issues
search
ahrefs
/
ocannl
OCANNL: OCaml Compiles Algorithms for Neural Networks Learning
BSD 2-Clause "Simplified" License
62
stars
2
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Optimization: Don't generate the addition for a zero-out accumulation when redundant
#236
lukstafi
closed
6 months ago
0
Fix: Apply bindings before sending to worker
#235
lukstafi
closed
6 months ago
0
Consider programmatically enforcing that the order of first-time executions is the same as of jitting
#234
lukstafi
closed
2 weeks ago
2
Consider adding a "hosted on device" setup for uniform-memory devices
#233
lukstafi
opened
6 months ago
0
Gradient update contexts are missing arrays except for the loss
#232
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#231
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#230
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#229
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#228
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#227
lukstafi
closed
6 months ago
1
Gradient update contexts are missing arrays except for the loss
#226
lukstafi
closed
6 months ago
1
Implement and benchmark against vdW_grid_bench
#225
lukstafi
closed
5 months ago
1
Make it easy to avoid recompiling code when jitting for different contexts / devices
#224
lukstafi
closed
6 months ago
1
Does the OCaml string passed to `RValue.string_literal` need to be null-terminated?
#223
lukstafi
opened
6 months ago
0
Try out "Multilinear Operator Networks"
#222
lukstafi
opened
7 months ago
0
An incompleteness bug deep inside shape inference, with a workaround hiding it for now
#221
lukstafi
closed
8 months ago
0
Reproduce Mamba
#220
lukstafi
opened
9 months ago
1
Integrate with `polars-ocaml`
#219
lukstafi
opened
10 months ago
0
Factor out gccjit and cudajit as separate independently hosted packages, provide `conf-gccjit` and `conf-cuda`
#218
lukstafi
closed
9 months ago
1
Shape inference: fix updating an inequalities-and-equations entry (don't self-eliminate)
#217
lukstafi
closed
9 months ago
1
Incorporate examples from François Fleuret's lectures
#216
lukstafi
opened
11 months ago
1
Optimize shape inference
#215
lukstafi
opened
11 months ago
1
Regression in simplification of literals
#214
lukstafi
closed
6 months ago
2
Consider renaming the types `Shape.dims` and `Shape.row`, and the fields of `Shape.dims`, to be less confusing
#213
lukstafi
closed
9 months ago
1
Uniform naming convention for `Lazy_array.t`; refactor and possibly rename `Lazy_array.t`
#212
lukstafi
closed
6 months ago
1
Rename `Assignments.to_low_level` to `reference_compile`, and introduce `cpu_friendly_compile` (later also `cuda_friendly`)
#211
lukstafi
opened
11 months ago
1
Consider incorporating the (runtime) argument's label for tensors generated via `let%op f x = ... in ...`
#210
lukstafi
closed
2 weeks ago
2
Accept `:=` syntax in the `%cd` extension for the `Fetch` constructor
#209
lukstafi
opened
11 months ago
0
Deriving projections for einsum and permute specs is broken
#208
lukstafi
closed
11 months ago
1
Some baffling regression in `Tensor.print` due to `Stdlib.Format`
#207
lukstafi
closed
11 months ago
2
Enable (and ensure proper) memory management for arrays embedded in temporary or only-non-top-level tensors
#206
lukstafi
closed
2 months ago
3
Equational projection inference should refresh the projection ids for every `update_step`
#205
lukstafi
closed
11 months ago
2
Support bfloat16
#204
lukstafi
opened
12 months ago
1
Consider re-introducing within-shape scaling constraints
#203
lukstafi
opened
12 months ago
0
Incorporate axis labels into dim unification
#202
lukstafi
closed
12 months ago
1
Rewrite shape inference equationally
#201
lukstafi
closed
12 months ago
1
Add an LLVM / clang backend
#200
lukstafi
opened
1 year ago
1
Extend the `%cd` syntax with commenting (`Block_comment` with string interpolation)
#199
lukstafi
closed
2 months ago
1
Make name passing optional by: extracting names from the comments in `Assignments.t`, generating more comments
#198
lukstafi
closed
1 year ago
0
Add an operator to convert static indices to tensors
#197
lukstafi
closed
11 months ago
1
Work out a style guideline and audit all code to follow the guideline
#196
lukstafi
opened
1 year ago
0
Consider re-introducing cuda `__constant__` arrays, but beware
#195
lukstafi
opened
1 year ago
0
Audit backends for mixed-precision safety
#194
lukstafi
closed
4 months ago
0
Consider getting rid of `backprop_roots`, `diff.backprop`, and compiling backpropagation by explicit traversal
#193
lukstafi
closed
11 months ago
1
Investigate what happens when division is used with `Compose` composition type
#192
lukstafi
opened
1 year ago
0
Reduce the huge amount of code duplication in `ppx_nn_cd`
#191
lukstafi
closed
2 weeks ago
0
Introduce a `strict` mode that enforces naming explicit axes; "Named Tensors" lessons
#190
lukstafi
opened
1 year ago
0
Introduce static indexing, use dynamic indexing's operations and shape inference code
#189
lukstafi
closed
1 year ago
1
Add `~initial` / `~final` boolean flags to JITted functions to determine if copying from / to host is needed
#188
lukstafi
closed
1 year ago
1
Consider re-introducing `frozen` axes
#187
lukstafi
closed
11 months ago
0
Previous
Next