Simultaneous fwd/bwd in peft finetuning + layer-level finetuning

benchmarking/add_arrival_times.py

@@ -0,0 +1,41 @@
+import numpy as np
+
+def generate_poisson_arrivals(n, arrival_rate):
+    """
+    Generate arrival times for n requests following a Poisson process.
+
+    Parameters:
+    n (int): Number of requests to generate
+    arrival_rate (float): Average arrival rate (requests per second)
+
+    Returns:
+    numpy.ndarray: Array of arrival times in seconds
+    """
+    # Generate n exponentially distributed inter-arrival times
+    # For a Poisson process, inter-arrival times follow exponential distribution
+    inter_arrival_times = np.random.exponential(scale=1/arrival_rate, size=n)
+
+    # Calculate cumulative sum to get arrival times
+    arrival_times = np.cumsum(inter_arrival_times)
+
+    # Round to 6 decimal places for practical purposes (microsecond precision)
+    arrival_times = np.round(arrival_times, decimals=6)
+
+    return arrival_times
+
+# Example usage
+if __name__ == "__main__":
+    # Example: Generate 10 arrivals with rate of 2 requests per second
+    n_requests = 250
+    rate = 0.1  # requests per second
+
+    arrivals = generate_poisson_arrivals(n_requests, rate)
+
+    print("Arrival times (seconds):")
+    for i, time in enumerate(arrivals, 1):
+        print(f"Request {i}: {time:.6f} s")
+
+    # Calculate average inter-arrival time to verify
+    avg_inter_arrival = np.diff(arrivals).mean()
+    print(f"\nAverage inter-arrival time: {avg_inter_arrival:.6f} s")
+    print(f"Expected inter-arrival time: {1/rate:.6f} s")