whisper.cpp / ggml /src /ggml-vulkan /ggml-vulkan.cpp

Commit History

vulkan: implement initial support for IQ2 and IQ3 quantizations (llama/11360)
bd93c1b

Rémy Oudompheng jeffbolznv commited on

vulkan: Catch pipeline creation failure and print an error message (llama/11436)
d4f6b2c

jeffbolznv commited on

vulkan: compile shaders on-demand (llama/11406)
5c008f7

jeffbolznv commited on

Vulkan-run-test: fix mmq_wg_denoms (llama/11343)
133a580

amd-dwang commited on

vulkan: fix diag_mask_inf (llama/11323)
f76204e

jeffbolznv commited on

vulkan: fix coopmat2 validation failures (llama/11284)
f2cc7e9

jeffbolznv commited on

vulkan: fix coopmat2 flash attention for non-contiguous inputs (llama/11281)
e0e73fa

jeffbolznv commited on

vulkan: support copy from f32 to q4_0/q4_1/q5_0/q5_1/q8_0/iq4_nl (llama/11166)
3bb9e77

jeffbolznv commited on

Vulkan: Fix float16 use on devices without float16 support + fix subgroup_size_control validation error (llama/11161)
5ad3f1d

OccamRazor commited on

Disable GL_KHR_cooperative_matrix Vulkan extension if not available. (llama/11117)
623b74d

mbaudier commited on

Vulkan: Add device-specific blacklist for coopmat for the AMD proprietary driver (llama/11074)
4d90c3d

OccamRazor commited on

vulkan: optimize mul_mat for small values of N (llama/10991)
5fc8eea

jeffbolznv commited on

vulkan: im2col and matmul optimizations for stable diffusion (llama/10942)
beef268

jeffbolznv commited on

vulkan: Use push constant offset to handle misaligned descriptors (llama/10987)
04e729a

jeffbolznv commited on

vulkan: multi-row k quants (llama/10846)
3bf5be1

Eve commited on

vulkan: build fixes for 32b (llama/10927)
f1e76ce

jeffbolznv commited on

vulkan: bugfixes for small subgroup size systems + llvmpipe test (llama/10809)
9220b51

Eve commited on

rwkv6: add wkv6 support for Vulkan backend (llama/10829)
c7285d6

Zhiyuan Li mollysama commited on

llama : add Qwen2VL support + multimodal RoPE (llama/10361)
219d12b

RzZ ggerganov commited on

vulkan: small mul_mat_vec optimizations (llama/10665)
ec98109

Eve commited on

Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgroups for coopmats (llama/10721)
488f19e

OccamRazor commited on

vulkan: request round-to-even for fp16 in im2col/rope_head (llama/10767)
461484c

jeffbolznv commited on

vulkan: dynamic subgroup size for the remaining k quants (llama/10745)
1bbdb81

Eve commited on

vulkan: disable spirv-opt for coopmat shaders (llama/10763)
2ac53b2

jeffbolznv commited on

vulkan: fix compile warnings (llama/10731)
cdcb67c

jeffbolznv commited on

vulkan: compile a test shader in cmake to check for coopmat2 support (llama/10713)
980eeb3

jeffbolznv commited on

Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597)
9a4de04

OccamRazor commited on

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)
d10b47b

jeffbolznv commited on

vulkan: Implement "fast divide" (mul+shift) for unary ops like copy (llama/10642)
e9ee893

jeffbolznv commited on

vulkan: optimize and reenable split_k (llama/10637)
bca95f5

jeffbolznv commited on

vulkan: Dynamic subgroup size support for Q6_K mat_vec (llama/10536)
59600b5

Eve commited on

vulkan: get the first command buffer submitted sooner (llama/10499)
e1c1e73

jeffbolznv commited on

vulkan: Handle GPUs with less shared memory (llama/10468)
18a0ad1

jeffbolznv commited on

vulkan: fix group_norm (llama/10496)
8f5eeb8

jeffbolznv commited on

ggml : add support for dynamic loading of backends (llama/10469)
b73266f

Diego Devesa ggerganov commited on

vulkan: further optimize mul_mat_vec using larger loads (llama/10387)
50a2978

jeffbolznv commited on

vulkan: Optimize soft_max (llama/10301)
5cb851d

jeffbolznv commited on

Vulkan: Fix device info output format specifiers (llama/10366)
8000df9

OccamRazor commited on

vulkan: Optimize some mat-vec mul quant shaders (llama/10296)
dc0e685

jeffbolznv commited on

sync : leftovers (ggml/0)
0f6c498

ggerganov commited on

ggml : build backends as libraries (llama/10256)
3dc93f3

Diego Devesa ggerganov R0CKSTAR commited on