ROCm AWQ support #1514

Narsil merged 27 commits into main from rocm-awq-support
IlyasMoutawwakil
IlyasMoutawwakil fix exllama overflows
6cb6020e
IlyasMoutawwakil awq fallback to exllama
12c1f545
IlyasMoutawwakil post process exllama model
5766c55b
IlyasMoutawwakil
IlyasMoutawwakil add triton fallback to awq
8acbcb31
IlyasMoutawwakil fix missing g_idx and eventual overflow in triton kernel
0b5b8587
IlyasMoutawwakil
Narsil
Narsil commented on 2024-02-01
Narsil
IlyasMoutawwakil
Narsil
Narsil
IlyasMoutawwakil
IlyasMoutawwakil revert changes
8665ab07
IlyasMoutawwakil adapt awq weights to exllama/gptq kernels
fb59c562
IlyasMoutawwakil typing
bcdb02e4
IlyasMoutawwakil pass g_idx instead of changing triton kernel
994ed8e1
IlyasMoutawwakil none g_idx
af2c589c
IlyasMoutawwakil log message
cda5751b
IlyasMoutawwakil fix exllama overflows
461dd6f1
IlyasMoutawwakil awq fallback to exllama
75086526
IlyasMoutawwakil post process exllama model
aa2014fc
IlyasMoutawwakil add triton fallback to awq
3963074c
IlyasMoutawwakil fix missing g_idx and eventual overflow in triton kernel
3ceeb858
IlyasMoutawwakil revert changes
212fdfff
IlyasMoutawwakil adapt awq weights to exllama/gptq kernels
8074c404
IlyasMoutawwakil typing
646ab282
IlyasMoutawwakil pass g_idx instead of changing triton kernel
bbe5bede
IlyasMoutawwakil none g_idx
76834c99
IlyasMoutawwakil log message
2629193e
Narsil Narsil force pushed from cda5751b to 2629193e 1 year ago
Narsil
Narsil dismissed these changes on 2024-02-08
Narsil Updating the tests.
04d38a83
Narsil Narsil dismissed their stale review via 04d38a83 1 year ago
IlyasMoutawwakil Merge branch 'rocm-awq-support' of https://github.com/huggingface/tex…
e29fb799
IlyasMoutawwakil generate g_idx only for triton kernel
bc157af9
Narsil Update llama gptq.
a76821e0
IlyasMoutawwakil
Narsil Better error message on non rocm.
326f8e30
Narsil Narsil merged a4e58016 into main 1 year ago
Narsil Narsil deleted the rocm-awq-support branch 1 year ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone