Skip to content

Hy3 (295B): CUDA backend support, GQA attention path #535

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
giannisanni wants to merge 7 commits into
base: glm5.2
Choose a base branch
from
Open

Hy3 (295B): CUDA backend support, GQA attention path #535

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
0ed3997
GLM 5.2: CUDA backend support
Jul 7, 2026
523ebbf
Hy3 (hy_v3) port foundation: model recognition + GQA attention kernels
Jul 8, 2026
685955c
Hy3 inference path: forward_token, generate_hy3_cuda_argmax, dispatch…
Jul 8, 2026
059019c
Hy3: bind exp_probs_b without .bias suffix (matches real hy-v3 GGUF)
Jul 8, 2026
04ceb1a
Hy3: resolve bos/eos by metadata token ids, not DeepSeek literals
Jul 8, 2026
159873a
GQA attention: single-pass online softmax, kill the pass-2 score reco…
Jul 9, 2026
526ccee
Hy3: interactive chat mode with KV retained across turns (GLM parity)
Jul 9, 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
4 changes: 3 additions & 1 deletion Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,7 +215,9 @@ ds4_agent_cpu.o: ds4_agent.c ds4.h ds4_ssd.h ds4_distributed.h ds4_help.h ds4_kv
ds4_metal.o: ds4_metal.m ds4_gpu.h $(METAL_SRCS)
$(CC) $(OBJCFLAGS) -c -o $@ ds4_metal.m

ds4_cuda.o: ds4_cuda.cu ds4_gpu.h ds4_iq2_tables_cuda.inc
ds4_cuda.o: ds4_cuda.cu ds4_gpu.h ds4_iq2_tables_cuda.inc ds4_cuda_gqa.inc \
ds4_cuda_glm_kv.inc ds4_cuda_glm_indexer.inc ds4_cuda_glm_attn.inc \
ds4_cuda_glm_moe.inc ds4_cuda_glm_stubs.inc
$(NVCC) $(NVCCFLAGS) -c -o $@ ds4_cuda.cu

ds4_rocm.o: ds4_rocm.cu ds4_gpu.h ds4_iq2_tables_cuda.inc $(ROCM_SRCS)
Expand Down
870 changes: 851 additions & 19 deletions ds4.c
View file
Open in desktop

Large diffs are not rendered by default.

13 changes: 13 additions & 0 deletions ds4_cli.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1937,6 +1937,19 @@ static cli_config parse_options(int argc, char **argv) {
}

int main(int argc, char **argv) {
/* Kernel self-tests that need no model file; handled before option
* parsing so no other flags are required. */
for (int i = 1; i < argc; i++) {
if (!strcmp(argv[i], "--gqa-selftest")) {
#ifndef DS4_NO_GPU
extern int ds4_gpu_gqa_selftest(void);
return ds4_gpu_gqa_selftest() ? 0 : 1;
#else
fprintf(stderr, "ds4: --gqa-selftest requires a GPU build\n");
return 2;
#endif
}
}
cli_config cfg = parse_options(argc, argv);
if (cfg.gen.dump_tokens) {
if (cfg.gen.prompt == NULL) {
Expand Down
1,797 changes: 1,627 additions & 170 deletions ds4_cuda.cu
View file
Open in desktop

Large diffs are not rendered by default.

Loading