Releases: dfdx/Yota.jl
Releases · dfdx/Yota.jl
v0.8.5
v0.8.4
v0.8.2
v0.8.1
v0.8.0
Yota v0.8.0
Closed issues:
- Functions with Varargs are nor properly handled (#84)
- Function to break down non-primitives on a tape into primitives (#94)
- Generic broadcast (#102)
- Primitivize rrules (#103)
Core.Typeof
instead ofinstance_type
? (#109)op
not defined inrecord_primitive
assertion error (#110)Umlaut.getcode
making invalid call (#113)- ERROR: UndefVarError: to_same_device not defined (#115)
- MethodError: no method matching length(::Type{Val{2}}) when differentiating log-likelihood (#121)
Merged pull requests:
- More dynamic rrule via ad (#101) (@dfdx)
- Migrate from Ghost to Umlaut (#104) (@dfdx)
- Write code like it's 2022 (#105) (@dfdx)
- Docs (#106) (@dfdx)
- Even more generic broadcasted (#107) (@dfdx)
- fix isprimitive bug (#108) (@cscherrer)
- Fixes (#111) (@dfdx)
- Fix rule for
getindex(::Tuple)
(#116) (@mcabbott) - Fix to same device (#119) (@dfdx)
- Adapt for Umlaut 0.4 (#120) (@dfdx)
- Fix kw rrule (#122) (@dfdx)
- Revert single-pass tracing/rrule transformation (#124) (@dfdx)
- Fix rrule for iterate. Update TagBot config (#125) (@dfdx)
v0.6.2
v0.6.1
v0.6.0
Yota v0.6.0
Update the list and versions of essential dependencies
Closed issues:
- Can't use broadcasting on non-primitives (#3)
- More gradients (#14)
- Investigate poor performance (#28)
- Tape BoundsError using dictionaries (#34)
- Nested Differentiation (#35)
- Use
rrule
from ChainRules.jl (#65) - grad() fails for simple case since v0.5.0 (#90)
- Add docs for update_chain_rules!() (#91)
- Multi argument + not supported (#92)
- rrule piracy (#95)
Merged pull requests: