Commits · a8e83a7654fffa169b90fa927e6d19c4c0c765d7 · OpenDAS / ollama

11 Mar, 2025 14 commits
- Disable causal attention based on batch index · a8e83a76
  Jesse Gross authored Mar 10, 2025
```
Currently we are using positions, which are relative to a
sequence and may not be unique.
```
  a8e83a76
- Fix follow up images and images split across batches · 2c40c4d3
  Jesse Gross authored Mar 09, 2025
  
  2c40c4d3
- use non-causal mask only for image positions · e9527893
  Michael Yang authored Mar 10, 2025
  
  e9527893
- use non-causal mask for inputs with images · 9d2a20a7
  Michael Yang authored Mar 10, 2025
  
  9d2a20a7
- compat with upstream gguf · 6b32a2d5
  Michael Yang authored Mar 10, 2025
  
  6b32a2d5
- fix vision encoder · f8889128
  Michael Yang authored Mar 09, 2025
  
  f8889128
- fix configs · 9b54267e
  Patrick Devine authored Mar 08, 2025
  
  9b54267e
- update model · 46bb0169
  Michael Yang authored Mar 08, 2025
  
  46bb0169
- use fast attention · 8934324b
  Michael Yang authored Mar 07, 2025
  
  8934324b
- fix conversion · c62861f4
  Patrick Devine authored Mar 07, 2025
  
  c62861f4
- set non-causal attention · 0df18004
  Michael Yang authored Mar 07, 2025
  
  0df18004
- fix drift from main · 4346c240
  Jesse Gross authored Mar 07, 2025
  
  4346c240
- add gemma vision encoder · 4b037a97
  Michael Yang authored Mar 06, 2025
  
  4b037a97
- gemma2 impl · 5f74d1fd
  Patrick Devine authored Feb 07, 2025
  
  5f74d1fd