Releases: PumasAI/SimpleChains.jl
Releases · PumasAI/SimpleChains.jl
v0.4.1
SimpleChains v0.4.1
Merged pull requests:
- reduce MAX_NUM_LV_EXTRACT (#117) (@chriselrod)
- backport MAX_NUM_LV_EXTRACT = 32 to SimpleChains 0.3 (#118) (@chriselrod)
v0.3.5
SimpleChains v0.3.5
Merged pull requests:
- fix arg order of
init_params!
(#116) (@chriselrod) - reduce MAX_NUM_LV_EXTRACT (#117) (@chriselrod)
- backport MAX_NUM_LV_EXTRACT = 32 to SimpleChains 0.3 (#118) (@chriselrod)
v0.4.0
v0.3.4
SimpleChains v0.3.4
Merged pull requests:
init_params!
should always have a default kwarg, rename tuple version_init_params!
(#115) (@chriselrod)
v0.3.3
SimpleChains v0.3.3
Closed issues:
init_params
bug (#112)
Merged pull requests:
- allow passing
AbstractRNG
s toinit_params
(#113) (@chriselrod)
v0.3.2
SimpleChains v0.3.2
Merged pull requests:
- Add docs for custom loss layer (#107) (@JamieMair)
- scale penalties by number of batches (#109) (@chriselrod)
- support more RNGs with glorot (#110) (@chriselrod)
- move
train_batched!
docstring to correct function (#111) (@chriselrod)
v0.3.1
SimpleChains v0.3.1
Closed issues:
GC.@preserve
is not reliable enough to actually rely on (#104)
Merged pull requests:
- promote types for calculating num_bytes (#105) (@chriselrod)
v0.3.0
SimpleChains v0.3.0
Closed issues:
- Issue with introductory example (#99)
Merged pull requests:
- More robust dual eval (#90) (@chriselrod)
- Don't promote vectors to matrices in dense layers (#92) (@chriselrod)
- StaticArrays in, StaticArrays out (#93) (@chriselrod)
- Stack Memory (#97) (@chriselrod)
- (fix): remove depreciated function call (#98) (@rahulunair)
- (fix): fixes example on readme (#100) (@rahulunair)
- Use 4 core laptop for smallmlp and update printed losses (#103) (@chriselrod)
v0.2.12
SimpleChains v0.2.12
Merged pull requests:
- TurboDense constructor static (#88) (@chriselrod)
- Static 0.7 (#91) (@chriselrod)
- 10x faster task_local_memory (#94) (@chriselrod)
- Update mnist.md (#95) (@pitmonticone)
v0.2.11
SimpleChains v0.2.11
Merged pull requests:
- thread safety through using task-local storage on non-inplace (#86) (@chriselrod)