fairbench-ai 0.3.0__tar.gz

fairbench_ai-0.3.0/LICENSE

@@ -0,0 +1,99 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work.
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      revisions are patentable; regardless of whether the work is made
+      available under the License.
+
+      "Contribution" shall mean any Work of authorship that is submitted
+      by a Contributor.
+
+      "Contributor" shall mean any individual or entity that creates or
+      contributes to the Work.
+
+      "Submission" shall mean any original work of authorship that is
+      submitted to the Work for inclusion in the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, sublicensable, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and Derivative Works thereof.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, sublicensable, royalty-free, irrevocable
+      patent license to make, have made, use, offer to sell, sell, import,
+      and otherwise transfer the Work.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, provided that you meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work, excluding
+          those notices that do not pertain to any part of the Derivative
+          Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding any notices that do not
+          pertain to any part of the Derivative Works.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution submitted is under this License.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, the Work is provided on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied.
+
+   8. Limitation of Liability. In no event shall the Licensor be liable
+      for any damages arising from use of the Work.
+
+   9. Acceptance. By using, copying, modifying, or distributing the Work,
+      You accept this License.

fairbench_ai-0.3.0/MANIFEST.in

@@ -0,0 +1,3 @@
+include LICENSE
+include README.md
+recursive-include fairbench/tests *.py

fairbench_ai-0.3.0/PKG-INFO

@@ -0,0 +1,65 @@
+Metadata-Version: 2.4
+Name: fairbench-ai
+Version: 0.3.0
+Summary: FairBench - AI model benchmarking suite with SWE-bench integration
+Home-page: https://github.com/AlessandroBernini/fairbench
+Author: Alessandro Bernini
+Author-email: alessandro@mair.it
+License: Apache-2.0
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: click>=8.1
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: requests>=2.31
+Requires-Dist: psutil>=5.9
+Provides-Extra: ml
+Requires-Dist: lm-eval>=0.4; extra == "ml"
+Requires-Dist: llama-cpp-python>=0.2; extra == "ml"
+Provides-Extra: swebench
+Requires-Dist: swebench>=2.0; extra == "swebench"
+Requires-Dist: datasets>=2.14; extra == "swebench"
+Requires-Dist: rich>=13.0; extra == "swebench"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: ruff>=0.1; extra == "dev"
+Requires-Dist: mypy>=1.0; extra == "dev"
+Dynamic: author
+Dynamic: author-email
+Dynamic: home-page
+Dynamic: requires-python
+
+# FairBench
+
+AI model benchmarking suite with SWE-bench integration.
+
+## Quick Start
+
+```bash
+pip install fair-bench
+fairbench run -m model.gguf
+```
+
+## Benchmarks
+
+- **Quality**: MMLU, HellaSwag, GSM8K, HumanEval, IFEval, TruthfulQA, ARC-Challenge, LiveBench
+- **Speed**: TTFT, TPOT, throughput (llama-cpp-python)
+- **Memory**: VRAM peak, RAM peak
+- **SWE-bench**: GitHub issue resolution capability
+
+## Configuration
+
+Edit `~/.fairbench/config.yaml`:
+
+```yaml
+api_url: https://fairbench.vercel.app/api
+default_n_ctx: 4096
+default_n_gpu_layers: -1
+default_quant: q4_k_m
+default_threads: 0
+publish: true
+results_dir: /home/jarvispi/.fairbench/results
+```
+
+## Publishing
+
+By default, every benchmark run is **automatically published** to the public leaderboard. Set `publish: false` to disable.

fairbench_ai-0.3.0/README.md

@@ -0,0 +1,35 @@
+# FairBench
+
+AI model benchmarking suite with SWE-bench integration.
+
+## Quick Start
+
+```bash
+pip install fair-bench
+fairbench run -m model.gguf
+```
+
+## Benchmarks
+
+- **Quality**: MMLU, HellaSwag, GSM8K, HumanEval, IFEval, TruthfulQA, ARC-Challenge, LiveBench
+- **Speed**: TTFT, TPOT, throughput (llama-cpp-python)
+- **Memory**: VRAM peak, RAM peak
+- **SWE-bench**: GitHub issue resolution capability
+
+## Configuration
+
+Edit `~/.fairbench/config.yaml`:
+
+```yaml
+api_url: https://fairbench.vercel.app/api
+default_n_ctx: 4096
+default_n_gpu_layers: -1
+default_quant: q4_k_m
+default_threads: 0
+publish: true
+results_dir: /home/jarvispi/.fairbench/results
+```
+
+## Publishing
+
+By default, every benchmark run is **automatically published** to the public leaderboard. Set `publish: false` to disable.

fairbench_ai-0.3.0/fairbench/__init__.py

@@ -0,0 +1,4 @@
+"""FairBench - AI model benchmarking suite with SWE-bench integration."""
+
+__version__ = "0.3.0"
+__all__ = ["backend", "cli", "config", "data", "engine", "quality", "reporter"]

fairbench_ai-0.3.0/fairbench/backend.py

@@ -0,0 +1,211 @@
+"""Speed and memory benchmarking for AI models."""
+
+from __future__ import annotations
+
+import os
+import subprocess
+import time
+
+import psutil
+
+from fairbench.types import MemoryResult, SpeedResult
+
+
+def benchmark_speed(
+    model_path: str,
+    prompt: str = "The quick brown fox jumps over the lazy dog. " * 10,
+    n_ctx: int = 4096,
+    n_gpu_layers: int = -1,
+    n_threads: int = 0,
+    runs: int = 3,
+) -> SpeedResult:
+    """Measure TTFT, TPOT, and throughput using llama-cpp-python bindings.
+
+    Falls back to llama-bench CLI if bindings unavailable.
+    """
+    try:
+        from llama_cpp import Llama as _Llama  # type: ignore
+    except ImportError:
+        return _benchmark_with_llama_bench(model_path, prompt, runs)
+
+    return _benchmark_with_llama_cpp(
+        model_path, prompt, n_ctx, n_gpu_layers, n_threads, runs, _Llama
+    )
+
+
+def _benchmark_with_llama_cpp(
+    model_path: str,
+    prompt: str,
+    n_ctx: int,
+    n_gpu_layers: int,
+    n_threads: int,
+    runs: int,
+    Llama: type,  # type: ignore[type-arg]
+) -> SpeedResult:
+    """Benchmark using llama-cpp-python directly."""
+    ttfts: list[float] = []
+    tpots: list[float] = []
+    throughputs: list[float] = []
+
+    for i in range(runs):
+        try:
+            llm = Llama(  # type: ignore[no-untyped-call]
+                model_path=model_path,
+                n_ctx=n_ctx,
+                n_gpu_layers=n_gpu_layers,
+                n_threads=n_threads if n_threads > 0 else None,
+                verbose=False,
+            )
+
+            start = time.time()
+            response = llm.create_completion(
+                prompt=prompt,
+                max_tokens=512,
+                temperature=0.7,
+                stream=True,
+            )
+
+            ttft = time.time() - start
+            ttfts.append(ttft * 1000)  # ms
+
+            token_count = 0
+            last_time = time.time()
+            for chunk in response:
+                if time.time() - last_time > 0.01:
+                    elapsed = time.time() - last_time
+                    tpots.append((token_count / elapsed) if elapsed > 0 else 0)
+                    token_count = 0
+                    last_time = time.time()
+                token_count += 1
+
+            total_time = time.time() - start
+            throughput = token_count / total_time if total_time > 0 else 0
+            throughputs.append(throughput)
+
+            llm.close()
+        except (OSError, RuntimeError) as e:
+            print(f"    Run {i + 1} failed: {e}")
+
+    return SpeedResult(
+        ttft_avg=sum(ttfts) / len(ttfts) if ttfts else 0.0,
+        tpot_avg=sum(tpots) / len(tpots) if tpots else 0.0,
+        throughput_avg=sum(throughputs) / len(throughputs) if throughputs else 0.0,
+        runs=len(ttfts),
+    )
+
+
+def _benchmark_with_llama_bench(model_path: str, prompt: str, runs: int) -> SpeedResult:
+    """Fallback: use llama-bench CLI tool."""
+    try:
+        cmd = [
+            "llama-bench",
+            "-m",
+            model_path,
+            "-n",
+            "512",
+            "-ngl",
+            "0",
+            "-p",
+            str(len(prompt.split())),
+            "-t",
+            "0",
+            "-r",
+            str(runs),
+        ]
+        result = subprocess.run(cmd, check=False, capture_output=True, text=True, timeout=300)
+        # Parse output - llama-bench format varies by version
+        output = result.stdout
+        return SpeedResult(runs=runs, metadata={"raw_output": output})
+    except (FileNotFoundError, subprocess.TimeoutExpired) as e:
+        return SpeedResult(runs=runs, metadata={"error": str(e)})
+
+
+def benchmark_memory(
+    model_path: str,
+    prompt: str = "The quick brown fox jumps over the lazy dog. " * 10,
+    n_ctx: int = 4096,
+    n_gpu_layers: int = -1,
+    n_threads: int = 0,
+) -> MemoryResult:
+    """Measure VRAM and RAM peak usage during inference."""
+    try:
+        from llama_cpp import Llama as _Llama  # type: ignore
+    except ImportError:
+        return MemoryResult(metadata={"reason": "llama-cpp-python not installed"})
+
+    return _measure_memory_llama_cpp(
+        model_path,
+        prompt,
+        n_ctx,
+        n_gpu_layers,
+        n_threads,
+        _Llama,  # type: ignore[arg-type]
+    )
+
+
+def _measure_memory_llama_cpp(
+    model_path: str,
+    prompt: str,
+    n_ctx: int,
+    n_gpu_layers: int,
+    n_threads: int,
+    Llama: type,  # type: ignore[type-arg]
+) -> MemoryResult:
+    """Measure memory using llama-cpp-python with peak tracking."""
+    vram_samples: list[float] = []
+    ram_samples: list[float] = []
+
+    try:
+        # Get initial memory state
+        initial_ram = psutil.Process().memory_info().rss
+
+        llm = Llama(  # type: ignore[no-untyped-call]
+            model_path=model_path,
+            n_ctx=n_ctx,
+            n_gpu_layers=n_gpu_layers,
+            n_threads=n_threads if n_threads > 0 else None,
+            verbose=False,
+        )
+
+        # Measure VRAM via nvidia-smi if available
+        if os.path.exists("/usr/bin/nvidia-smi"):
+            result = subprocess.run(
+                [
+                    "nvidia-smi",
+                    "--query-gpu=memory.used",
+                    "--format=csv,noheader,nounits",
+                ],
+                check=False,
+                capture_output=True,
+                text=True,
+            )
+            if result.returncode == 0:
+                vram_samples = [
+                    float(line.strip())
+                    for line in result.stdout.strip().split("\n")
+                    if line.strip()
+                ]
+
+        # Run inference
+        llm.create_completion(
+            prompt=prompt,
+            max_tokens=256,
+            temperature=0.7,
+        )
+
+        # Peak RAM
+        current_ram = psutil.Process().memory_info().rss
+        ram_peak = (current_ram - initial_ram) / (1024 * 1024)  # MB
+        ram_samples.append(ram_peak)
+
+        llm.close()
+
+        return MemoryResult(
+            vram_peak_mb=sum(vram_samples) / len(vram_samples) if vram_samples else 0.0,
+            ram_peak_mb=ram_peak,
+            vram_avg_mb=sum(vram_samples) / len(vram_samples) if vram_samples else 0.0,
+            ram_avg_mb=sum(ram_samples) / len(ram_samples) if ram_samples else 0.0,
+            metadata={"model_size_gb": os.path.getsize(model_path) / (1024**3)},
+        )
+    except (OSError, RuntimeError) as e:
+        return MemoryResult(metadata={"error": str(e)})