accelerate
c7e59dd7 - Deepspeed Ulysses/ALST integration (#3817)

Commit
48 days ago
Deepspeed Ulysses/ALST integration (#3817) * Feat: initial impl Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * improve Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * s/flavour/backend/ Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * style + ver Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * better check Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * check Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * docs + example Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * add tests Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * add tests Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * cleanup Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * cleanup Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestions from code review Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * add experimental notice Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * style Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * new deepspeed version Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * additional checks + tests Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * more docs Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * more docs Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * working now Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * style Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * update docs Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * more robust config parsing Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * fix Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestions from code review Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * check backend, integrate ulysses API improvement Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * style Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * fix default to match the doc Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestions from code review Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * fix Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * deepspeed=0.18.2 is out Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestions from code review Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com> * s/cp/sp Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * fixes Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestions from code review Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * Update src/accelerate/parallelism_config.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * suggestion Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Update docs/source/concept_guides/sequence_parallelism.md Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * Update sequence_parallelism.md * fix Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * fix Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * fix Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif * Apply suggestion from @kashif --------- Signed-off-by: Stas Bekman <stas.bekman@snowflake.com> Co-authored-by: S1ro1 <matej.sirovatka@gmail.com> Co-authored-by: Stas Bekman <stas.bekman@snowflake.com> Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>
Author
Parents
Loading