Hugging Face's logo

Csed-dev
/
matrixpfn-base

PyTorch
matrixpfn
preconditioner
sparse-linear-systems
graph-neural-network
Model card Files
xet
Community
Csed-dev commited on
Commit
409656d
·
verified ·
1 Parent(s): 16bf72c

v20260309-0350

Browse files
Files changed (2) hide show
  1. README.md +19 -20
  2. model.pt +2 -2
README.md CHANGED
@@ -9,11 +9,9 @@ tags:
9
 
10
  # matrixpfn-base
11
 
12
- (Currently in development. Not meant for use.)
13
-
14
  GNN-based learned preconditioner for sparse linear systems.
15
 
16
- **Version**: 0.1.0
17
 
18
  ## Usage
19
 
@@ -36,32 +34,33 @@ print(f"Converged: {result.converged} in {result.iterations} iterations")
36
  | Parameter | Value |
37
  |-----------|-------|
38
  | Network | ContextResGCN |
39
- | Layers | 12 |
40
- | Embed | 64 |
41
- | Hidden | 256 |
42
  | Context pairs | 10 |
43
- | Parameters | 419,074 |
44
  | dtype | float32 |
45
 
46
  ## Training
47
 
48
- - **epochs**: 2000
49
- - **best_loss**: 0.071401
50
  - **loss_function**: l1_direct
51
- - **batch_size**: 512
52
- - **domains**: diffusion, diffusion_advection
 
53
  - **grid_sizes**: [16, 24, 32, 48]
54
 
55
  ## Benchmark
56
 
57
  | domain | grid | converged | avg_iters | avg_residual |
58
  |--------|------|-----------|-----------|--------------|
59
- | diffusion | 16x16 | 0/20 | 300.0 | 5.06e-02 |
60
- | diffusion | 24x24 | 0/20 | 300.0 | 3.02e-02 |
61
- | diffusion | 32x32 | 0/20 | 300.0 | 2.51e-02 |
62
- | diffusion | 48x48 | 0/20 | 300.0 | 1.71e-02 |
63
- | diffusion | 64x64 | 0/20 | 300.0 | 9.03e-03 |
64
- | diffusion_advection | 16x16 | 20/20 | 87.7 | 6.12e-09 |
65
- | diffusion_advection | 24x24 | 7/20 | 263.1 | 5.34e-04 |
66
- | diffusion_advection | 32x32 | 0/20 | 300.0 | 5.15e-03 |
67
- | diffusion_advection | 48x48 | 0/20 | 300.0 | 2.14e-02 |
 
9
 
10
  # matrixpfn-base
11
 
 
 
12
  GNN-based learned preconditioner for sparse linear systems.
13
 
14
+ **Version**: 20260309-0350
15
 
16
  ## Usage
17
 
 
34
  | Parameter | Value |
35
  |-----------|-------|
36
  | Network | ContextResGCN |
37
+ | Layers | 24 |
38
+ | Embed | 128 |
39
+ | Hidden | 512 |
40
  | Context pairs | 10 |
41
+ | Parameters | 2,056,706 |
42
  | dtype | float32 |
43
 
44
  ## Training
45
 
46
+ - **epochs**: 5000
47
+ - **best_loss**: 0.006055
48
  - **loss_function**: l1_direct
49
+ - **batch_size**: 256
50
+ - **matrices_per_epoch**: 4
51
+ - **domains**: diffusion, diffusion_advection, graph_laplacian, elasticity, stokes, sbm, spectral_stress, variable_diffusion, variable_advection, enhanced_diffusion, enhanced_advection, random_sparse, directed_power_law
52
  - **grid_sizes**: [16, 24, 32, 48]
53
 
54
  ## Benchmark
55
 
56
  | domain | grid | converged | avg_iters | avg_residual |
57
  |--------|------|-----------|-----------|--------------|
58
+ | diffusion | 16x16 | 0/20 | 300.0 | 3.96e-02 |
59
+ | diffusion | 24x24 | 0/20 | 300.0 | 3.40e-02 |
60
+ | diffusion | 32x32 | 0/20 | 300.0 | 2.43e-02 |
61
+ | diffusion | 48x48 | 0/20 | 300.0 | 3.41e-02 |
62
+ | diffusion | 64x64 | 0/20 | 300.0 | 4.11e-02 |
63
+ | diffusion_advection | 16x16 | 0/20 | 300.0 | 3.69e-02 |
64
+ | diffusion_advection | 24x24 | 0/20 | 300.0 | 2.91e-02 |
65
+ | diffusion_advection | 32x32 | 0/20 | 300.0 | 2.59e-02 |
66
+ | diffusion_advection | 48x48 | 0/20 | 300.0 | 2.95e-02 |
model.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8b4b6368be29a328628295bd48f139f21a8a80053ca487020f3eb580ca7bcf76
3
- size 1741983
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:690ba93a9ce390d5ddb01f90389a6d909abe518b0c353cbd0ce18d439a6e9150
3
+ size 8352431