mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
de409e8468
* [AutoDeploy] HF factory improvements Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com> * improve monkey-patches and add unit tests Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com> --------- Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
33 lines
1.1 KiB
JSON
33 lines
1.1 KiB
JSON
{
|
|
"version": "0.2.0",
|
|
"configurations": [
|
|
{
|
|
"name": "build_and_run_ad.py",
|
|
"type": "debugpy",
|
|
"request": "launch",
|
|
"program": "build_and_run_ad.py",
|
|
"args": [
|
|
"--config",
|
|
"{\"batch_size\": 2, \"page_size\": 16, \"world_size\": 2, \"compile_backend\": \"torch-simple\", \"attn_backend\": \"FlashInfer\",\"model_factory\": \"AutoModelForCausalLM\", \"model\": \"meta-llama/Meta-Llama-3.1-8B-Instruct\", \"benchmark\": false}",
|
|
"--model-kwargs",
|
|
"{}",
|
|
// "{\"num_hidden_layers\": 3}",
|
|
],
|
|
"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
|
|
},
|
|
]
|
|
}
|