Tags

Tags give the ability to mark specific points in history as being important

master-703ef9c

703ef9c1 · Set the singleton to nullptr here. · Sep 14, 2023
b1226

990a5e22 · cmake : add relocatable Llama package (#2960) · Sep 14, 2023
b1225

980ab41a · docker : add gpu image CI builds (#3103) · Sep 14, 2023
b1223

4c8643dd · feature : support Baichuan serial models (#3009) · Sep 14, 2023
b1222

35f73049 · speculative : add heuristic algorithm (#3006) · Sep 14, 2023
master-7ff671e

7ff671e1 · Only use vulkan with known quant that work. · Sep 14, 2023
master-8616ce0

8616ce08 · Sync from device back to host at begin of new prompt. · Sep 13, 2023
master-80da9b8

80da9b89 · Don't try and install kompute artifacts. · Sep 13, 2023
master-e5ab32a

e5ab32aa · vulkan: disambiguate gpus with the same name · Sep 13, 2023
master-2f7732b

2f7732b6 · Throw an exception when allocation fails for vulkan. · Sep 13, 2023
b1221

71ca2fad · whisper : tokenizer fix + re-enable tokenizer test for LLaMa (#3096) · Sep 13, 2023
b1220

1b6c650d · cmake : add a compiler flag check for FP16 format (#3086) · Sep 13, 2023
b1219

0a5eebb4 · CUDA: mul_mat_q RDNA2 tunings (#2910) · Sep 13, 2023
b1218

84e72365 · speculative: add --n-gpu-layers-draft option (#3063) · Sep 13, 2023
b1217

b52b29ab · arm64 support for windows (#3007) · Sep 12, 2023
b1216

4f7cd6ba · CUDA: fix LoRAs (#3130) · Sep 13, 2023
master-9bee309

9bee309a · Make kompute actually include external SDK headers when requested · Sep 12, 2023
master-0412ec2

0412ec28 · Completely revamp how we do object management with the vulkan backend and · Sep 12, 2023
b1215

89e89599 · CUDA: fix mul_mat_q not used for output tensor (#3127) · Sep 11, 2023
b1214

d54a4027 · CUDA: lower GPU latency + fix Windows performance (#3110) · Sep 11, 2023

Prev
1
…
56
57
58
59
60
61
62
63
64
…
99
Next