Skip to content

feat(loader): support stock-recipe (Q8_0/F32) abliterated GGUFs end-to-end on Metal #60

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

Closed
audreyt wants to merge 8 commits into from
Closed

feat(loader): support stock-recipe (Q8_0/F32) abliterated GGUFs end-to-end on Metal #60

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
24022c2
feat(loader): support stock-recipe (Q8_0/F32) GGUFs end-to-end on Metal
audreyt May 10, 2026
431943f
fix(metal): correct Q8_0 ape compressor paths
audreyt May 10, 2026
c2144e5
fix(loader): relax indexer decode F16-only validators to accept Q8_0
audreyt May 10, 2026
86aa248
Merge remote-tracking branch 'origin/main' into support-q8_0-token-embd
audreyt May 10, 2026
d624188
chore(metal): drop dead Q8_0->F32 cpy encoder
audreyt May 10, 2026
a0f1a0b
Merge remote-tracking branch 'origin/main' into support-q8_0-token-embd
audreyt May 10, 2026
147b0d7
Merge remote-tracking branch 'origin/main' into support-q8_0-token-embd
audreyt May 11, 2026
e8f5bdc
Merge remote-tracking branch 'origin/main' into support-q8_0-token-embd
audreyt May 12, 2026
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
421 changes: 213 additions & 208 deletions ds4.c
View file
Open in desktop

Large diffs are not rendered by default.

7 changes: 5 additions & 2 deletions ds4_gpu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,7 @@ void ds4_gpu_print_memory_report(const char *label);
* compressed-attention indexer that chooses visible compressed rows.
*/

/* `weight_type` is the GGUF tensor type code (1 for F16, 8 for Q8_0). */
int ds4_gpu_embed_token_hc_tensor(
ds4_gpu_tensor *out_hc,
const void *model_map,
Expand All @@ -58,7 +59,8 @@ int ds4_gpu_embed_token_hc_tensor(
uint32_t n_vocab,
uint32_t token,
uint32_t n_embd,
uint32_t n_hc);
uint32_t n_hc,
uint32_t weight_type);

int ds4_gpu_embed_tokens_hc_tensor(
ds4_gpu_tensor *out_hc,
Expand All @@ -69,7 +71,8 @@ int ds4_gpu_embed_tokens_hc_tensor(
uint32_t n_vocab,
uint32_t n_tokens,
uint32_t n_embd,
uint32_t n_hc);
uint32_t n_hc,
uint32_t weight_type);

int ds4_gpu_indexer_score_one_tensor(
ds4_gpu_tensor *scores,
Expand Down
Loading