Smoothquant LLaMA builds not working on 0.8.0 release

[ghost]

System Info

GPU : NVIDIA A100 80GB

package version
tensorrt-9.2.0.post12.dev5-cp310-none-linux_x86_64.whl
[TensorRT-LLM] TensorRT-LLM version: 0.8.00.8.0

Who can help?

@Tracin @byshiue

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

1. Installation
   python -m pip install tensorrt_llm==0.8.0 --extra-index-url https://pypi.nvidia.com

2. Create smoothquant checkpoint for LLaMA
   python ./examples/llama/convert_checkpoint.py --model_dir ~/Llama-2-13b-chat-hf --output_dir ~/fp16-tp4-sq5 --dtype float16 --tp_size 4 --smoothquant 0.5 --per_token --per_channel --workers 4

Expected behavior

Checkpoint should be created.

actual behavior

Error at line - https://github.com/NVIDIA/TensorRT-LLM/blob/v0.8.0/examples/llama/convert_checkpoint.py#L1502

ValueError: You are trying to save a non contiguous tensor: transformer.layers.0.attention.qkv.weight which is not allowed. It either means you are trying to save tensors which are reference of each other in which case it's recommended to save only the full tensors, and reslice at load time, or simply call .contiguous() on your tensor to pack it before saving.

additional notes

No such error seen on release 0.7.1

My guess is that the function get_tllm_linear_sq_weight returns some non-contiguous tensors.

[byshiue]

Thank you for the report. We will fix it in next update. As you mentioned, you could add .contiguous() for

results[prefix + 'weight'] = torch.from_numpy(cur_weights).t().clone()

in get_tllm_linear_sq_weight.

[Hukongtao]

Qwen has the same error, When I run

python3 examples/qwen/convert_checkpoint.py \
    --model_dir ./Qwen-72B-Chat-Int4/ \
    --output_dir./Qwen-72B-Chat-Int4-TRT/tllm_checkpoint_1gpu_gptq/ \
    --dtype float16 \
    --use_weight_only \
    --weight_only_precision int4_gptq \
    --tp_size 4 \
    --pp_size 1 \
    --per_group

I got:
ValueError: You are trying to save a non contiguous tensor: transformer.layers.0.mlp.gate.weights_scaling_factor which is not allowed. It either means you are trying to save tensors which are reference of each other in which case it's recommended to save only the full tensors, and reslice at load time, or simply call .contiguous() on your tensor to pack it before saving.