Repository navigation

madspace: float32 networks - #274

Merged

oliviermattelaer merged 30 commits into

mainfrom

feat-madspace-fp32

Oct 30, 2026

Merged

madspace: float32 networks#274
oliviermattelaer merged 30 commits into
mainfrom
feat-madspace-fp32

Commits on Oct 13, 2026
- madspace: rename dtypes to int32/float64 and add float32
  theoheimel
  committed
  7cc382f
- madspace: float32/float64 variants of NN instructions
  theoheimel
  committed
  1933c24
- madspace: float32 SIMD vectors with twice the lanes
  theoheimel
  committed
  d6f1fe5
- madspace: train float32 parameters
  theoheimel
  committed
  c33c5fc
- madspace: precision argument for MLP and the networks using it
  theoheimel
  committed
  818d0e5
- madspace: integer gradient views for gather and select
  theoheimel
  committed
  52b1683
- mg7: run card setting madnis.nn_float_precision
  theoheimel
  committed
  61dc6a4
- madspace: build OpenBLAS with single precision
  theoheimel
  committed
  f002872
- madspace: stub location of DataType default arguments
  theoheimel
  committed
  399349e
Commits on Oct 14, 2026
Commits on Oct 17, 2026
- Merge branch 'fix-madspace-simd' into feat-madspace-fp32
  theoheimel
  committed
  aae84f5
- madspace: int32 indices for empty batch in GPU unweight
  
  theoheimel
  and
  claude
  committed
  c534fdb
- madspace: single-kernel bias gradient in GPU matmul backward
  
  theoheimel
  and
  claude
  committed
  3125be0
Commits on Oct 27, 2026
- Merge branch 'fix-madspace-simd' into feat-madspace-fp32
  theoheimel
  committed
  acb9b88
- Merge branch 'feat-madspace-fp32' of github.com:MadGraphTeam/MadGraph7 into feat-madspace-fp32
  theoheimel
  committed
  20e8722
Commits on Oct 28, 2026
- madspace: fuse float32/float64 conversions around network layers
  
  theoheimel
  and
  claude
  committed
  38b4205
- Merge branch 'fix-madspace-simd' into feat-madspace-fp32
  theoheimel
  committed
  9738c30
- madspace: train the MLP test network with a smooth activation
  
  theoheimel
  and
  claude
  committed
  a9ceb11
Commits on Oct 29, 2026
- madspace: 64-bit offsets in the x86 SIMD gathers and scatters
  
  theoheimel
  and
  claude
  committed
  11be642
- madspace: add the matmul bias after the gemm
  
  theoheimel
  and
  claude
  committed
  88ab19d
- madspace: no gradients of integer inputs in structural backward ops
  
  theoheimel
  and
  claude
  committed
  7e899af
- madspace: fuse network activations into the matmul bias kernel
  
  theoheimel
  and
  claude
  committed
  e9e86c0
- madspace: rename fuse_conversions to fuse_instructions
  
  theoheimel
  and
  claude
  committed
  e5a0eea
- madspace: atomic accumulation in GPU tensor views, GPU-safe test_float32
  
  theoheimel
  and
  claude
  committed
  a170428
- Merge branch 'feat-madspace-fp32' of github.com:MadGraphTeam/MadGraph7 into feat-madspace-fp32
  theoheimel
  committed
  cf30447
- madspace tests: test_float32 uses the default context of the test device
  
  theoheimel
  and
  claude
  committed
  059fbc6
- Merge remote-tracking branch 'origin/main' into feat-madspace-fp32
  theoheimel
  committed
  81f7a39
Commits on Oct 30, 2026
- madspace: review fixes for float32 networks
  
  theoheimel
  and
  claude
  committed
  eb6338c

Commits

Commits on Oct 13, 2026

Commits on Oct 14, 2026

Commits on Oct 17, 2026

Commits on Oct 27, 2026

Commits on Oct 28, 2026

Commits on Oct 29, 2026

Commits on Oct 30, 2026