[Diff since v1.49.0](https://github.com/SciML/DiffEqFlux.jl/compare/v1.49.0...v1.49.1)
**Closed issues:**
- NeuralDSDE Adjoint Fails on GPU (#138)
- Which types work as container of parameters? (#178)
- NeuralODE: modeling dynamics of system with (discontinuous) control inputs (#190)
- Nested reverse modes with fast layers (#207)
- BFGS Errors with GPU (#238)
- BoundsError: attempt to access () at index [0] (#253)
- Code in LV-Jump.md blog post no longer runs (#272)
- Reproducible TypeError in sciml_train (#278)
- Why the need for Fast* layers? (#328)
- Rename DiffEqFlux? (#354)
- The MNIST Neural ODE Classifier tutorial is broken (#368)
- sciml_train crashes when fed Fluxs ::Params type (#379)
- Neural Graph Differential Equation tests disabled (#405)
- Link to stable documentation broken (#407)
- neural_ode_sciml example fails when Dense layer replaced by GRU (#432)
- Indexing issue caused by applychain (#436)
- Mutating Arrays not Allowed (#443)
- Optimal Control ref in exogenous input page is broken (#445)
- Support for sparse matrix ODEs (#462)
- implicit layers with neuralODEs ? (#478)
- Disable events tests (#483)
- PresetTimeCallback errors with Zygote differentiation (#502)
- Suggestion for Memory Limitations (#527)
- Unstable `FastDense` `hash` and comparison. (#555)
- SteadyStateAdjoint + GPUs: scalar getindex is disallowed (#557)
- sample code running slow on Atom (#560)
- ReverseDiff tape generation over newer Zygote versions segfaults (#586)
- ERROR: UndefVarError: eJ not defined (#588)
- Neural Graph Tutorial not working (#609)
- Note on hybrid sensitivity analysis methods needs an update? (#612)
- ERROR: LoadError: MethodError: no method matching Float32(::ReverseDiff.TrackedReal{Float64, Float32, Nothing}) (#615)
- type HMCKernel has no field integrator (#618)
- Reverse-mode VJPs with mixed scalar+vector GPU code (#632)
- Error in Neural Graph Ordinary Differential Equations example (#640)
- Best practice for taking backpropagation with logging tools for DiffEq.jl (#661)
- Add batch support to FastDense layers, mimicking Flux (#672)
- Missing optimizations in preallocation form (#680)
- Error in the UODEs training when the network enlarges (#689)
- initial_params() always identical (#690)
- PINN with TensorLayer example fails - maybe Enzyme internal error (#694)
- MethodError: no method matching Flux.Dense(::Pair{Int64, Int64}, ::typeof(NNlib.elu)) (#700)
- `sciml_train` output API documentation (#702)
- LoadError with sciml_train (#713)
- Training a NODE using multiple shooting. (#717)
**Merged pull requests:**
- Corrections in keyword arguments for NeuralODE example (#720) (@Abhishek-1Bhatt)
- Removal of deprecations (#722) (@Abhishek-1Bhatt)
- Fix Some Doctests (#723) (@DhairyaLGandhi)
- Fixed NeuralDSDE (#724) (@Abhishek-1Bhatt)
- Update neural_de tests (#725) (@frankschae)