mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-30 15:43:19 +08:00
39 lines
1.2 KiB
JSON
39 lines
1.2 KiB
JSON
{
|
|
"version": "0.2.0",
|
|
"configurations": [
|
|
{
|
|
"name": "build_and_run_ad.py",
|
|
"type": "debugpy",
|
|
"request": "launch",
|
|
"program": "build_and_run_ad.py",
|
|
"args": [
|
|
"--model=meta-llama/Meta-Llama-3.1-8B-Instruct",
|
|
"--args.world_size=2",
|
|
"--args.runtime=demollm",
|
|
"--args.compile_backend=torch-simple",
|
|
"--args.attn_page_size=16",
|
|
"--args.attn_backend=flashinfer",
|
|
"--args.model_factory=AutoModelForCausalLM",
|
|
"--benchmark.enabled=false",
|
|
"--prompt.batch_size=2",
|
|
"--args.model_kwargs",
|
|
"num_hidden_layers=3,num_attention_heads=32",
|
|
],
|
|
"console": "integratedTerminal",
|
|
"justMyCode": false,
|
|
"cwd": "${workspaceFolder}/examples/auto_deploy"
|
|
},
|
|
{
|
|
"name": "Python: Debug Tests",
|
|
"type": "debugpy",
|
|
"request": "launch",
|
|
"program": "${file}",
|
|
"purpose": [
|
|
"debug-test",
|
|
],
|
|
"console": "integratedTerminal",
|
|
"justMyCode": false
|
|
},
|
|
]
|
|
}
|