68 lines
1.6 KiB
JSON
68 lines
1.6 KiB
JSON
[
|
|
{
|
|
"build_commit": "baseline",
|
|
"backends": "Metal",
|
|
"cpu_info": "Apple M4 Max",
|
|
"gpu_info": "Apple M4 Max",
|
|
"model_filename": "gemma-4-31b-q4.gguf",
|
|
"model_type": "31B",
|
|
"model_size": 20000000000,
|
|
"model_n_params": 30700000000,
|
|
"n_batch": 2048,
|
|
"n_ubatch": 512,
|
|
"n_threads": 12,
|
|
"type_k": "f16",
|
|
"type_v": "f16",
|
|
"n_gpu_layers": 999,
|
|
"n_cpu_moe": 0,
|
|
"split_mode": "layer",
|
|
"main_gpu": 0,
|
|
"no_kv_offload": false,
|
|
"flash_attn": -1,
|
|
"attention_kernel": "parallel",
|
|
"devices": "Metal0",
|
|
"tensor_split": "0.00",
|
|
"tensor_buft_overrides": "none",
|
|
"load_mode": "mmap",
|
|
"no_op_offload": 0,
|
|
"no_host": false,
|
|
"n_prompt": 512,
|
|
"n_gen": 0,
|
|
"avg_ts": 1000.0,
|
|
"stddev_ts": 5.0,
|
|
"samples_ts": [995.0, 1000.0, 1005.0]
|
|
},
|
|
{
|
|
"build_commit": "baseline",
|
|
"backends": "Metal",
|
|
"cpu_info": "Apple M4 Max",
|
|
"gpu_info": "Apple M4 Max",
|
|
"model_filename": "gemma-4-31b-q4.gguf",
|
|
"model_type": "31B",
|
|
"model_size": 20000000000,
|
|
"model_n_params": 30700000000,
|
|
"n_batch": 2048,
|
|
"n_ubatch": 512,
|
|
"n_threads": 12,
|
|
"type_k": "f16",
|
|
"type_v": "f16",
|
|
"n_gpu_layers": 999,
|
|
"n_cpu_moe": 0,
|
|
"split_mode": "layer",
|
|
"main_gpu": 0,
|
|
"no_kv_offload": false,
|
|
"flash_attn": -1,
|
|
"attention_kernel": "parallel",
|
|
"devices": "Metal0",
|
|
"tensor_split": "0.00",
|
|
"tensor_buft_overrides": "none",
|
|
"load_mode": "mmap",
|
|
"no_op_offload": 0,
|
|
"no_host": false,
|
|
"n_prompt": 0,
|
|
"n_gen": 128,
|
|
"avg_ts": 20.0,
|
|
"stddev_ts": 0.1,
|
|
"samples_ts": [19.9, 20.0, 20.1]
|
|
}
|
|
]
|