Skip to content

Metal: FP8-packed compressed-KV cache + long-context memory optimizations #418

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
aledesogusbusiness-hue wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Metal: FP8-packed compressed-KV cache + long-context memory optimizations #418

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
343 changes: 225 additions & 118 deletions ds4.c
View file
Open in desktop

Large diffs are not rendered by default.

42 changes: 42 additions & 0 deletions ds4_gpu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -179,6 +179,40 @@ int ds4_gpu_indexer_scores_prefill_tensor(
uint32_t ratio,
float scale);

/* comp_mask f16 (1) unless KV opts disabled via DS4_DISABLE_KV_OPTS (0 = f32). */
int ds4_gpu_comp_mask_f16(void);

int ds4_gpu_indexer_prefill_score_topk_tiled(
ds4_gpu_tensor *scores,
ds4_gpu_tensor *selected,
const ds4_gpu_tensor *q,
const ds4_gpu_tensor *weights,
const ds4_gpu_tensor *index_comp,
uint32_t n_comp,
uint32_t n_tokens,
uint32_t n_head,
uint32_t head_dim,
uint32_t ratio,
float scale,
uint32_t top_k,
uint32_t score_tile);

int ds4_gpu_indexer_decode_batch_score_topk_tiled(
ds4_gpu_tensor *scores,
ds4_gpu_tensor *selected,
const ds4_gpu_tensor *q,
const ds4_gpu_tensor *weights,
const ds4_gpu_tensor *index_comp,
uint32_t n_comp,
uint32_t n_tokens,
uint32_t pos0,
uint32_t n_head,
uint32_t head_dim,
uint32_t ratio,
float scale,
uint32_t top_k,
uint32_t score_tile);

int ds4_gpu_indexer_scores_decode_batch_tensor(
ds4_gpu_tensor *scores,
const ds4_gpu_tensor *q,
Expand Down Expand Up @@ -413,6 +447,14 @@ int ds4_gpu_dsv4_fp8_kv_quantize_tensor(
uint32_t head_dim,
uint32_t n_rot);

int ds4_gpu_dsv4_kv_pack_comp_rows(
ds4_gpu_tensor *dst,
uint64_t dst_byte_offset,
ds4_gpu_tensor *src,
uint32_t n_rows,
uint32_t head_dim,
uint32_t n_rot);

int ds4_gpu_dsv4_indexer_qat_tensor(
ds4_gpu_tensor *x,
uint32_t n_rows,
Expand Down
Loading