Update app.py

app.py

@@ -21,8 +21,35 @@ def main():
     print(f" num_gpu = {num_gpu}, prompt_size = {prompt_size} tokens, response_size = {response_size} tokens")
     print(f" n_concurrent_request = {n_concurrent_request}, avg_context_window = {avg_context_window} tokens")
 
-    # Resto del codice rimane invariato...
-
+    # Define variables
+    gpu_specs = [
+        {"name": "A10", "fp16_tflops": 125, "memory_gb": 24, "memory_bandwidth_gbps": 600},
+        {"name": "A30", "fp16_tflops": 330, "memory_gb": 24, "memory_bandwidth_gbps": 933},
+        {"name": "L40", "fp16_tflops": 181, "memory_gb": 48, "memory_bandwidth_gbps": 864},
+        {"name": "L40s", "fp16_tflops": 362, "memory_gb": 48, "memory_bandwidth_gbps": 864},
+        {"name": "A100 40 GB", "fp16_tflops": 312, "memory_gb": 40, "memory_bandwidth_gbps": 1555},
+        {"name": "A100 40 GB SXM", "fp16_tflops": 312, "memory_gb": 40, "memory_bandwidth_gbps": 1555},
+        {"name": "A100 80 GB PCIe", "fp16_tflops": 312, "memory_gb": 80, "memory_bandwidth_gbps": 1935},
+        {"name": "A100 80 GB SXM", "fp16_tflops": 312, "memory_gb": 80, "memory_bandwidth_gbps": 2039},
+        {"name": "H100 PCIe", "fp16_tflops": 1513, "memory_gb": 80, "memory_bandwidth_gbps": 2000},
+        {"name": "H100 SXM", "fp16_tflops": 1979, "memory_gb": 80, "memory_bandwidth_gbps": 3350},
+        {"name": "H100 NVL", "fp16_tflops": 3958, "memory_gb": 188, "memory_bandwidth_gbps": 7800}
+        # Add or comment out GPU types as needed
+    ]
+
+    model_specs = [
+        {"name": "Llama-3-8B", "params_billion": 8, "d_model": 4096, "n_heads": 32, "n_layers": 32, "max_context_window": 8192, "d_head": 128},
+        {"name": "Llama-3-70B", "params_billion": 70, "d_model": 8192, "n_heads": 64, "n_layers": 80, "max_context_window": 8192, "d_head": 128},
+        {"name": "Llama-3.1-8B", "params_billion": 8, "d_model": 4096, "n_heads": 32, "n_layers": 32, "max_context_window": 131072, "d_head": 128},
+        {"name": "Llama-3.1-70B", "params_billion": 70, "d_model": 8192, "n_heads": 64, "n_layers": 80, "max_context_window": 131072, "d_head": 128},
+        {"name": "Mistral-7B-v0.3", "params_billion": 7, "d_model": 4096, "n_heads": 32, "n_layers": 32, "max_context_window": 32768, "d_head": 128},
+        {"name": "Falcon-7B", "params_billion": 7, "d_model": 4544, "n_heads": 71, "n_layers": 32, "max_context_window": 2048, "d_head": 64},
+        {"name": "Falcon-40B", "params_billion": 40, "d_model": 8192, "n_heads": 128, "n_layers": 60, "max_context_window": 2048, "d_head": 64},
+        {"name": "Falcon-180B", "params_billion": 180, "d_model": 14848, "n_heads": 232, "n_layers": 80, "max_context_window": 2048, "d_head": 64}
+        # Add or comment out model specifications as needed
+    ]
+
+    BYTES_IN_GB = 1_073_741_824  # 1 GB = 1,073,741,824 bytes
     def calc_kv_cache_size_per_token(n_layers, d_model):
         return 2 * 2 * n_layers * d_model / BYTES_IN_GB  # GB/token