diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-03-01 08:25:27 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-03-01 08:25:27 +0200 |
| commit | a79ab8f34222e1e0142a30eaa97e78ad077abca9 (patch) | |
| tree | 24f89079780736d697347e1ebbe6544750534e22 /common/common.h | |
| parent | b762db7c9264199c2d0f66e7d63e3b4884f3fc0c (diff) | |
Reduce size of compute buffers (#237)
* This reduces compute buffer size for MLA
* This should accomplish it for standard attention
* Much better
* Better concat for contiguous tensors
If all the op does is to concatenate the second tensor
to the first, why would we want to have a loop?
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'common/common.h')
| -rw-r--r-- | common/common.h | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/common/common.h b/common/common.h index ef5175f3..f35f3558 100644 --- a/common/common.h +++ b/common/common.h @@ -175,7 +175,8 @@ struct gpt_params { bool simple_io = false; // improves compatibility with subprocesses and limited consoles bool cont_batching = true; // insert new sequences for decoding on-the-fly bool flash_attn = false; // flash attention - int mla_attn = false; // MLA 0: standard attention, 1: MLA with K and transposed V cache, 2: MLA with just K cache + int mla_attn = 0; // MLA 0: standard attention, 1: MLA with K and transposed V cache, 2: MLA with just K cache + int attn_max_batch = 0; // Max batch size to use when computing attention (only applicable if flash_attn = false) bool fused_moe_up_gate = false; // fused up*unary(gate) op for MoE models bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix |