transformers
3491f116 - Idefics3 without cache fix (#44607)

Commit
1 day ago
Idefics3 without cache fix (#44607) * fix: Remove cache_position kwargs everywhere in idefics3 Branch: Idefics3WithoutCacheFix AI-usage: none Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Only purge pixel model inputs for subsequent turns if using cache This was the root of the incorrect results without cache! Branch: Idefics3WithoutCacheFix AI-usage: none Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Use named argument for use_cache Branch: Idefics3WithoutCacheFix AI-usage: none Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Remove cache_position and fix pixel purge logic for idefics2 Branch: Idefics3WithoutCacheFix AI-usage: none Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Copy fixes over to smolvlm Branch: Idefics3WithoutCacheFix AI-usage: none Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * Apply repo consistency fixes * repo bot didn't do the job, so fix-repo --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Raushan Turganbay <raushan@huggingface.co> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Author
Parents
Loading