Commit b00533e
Use ninja to speed up build (#5088)
Deepspeed have too many ops now, and it take too many time to pre-build
all ops.
I notice deepspeed disabled `ninja` 4 years ago
(#298) and I think we should
consider enable it now.
The issue mentioned in #298
can be solved by resolving `include_dirs` to absolute path.
---------
Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com>
Co-authored-by: Logan Adams <loadams@microsoft.com>
Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com>
Co-authored-by: Michael Wyatt <michaelwyatt@microsoft.com>1 parent 005afe1 commit b00533e
File tree
17 files changed
+34
-21
lines changed- deepspeed/inference/v2/kernels
- core_ops
- bias_activations
- cuda_layer_norm
- cuda_rms_norm
- gated_activations
- ragged_ops
- embed
- linear_blocked_kv_rotary
- logits_gather
- moe_gather
- moe_scatter
- top_k_gating
- op_builder
- cpu
- hpu
- xpu
17 files changed
+34
-21
lines changed
0 commit comments