Hui Liu

Results 14 comments of Hui Liu

Check failed: [max_elem + n_columns*(max_elem + 1) < 0x7fffffff] Max_values is too large to be transformed.请问这个问题怎么解决？

0x7fffffff, that is the maximal integer for int32. int64 should be used for index.

Struggling to compile/run PyTorch in Docker (gfx803, no AVX)

it looks for libncurses-dev. you may modify build.ninja and add -lncurses to LINK_LIBRARIES.

PyPI wheel

Let us assume user has a ROCM installed already. the PyPI version can detect ROCm using ROCM_PATH and find the libraries it needs. So amdsmi or pyamdsmi can be independent...

[Misc]: Cohere models are not working due to transformers library outdated?

use 4.39.1?

[Core] Multiprocessing executor for single-node multi-GPU deployment

Any ETA when this PR will be merged? Thank you.

Can you choose which GPU to use. like tf inference device_map="cuda:0"

you may use CUDA_VISIBLE_DEVICES / HIP_VISIBLE_DEVICES to isolate GPU. PyTorch picks which GPUs to use.

Can you choose which GPU to use. like tf inference device_map="cuda:0"

torchrun should be able to solve your problem? each process runs 1 vllm...

Add MI300X Spec

Hello, the agreement has been signed before. Thank you.

[Feature]: Parity With NVIDIA DCGM - Out of the Box Diagnostics Levels

@functionstackx I have created a ticket, and its internal tracking ID is SWDEV-542037. Thank you.

[Feature]: Parity With NVIDIA DCGM - Pulse Test

@functionstackx internal ticket created, SWDEV-542282, thank you.