Jeffrey Morgan
|
da0e345200
ml: use input context for extracting outputs (#9875)
|
пре 1 месец |
Jesse Gross
|
282bfaaa95
ollamarunner: Use a separate context per multimodal input
|
пре 1 месец |
Jesse Gross
|
9679f40146
ml: Allow models to constrain inputs to a single batch
|
пре 1 месец |
Michael Yang
|
5e2e0b46b1
fix: error if image requested without vision model
|
пре 1 месец |
Michael Yang
|
63a394068c
use 2d pooling
|
пре 1 месец |
jmorganca
|
11bfa62796
add trailing \n\n after <end_of_image> to match reference implementation
|
пре 1 месец |
jmorganca
|
f63e62e546
reduce kernel size, add TODO for loading from config
|
пре 1 месец |
jmorganca
|
65b0f329d1
Revert "Allow models to force a new batch"
|
пре 1 месец |
Jesse Gross
|
06007c0a18
Allow models to force a new batch
|
пре 1 месец |
Jesse Gross
|
2c40c4d35e
Fix follow up images and images split across batches
|
пре 1 месец |
Michael Yang
|
6b32a2d549
compat with upstream gguf
|
пре 1 месец |
Michael Yang
|
46bb0169c4
update model
|
пре 1 месец |
Michael Yang
|
8934324b72
use fast attention
|
пре 1 месец |
Michael Yang
|
0df1800436
set non-causal attention
|
пре 1 месец |
Michael Yang
|
4b037a97dc
add gemma vision encoder
|
пре 1 месец |
Patrick Devine
|
5f74d1fd47
gemma2 impl
|
пре 2 месеци |