id, classes=classes)
+        self.label = label
+        self.set_reactive(StatCard.value, value)
+        self.set_reactive(StatCard.hint, hint)
+
+    def compose(self) -> ComposeResult:
+        yield Static(self.label, classes="stat-label")
+        yield Static(self.value, classes="stat-value", id="value")
+        yield Static(self.hint, classes="stat-hint", id="hint")
+
+    def watch_value(self, value: str) -> None:
+        if self.is_mounted:
+            self.query_one("#value", Static).update(value)
+
+    def watch_hint(self, hint: str) -> None:
+        if self.is_mounted:
+            self.query_one("#hint", Static).update(hint)
+
+    def set_state(self, value: str, hint: str = "") -> None:
+        self.value = value
+        self.hint = hint
+
+
+class Banner(Static):
+    """The BenchKit wordmark used on the connection screen."""
+
+    ART = r"""
+██████╗ ███████╗███╗   ██╗ ██████╗██╗  ██╗██╗  ██╗██╗████████╗
+██╔══██╗██╔════╝████╗  ██║██╔════╝██║  ██║██║ ██╔╝██║╚══██╔══╝
+██████╔╝█████╗  ██╔██╗ ██║██║     ███████║█████╔╝ ██║   ██║
+██╔══██╗██╔══╝  ██║╚██╗██║██║     ██╔══██║██╔═██╗ ██║   ██║
+██████╔╝███████╗██║ ╚████║╚██████╗██║  ██║██║  ██╗██║   ██║
+╚═════╝ ╚══════╝╚═╝  ╚═══╝ ╚═════╝╚═╝  ╚═╝╚═╝  ╚═╝╚═╝   ╚═╝
+""".strip("\n")
+
+    def __init__(self, **kwargs: Any) -> None:
+        super().__init__(self.ART, **kwargs)
+
+
+class SectionTitle(Static):
+    """A pane heading with an optional right-aligned counter."""
+
+    def __init__(self, title: str, detail: str = "", **kwargs: Any) -> None:
+        super().__init__(**kwargs)
+        self._title = title
+        self._detail = detail
+        self.update(self._render())
+
+    def _render(self) -> Content:
+        if self._detail:
+            return Content.from_markup(
+                "[b]$title[/b]  [dim]$detail[/dim]",
+                title=self._title,
+                detail=self._detail,
+            )
+        return Content.from_markup("[b]$title[/b]", title=self._title)
+
+    def set_detail(self, detail: str) -> None:
+        self._detail = detail
+        self.update(self._render())
+
+
+def apply_compact(screen: Screen, height: int) -> None:
+    """Tighten the stat cards and progress block on short terminals."""
+    screen.set_class(height < 32, "compact")

benchkit_cli-0.1.0.dist-info/METADATA

@@ -0,0 +1,294 @@
+Metadata-Version: 2.4
+Name: benchkit-cli
+Version: 0.1.0
+Summary: Benchmark your local LLMs with real eval suites
+License-Expression: Apache-2.0
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: evalplus==0.3.1
+Requires-Dist: httpx>=0.28.1
+Requires-Dist: pyarrow>=25.0.0
+Requires-Dist: rich>=15.0.0
+Requires-Dist: python-dotenv>=1.2.3
+Requires-Dist: textual>=8.2.8
+Provides-Extra: browser
+Requires-Dist: playwright>=1.63.0; extra == "browser"
+Dynamic: license-file
+
+<div align="center">
+
+<img src="assets/benchkit-connect.png" alt="BenchKit" width="880">
+
+# BenchKit
+
+**Benchmark local LLMs with real evaluation suites from a full terminal UI.**
+
+Not vibes. Actual scores.
+
+<img src="https://img.shields.io/badge/Python-3.11%2B-2563EB?style=flat-square&logo=python&logoColor=white&labelColor=0b0b0b" alt="Python 3.11+">
+<img src="https://img.shields.io/badge/TUI-Textual-60A5FA?style=flat-square&labelColor=0b0b0b" alt="Built with Textual">
+<img src="https://img.shields.io/badge/Suites-25-34D399?style=flat-square&labelColor=0b0b0b" alt="25 benchmark suites">
+<img src="https://img.shields.io/badge/License-Apache%202.0-6B7280?style=flat-square&labelColor=0b0b0b" alt="Apache 2.0">
+
+</div>
+
+## What it does
+
+BenchKit runs established coding, reasoning, knowledge, instruction-following,
+vision, and long-context benchmarks against local models. It supports OpenAI-compatible
+servers such as llama.cpp, llama-swap, vLLM, and LM Studio, plus native Ollama.
+
+- Guided terminal UI for connecting, selecting models, running suites, and
+  exploring results
+- Headless mode for scripts and CI
+- Live streamed responses, reasoning traces, progress, speed, and loop detection
+- Direct model evaluation and optional Pi coding-agent evaluation
+- Benchmark slicing, tag filters, verifier repair, and choice-order robustness
+- JSON, CSV, Markdown, and standalone interactive HTML reports
+- Performance profiling and a local dashboard for historical runs
+
+<div align="center">
+  <img src="assets/benchkit-run.png" alt="BenchKit running a benchmark" width="940">
+</div>
+
+## Quick start
+
+BenchKit requires Python 3.11 or newer and [uv](https://docs.astral.sh/uv/).
+
+```bash
+uvx benchkit-cli                # try it without installing
+uv tool install benchkit-cli    # or install the `benchkit` command
+```
+
+From a clone of this repository:
+
+```bash
+uv sync
+cp .env.example .env
+uv run benchkit
+```
+
+Set your server in a `.env` in the directory you run `benchkit` from (parent
+directories are not searched), or enter it on the Connect screen:
+
+```env
+BENCHKIT_PROVIDER=openai
+BENCHKIT_HOST=http://localhost:8080/v1
+BENCHKIT_API_KEY=
+```
+
+For Ollama:
+
+```env
+BENCHKIT_PROVIDER=ollama
+OLLAMA_HOST=http://localhost:11434
+```
+
+No inference server available? Explore the complete UI offline:
+
+```bash
+uv run benchkit --demo
+```
+
+## Run benchmarks
+
+The TUI lets you select models, benchmarks, task limits, and optional run modes.
+For automation, use headless mode:
+
+```bash
+# Run the first 20 HumanEval tasks
+uv run benchkit --headless --models qwen3:8b --benchmarks humaneval:20
+
+# Run multiple models and benchmarks
+uv run benchkit --headless \
+  --models qwen3:8b,gemma3:12b \
+  --benchmarks sanity,gsm8k:50
+
+# Print prompts, reasoning traces, and responses
+uv run benchkit --headless --models qwen3:8b \
+  --benchmarks sanity --verbose
+```
+
+Task slices accept `20` for the first 20 tasks, `-20` for the last 20, and
+`40-80` for a range.
+
+Useful commands:
+
+```bash
+uv run benchkit --list
+uv run benchkit --help
+```
+
+## Benchmarks
+
+| Benchmark | Key | Tasks | What it tests |
+| --- | --- | ---: | --- |
+| PatchEval | `patcheval` | 20 | Real Python bug fixes verified by hidden tests with Pi |
+| Sanity | `sanity` | 25 | Fast checks across five core capabilities |
+| Three.js Arena | `treejs-arena` | 25 | Creative single-file scenes, scored only on whether they render |
+| Git Surgery | `git-surgery` | 10 | Stateful Git operations with Pi |
+| MedFailBench | `medfailbench` | 30 | Offline heuristic screen for high-risk clinical responses; not clinical validation, certification, deployment evidence, or model ranking |
+| Aider Polyglot | `aider-polyglot` | 225 | Repository editing across six languages with Pi |
+| HumanEval | `humaneval` | 164 | Python function completion |
+| HumanEval+ | `humaneval-plus` | 164 | HumanEval with expanded EvalPlus tests |
+| MBPP | `mbpp` | 500 | Short Python programming tasks |
+| MBPP+ | `mbpp-plus` | 378 | MBPP with expanded EvalPlus tests |
+| GSM8K | `gsm8k` | 1,319 | Multi-step grade-school math |
+| IFEval | `ifeval` | 541 | Verifiable instruction following |
+| RULER | `ruler` | 39 × 6 | Practical long-context retrieval from 4K to 128K |
+| RULER Full | `ruler-full` | 6,500 × 6 | Research-scale RULER (explicit opt-in) |
+| XSTest | `xstest` | 450 | Safe compliance and unsafe refusal with an offline checker |
+| GPQA | `gpqa` | 198 | Graduate-level science questions |
+| MMLU-Pro | `mmlu-pro` | 12,032 | Reasoning across 14 knowledge categories |
+| BANKING77 | `banking77` | 3,080 | Route a banking message to one of 77 intents |
+| Code Verifier | `code-verifier` | 500 | Judge whether a model-written HumanEval+ solution passes its tests |
+| OCRBench | `ocrbench` | 1,000 | Text recognition and text-centric VQA over images |
+| ChartQA | `chartqa` | 2,500 | Questions about real charts, 5% relaxed numeric accuracy |
+| MMVP | `mmvp` | 300 | Look-alike image pairs that differ in one detail; also reports pair accuracy |
+| V* Bench | `vstar` | 191 | Small details in high-resolution images |
+| MMLU | `mmlu` | 14,042 | Academic and professional knowledge |
+| ARC | `arc` | 1,172 | Grade-school science reasoning |
+| OpenBookQA | `openbookqa` | 500 | Elementary science knowledge |
+| WinoGrande | `winogrande` | 1,267 | Commonsense pronoun resolution |
+| PIQA | `piqa` | 1,838 | Physical commonsense reasoning |
+| BoolQ | `boolq` | 3,270 | Passage-based yes/no questions |
+| TruthfulQA | `truthfulqa` | 817 | Resistance to common misconceptions |
+| HellaSwag | `hellaswag` | 1,000 | Plausible real-world continuations |
+
+`treejs-arena` renders each generated page in headless Chromium, so it needs
+the browser extra first:
+
+```bash
+uv sync --extra browser
+uv run playwright install chromium
+```
+
+Vision suites (`ocrbench`, `chartqa`, `mmvp`, `vstar`) send images inline as
+base64, so they need a vision model (a llama.cpp server started with
+`--mmproj`, or an Ollama vision model). Each one downloads a pinned dataset
+revision on first use. `mmvp` and `vstar` are small enough to run in full on
+one consumer GPU and also run on decision models with an mmproj.
+
+Small suites ship with BenchKit and work offline. Larger datasets (`mmlu`,
+`mmlu-pro`, the vision suites, EvalPlus, RULER's haystacks, Aider Polyglot,
+PatchEval) are
+downloaded once into `~/.cache/benchkit` the first time a run needs them. List
+what is installed and fetch everything that is missing ahead of time with:
+
+```bash
+uv run benchkit datasets
+uv run benchkit datasets install            # every missing dataset
+uv run benchkit datasets install mmlu       # or just the ones you name
+```
+
+In the terminal app, press `d` on the setup screen (or click **Datasets**) for
+the same list with an **Install all missing** button.
+
+The CLI registry is the canonical source for current counts, descriptions,
+and supported perturbations:
+
+```bash
+uv run benchkit --list
+```
+
+## Advanced runs
+
+Compare raw generation with the stock Pi coding agent (Docker required):
+
+```bash
+uv run benchkit --headless --models qwen3:8b \
+  --benchmarks gsm8k:20 --harness both
+```
+
+Give incorrect answers one verifier-guided replacement attempt:
+
+```bash
+uv run benchkit --headless --models qwen3:8b \
+  --benchmarks gsm8k:20 --repair-attempts 1
+```
+
+Repairs default to off and can be set from 0–10. One is a practical starting
+point, but the right depth depends on what you want to measure.
+
+Test whether multiple-choice accuracy survives reordered answer choices:
+
+```bash
+uv run benchkit --headless --models qwen3:8b \
+  --benchmarks mmlu-pro:100 --perturbation choice-order
+```
+
+Decision models served by llama.cpp (`/v1/systemone`) are detected
+automatically and run on the multiple-choice and yes/no suites, scored on
+accuracy plus calibration (Brier, ECE):
+
+```bash
+llama-server -hf ggml-org/Kev-4B-GGUF
+uv run benchkit --headless --models <decision-model> \
+  --benchmarks mmlu-pro:100,boolq:100
+```
+
+Behind a proxy that hides this, such as llama-swap (v262+), name the model:
+
+```bash
+uv run benchkit --headless --models d1-3b,qwen3:8b --benchmarks gpqa \
+  --decision-models d1-3b
+```
+
+or mark it once in the llama-swap config:
+
+```yaml
+models:
+  d1-3b:
+    metadata:
+      decision: true
+```
+
+Profile inference performance:
+
+```bash
+uv run benchkit perf qwen3:8b
+uv run benchkit perf qwen3:8b --depths minimal,4k,16k --gen 256 --reps 10
+```
+
+## Results
+
+Every run creates a timestamped directory containing:
+
+```text
+results/<timestamp>/
+├── results.json
+├── results.csv
+├── results.md
+└── results.html
+```
+
+Open `results.html` for an interactive report with charts and task-level
+prompts, responses, reasoning traces, timing, and diagnostics.
+
+Browse completed benchmark and performance runs locally:
+
+```bash
+uv run benchkit history
+```
+
+## Development
+
+```bash
+uv sync
+uv run pre-commit install
+uv run ruff check .
+uv run ruff format .
+uv run pytest
+```
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for contributor setup and workflow.
+
+## License
+
+Apache License 2.0
+
+<div align="center">
+<br>
+<sub>Themed with <a href="https://github.com/DogukanUrker/DogiZed">Dogi</a> · built on <a href="https://github.com/Textualize/textual">Textual</a></sub>
+</div>