benchgr-agent 0.1.0__tar.gz

benchgr_agent-0.1.0/PKG-INFO

@@ -0,0 +1,31 @@
+Metadata-Version: 2.4
+Name: benchgr-agent
+Version: 0.1.0
+Summary: BenchGR GPU Benchmark Agent - Test your GPU and submit to the leaderboard
+Author: San S
+License: MIT
+Project-URL: Homepage, https://benchgr-frontend.vercel.app
+Keywords: gpu,benchmark,leaderboard,cuda,ai
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: torch>=2.0.0
+Requires-Dist: transformers>=4.30.0
+Requires-Dist: requests>=2.28.0
+Requires-Dist: numpy>=1.24.0
+
+# benchgr-agent
+
+GPU Benchmark Agent for [BenchGR Leaderboard](https://benchgr-frontend.vercel.app).
+
+## Install
+pip install benchgr-agent
+
+## Usage
+# Check your GPU
+benchgr info
+
+# Run benchmark (no submit)
+benchgr run --no-submit
+
+# Run and submit to leaderboard
+benchgr run --api-key YOUR_API_KEY

benchgr_agent-0.1.0/README.md

@@ -0,0 +1,16 @@
+# benchgr-agent
+
+GPU Benchmark Agent for [BenchGR Leaderboard](https://benchgr-frontend.vercel.app).
+
+## Install
+pip install benchgr-agent
+
+## Usage
+# Check your GPU
+benchgr info
+
+# Run benchmark (no submit)
+benchgr run --no-submit
+
+# Run and submit to leaderboard
+benchgr run --api-key YOUR_API_KEY

benchgr_agent-0.1.0/benchgr_agent/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.1.0"

benchgr_agent-0.1.0/benchgr_agent/benchmark.py

@@ -0,0 +1,111 @@
+import time
+import torch
+import numpy as np
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+BACKEND_URL = "https://benchgr-backend.vercel.app"
+MODEL_ID = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+
+
+def get_gpu_info():
+    if not torch.cuda.is_available():
+        return None, "No CUDA GPU detected. Please run on a CUDA-enabled GPU."
+    
+    gpu_name = torch.cuda.get_device_name(0)
+    vram_total = torch.cuda.get_device_properties(0).total_memory / (1024 ** 3)
+    return {"gpu_name": gpu_name, "vram_gb": round(vram_total, 2)}, None
+
+
+def run_inference_benchmark():
+    print("📥 Loading TinyLlama 1.1B (first run may take 1-2 mins to download ~600MB)...")
+    
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        torch_dtype=torch.float16,
+        device_map="cuda"
+    )
+    model.eval()
+
+    prompt = "Explain what a GPU does in simple terms."
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+
+    # Warmup pass
+    with torch.no_grad():
+        model.generate(**inputs, max_new_tokens=20)
+
+    # Actual benchmark — 3 runs averaged
+    total_tokens = 0
+    total_time = 0.0
+    runs = 3
+
+    print("⚡ Running inference benchmark (3 passes)...")
+    for i in range(runs):
+        torch.cuda.synchronize()
+        start = time.perf_counter()
+        with torch.no_grad():
+            output = model.generate(**inputs, max_new_tokens=100)
+        torch.cuda.synchronize()
+        end = time.perf_counter()
+
+        tokens_generated = output.shape[1] - inputs["input_ids"].shape[1]
+        total_tokens += tokens_generated
+        total_time += (end - start)
+        print(f"   Pass {i+1}: {tokens_generated / (end - start):.1f} tok/s")
+
+    tokens_per_sec = round(total_tokens / total_time, 2)
+    return tokens_per_sec
+
+
+def run_cuda_benchmark():
+    print("🔢 Running CUDA matrix benchmark...")
+    size = 4096
+    a = torch.randn(size, size, device="cuda", dtype=torch.float16)
+    b = torch.randn(size, size, device="cuda", dtype=torch.float16)
+
+    # Warmup
+    torch.matmul(a, b)
+    torch.cuda.synchronize()
+
+    runs = 10
+    start = time.perf_counter()
+    for _ in range(runs):
+        torch.matmul(a, b)
+    torch.cuda.synchronize()
+    end = time.perf_counter()
+
+    avg_ms = ((end - start) / runs) * 1000
+    # TFLOPS = (2 * N^3) / (time_in_seconds * 1e12)
+    tflops = round((2 * size**3) / ((avg_ms / 1000) * 1e12), 2)
+    return tflops
+
+
+def run_memory_bandwidth_benchmark():
+    print("💾 Running memory bandwidth benchmark...")
+    size = 256 * 1024 * 1024 // 4  # 256MB of float32
+    a = torch.ones(size, device="cuda", dtype=torch.float32)
+    b = torch.empty(size, device="cuda", dtype=torch.float32)
+
+    torch.cuda.synchronize()
+    start = time.perf_counter()
+    for _ in range(5):
+        b.copy_(a)
+    torch.cuda.synchronize()
+    end = time.perf_counter()
+
+    mem_bw_gbps = round((5 * 256 * 2) / (end - start), 2)  # GB/s (read + write)
+    return mem_bw_gbps
+
+
+def submit_results(api_key, results):
+    import requests
+    try:
+        resp = requests.post(
+            f"{BACKEND_URL}/api/results/submit",
+            params={"api_key": api_key},
+            json=results,
+            timeout=10
+        )
+        return resp.json()
+    except Exception as e:
+        return {"error": str(e)}

benchgr_agent-0.1.0/benchgr_agent/main.py

@@ -0,0 +1,96 @@
+import argparse
+import sys
+from benchgr_agent.benchmark import (
+    get_gpu_info,
+    run_inference_benchmark,
+    run_cuda_benchmark,
+    run_memory_bandwidth_benchmark,
+    submit_results
+)
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        prog="benchgr",
+        description="BenchGR — GPU Benchmark Agent"
+    )
+    subparsers = parser.add_subparsers(dest="command")
+
+    # benchgr run
+    run_parser = subparsers.add_parser("run", help="Run benchmark and optionally submit")
+    run_parser.add_argument("--api-key", type=str, help="Your BenchGR API key", default=None)
+    run_parser.add_argument("--no-submit", action="store_true", help="Run without submitting results")
+
+    # benchgr info
+    subparsers.add_parser("info", help="Show detected GPU info")
+
+    args = parser.parse_args()
+
+    if args.command == "info":
+        gpu, err = get_gpu_info()
+        if err:
+            print(f"❌ {err}")
+        else:
+            print(f"✅ GPU: {gpu['gpu_name']}")
+            print(f"   VRAM: {gpu['vram_gb']} GB")
+        return
+
+    if args.command == "run":
+        print("=" * 50)
+        print("       🚀 BenchGR GPU Benchmark Agent")
+        print("=" * 50)
+
+        gpu, err = get_gpu_info()
+        if err:
+            print(f"❌ {err}")
+            sys.exit(1)
+
+        print(f"✅ Detected GPU: {gpu['gpu_name']} ({gpu['vram_gb']} GB VRAM)\n")
+
+        tokens_per_sec = run_inference_benchmark()
+        tflops = run_cuda_benchmark()
+        mem_bw = run_memory_bandwidth_benchmark()
+
+        results = {
+            "gpu_name": gpu["gpu_name"],
+            "vram_gb": gpu["vram_gb"],
+            "tokens_per_sec": tokens_per_sec,
+            "tflops_fp16": tflops,
+            "memory_bw_gbps": mem_bw,
+        }
+
+        print("\n" + "=" * 50)
+        print("📊 BENCHMARK RESULTS")
+        print("=" * 50)
+        print(f"  GPU         : {gpu['gpu_name']}")
+        print(f"  Tokens/sec  : {tokens_per_sec} tok/s")
+        print(f"  TFLOPS FP16 : {tflops} TFLOPS")
+        print(f"  Mem BW      : {mem_bw} GB/s")
+        print("=" * 50)
+
+        if not args.no_submit:
+            if not args.api_key:
+                print("\n⚠️  No API key provided. Get yours at:")
+                print("   👉 https://benchgr-frontend.vercel.app/dashboard")
+                print("\nRun again with:")
+                print("   benchgr run --api-key YOUR_KEY_HERE")
+                return
+
+            print("\n📤 Submitting to leaderboard...")
+            response = submit_results(args.api_key, results)
+            if "error" in response:
+                print(f"❌ Submission failed: {response['error']}")
+            else:
+                print(f"✅ Submitted! Check your rank at:")
+                print(f"   👉 https://benchgr-frontend.vercel.app/leaderboard")
+        else:
+            print("\n✅ Done! (Results not submitted — remove --no-submit to submit)")
+
+    else:
+        print("Usage: benchgr run --api-key YOUR_KEY")
+        print("       benchgr info")
+        print("       benchgr run --no-submit")
+
+
+if __name__ == "__main__":
+    main()

benchgr_agent-0.1.0/benchgr_agent.egg-info/PKG-INFO

@@ -0,0 +1,31 @@
+Metadata-Version: 2.4
+Name: benchgr-agent
+Version: 0.1.0
+Summary: BenchGR GPU Benchmark Agent - Test your GPU and submit to the leaderboard
+Author: San S
+License: MIT
+Project-URL: Homepage, https://benchgr-frontend.vercel.app
+Keywords: gpu,benchmark,leaderboard,cuda,ai
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: torch>=2.0.0
+Requires-Dist: transformers>=4.30.0
+Requires-Dist: requests>=2.28.0
+Requires-Dist: numpy>=1.24.0
+
+# benchgr-agent
+
+GPU Benchmark Agent for [BenchGR Leaderboard](https://benchgr-frontend.vercel.app).
+
+## Install
+pip install benchgr-agent
+
+## Usage
+# Check your GPU
+benchgr info
+
+# Run benchmark (no submit)
+benchgr run --no-submit
+
+# Run and submit to leaderboard
+benchgr run --api-key YOUR_API_KEY

benchgr_agent-0.1.0/benchgr_agent.egg-info/SOURCES.txt

@@ -0,0 +1,11 @@
+README.md
+pyproject.toml
+benchgr_agent/__init__.py
+benchgr_agent/benchmark.py
+benchgr_agent/main.py
+benchgr_agent.egg-info/PKG-INFO
+benchgr_agent.egg-info/SOURCES.txt
+benchgr_agent.egg-info/dependency_links.txt
+benchgr_agent.egg-info/entry_points.txt
+benchgr_agent.egg-info/requires.txt
+benchgr_agent.egg-info/top_level.txt

benchgr_agent-0.1.0/benchgr_agent.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

benchgr_agent-0.1.0/benchgr_agent.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+benchgr = benchgr_agent.main:main

benchgr_agent-0.1.0/benchgr_agent.egg-info/requires.txt

@@ -0,0 +1,4 @@
+torch>=2.0.0
+transformers>=4.30.0
+requests>=2.28.0
+numpy>=1.24.0

benchgr_agent-0.1.0/benchgr_agent.egg-info/top_level.txt

@@ -0,0 +1 @@
+benchgr_agent

benchgr_agent-0.1.0/pyproject.toml

@@ -0,0 +1,27 @@
+[build-system]
+requires = ["setuptools>=61.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "benchgr-agent"
+version = "0.1.0"
+description = "BenchGR GPU Benchmark Agent - Test your GPU and submit to the leaderboard"
+readme = "README.md"
+requires-python = ">=3.8"
+license = {text = "MIT"}
+authors = [
+  {name = "San S"}
+]
+keywords = ["gpu", "benchmark", "leaderboard", "cuda", "ai"]
+dependencies = [
+    "torch>=2.0.0",
+    "transformers>=4.30.0",
+    "requests>=2.28.0",
+    "numpy>=1.24.0",
+]
+
+[project.scripts]
+benchgr = "benchgr_agent.main:main"
+
+[project.urls]
+Homepage = "https://benchgr-frontend.vercel.app"

benchgr_agent-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+