Merge pull request #451 from ZacharyZcR/feat/cuda-grouped-g64

cuda: grouped-int4 (fmt=4) support in the expert-group kernels — opens the GPU tier to g64 and E8 containers (#334)
2026-07-20 18:06:45 +02:00
parent 78a773e539 a74e3e0c3a
commit e78fcfcbca
5 changed files with 206 additions and 8 deletions
@@ -36,6 +36,9 @@ COLI_CUDA_DLLEXPORT void coli_cuda_group_stats(uint64_t *calls, uint64_t *expert
                           double *h2d_ms, double *kernel_ms, double *d2h_ms);

 /* Upload without executing, so capacity failures happen during model startup. */
+COLI_CUDA_DLLEXPORT int coli_cuda_tensor_upload_g(ColiCudaTensor **tensor,
+        const void *weights, const float *scales,
+        int fmt, int I, int O, int device, int gs);
 COLI_CUDA_DLLEXPORT int coli_cuda_tensor_upload(ColiCudaTensor **tensor,
                            const void *weights, const float *scales,
                            int fmt, int I, int O, int device);