Skip to content

Merge multi-precision support into main - #213

Merged
rouson merged 115 commits into
mainfrom
develop
Oct 17, 2024
Merged

Merge multi-precision support into main#213
rouson merged 115 commits into
mainfrom
develop

Conversation

@rouson

@rousonrouson commented Oct 2, 2024•
edited
Loading

Copy link
Copy Markdown
Contributor

This PR is will be ready to merge either when gfortran's support for parameterized derived types (PDTs) has been fixed or we have decided to drop support for gfortran.

The develop branch that this PR will merge into main resulted from merging three significant development branches:

  1. The parallel branch added support for automatic parallelization of training via do concurrent.
  2. The parameterized-derived-types branch added the ability to support both default-real and double precision inference without requiring recompiling.
  3. The rename-inference-engine_t branch, which globally renames the project from "Inference-Engine" to Fiats and correspondingly renames numerous program entities (files, modules, procedures, documentation, etc.)

The latter three branches have been deleted in order to clarify the path forward. Starting now, new work should branch off of develop.

rouson added 30 commits August 27, 2024 19:52
@rouson
refac(neuron_t): add kind parameter
e717d93
@rouson
refac(layer_t): add kind parameter
8efec6f
@rouson
refac(tensor_map): add kind parameter
1e25cd6
@rouson
refac(tensor_m): add kind parameter
c63d9bb
@rouson
refac(inference_engine_t): add kind parameter
8bd7ae7
@rouson
test(inference_engine_t): use eqv op for compare
dff93d9
This commit replaces the inference_engine_t type-bound, generic
subtraction operator(-) with an user-defined approximate
equivalence operator(==), which simplifies comparisons by
eliminating the diffence_t type. The new operator is used in
inference_engine_test_m instead of the difference_t, operator(-),
and the norm() function.
@rouson
refac: define generic activation/derivative
1c1a93e
This commit sets up for adding double-precision support by
making the activation and activation-derivative bindings generic.
@rouson
chore(tensor_map_test): match file name to module
216e59b
@rouson
feat(tensor_t):dble-prec constructor/accessor/test
72b6348
@rouson
feat(step): add double-precision activation
dd0cb91
@rouson
feat(sigmoid):double-precision activation/derivat
6ceabf0
@rouson
feat(relu): add double-precision activation/deriv
0d69cce
@rouson
feat(gelu): add double-prec activation/derivatie
59547e7
@rouson
feat: finish double-prec activation/derivs
fc54a18
@rouson
refac(tensor_map):mk type-bound proc names generic
84d21f7
@rouson
feat(tensor_map): add double-precision procedures
d878713
This commit adds double-precision versions of all tensor_map_t
type-bound and constructor procedures except the
default_real_from_json constructor function.
@rouson
refac(neuron): mk procedure names generic bindings
d011f0d
@rouson
refac(layer):make procedure names generic bindings
c496147
@rouson
feat(neuron_t): add double-precision procedures
85fba96
@rouson
feat(layer_t): add double-precision procedures
71bbff8
@rouson
feat(layer_t): add double-precision procedures
9c99739
@rouson
Merge remote-tracking branch 'refs/remotes/origin/parameterized-deriv…
d031662
…ed-types' into parameterized-derived-types
@rouson
feat(trainable_engine_t): add real kind parameter
d919e44
@rouson
feat(trainable_engine_t): add generic "predict"
070b4c3
This commit adds a generic "predict" binding to follow PyTorch
nomenclature. For now, the equivalent "infer" generic binding
remains to support existing code.
@rouson
refac(trainable_engine): make bindings generic
de37292
@rouson
feat(input_output_pair_t): add kind parameter
a5636dd
also make bindings generic
@rouson
feat(input_output_pair): support double precision
d82b498
@rouson
feat(mini_match): kind parameter +generic bindings
0e284a6
@rouson
feat(mini_match): support double precision
0ff2681
@rouson
feat(hyperparameters): support double precision
7ea4e3b
also update julienne dependency to 1.2.0 for double-precsion value
extraction from JSON files.
@rouson
fix(unmapped_network): rm ambiguous specific proc
4c76b7a
also adjust demo/app/infer-aerosol.f90 to reflect the
renaming of unmapped_engine_t to unmapped_network_t
@rouson
chore(trainable_network):replaces trainable_engine
e4f8ba9
@rouson
fix(train-cloud-micro):let write_lines() open file
e89c7b4
@rouson
Merge pull request#217from BerkeleyLab/simplify-class-relationships
e7794c5
Simplify class relationships
@rouson
chore(inference_engine): -> neural_network/fiats
b23524d
This commit
1. Renames the project from inference-engine to fiats.
2. Renames inference_engine_t to neural_network_t.
@rouson
doc(ford): update project name & summary
2fbb161
@rouson
chore: rename asymmetric_{engine,network}_test
bada395
@rouson
doc(README): update compiler status
be7960e
@rouson
refac(json): rename acceptable_engine_tag
4b09cdc
This commit renames "acceptable_engine_tag" to
"minimum_acceptable_tag" in the network JSON file format.
@rouson
chore: rm unused exhcnage_t type
d69324d
@rouson
doc(README.md): update
b8473d6
@rouson
chore: white space edits
6f03491
@rouson
chore(training_configuration): rm unused function
31ecd59
@rouson
doc(UML): update class diagram
c574fd0
@rouson
doc(uml): add components & type-bound procedures
7f606a4
@rouson
doc(README/uml): add link to UML diagram
5f16334
@rouson
doc(uml): add classes & relationships
5292c9b
@rouson
Revert "chore(training_configuration): rm unused function"
48f2268
This reverts commit 31ecd59.
@rouson
Merge pull request#218from BerkeleyLab/rename-inference_engine_t
5f844c8
Rename Inference-Engine to Fiats
@rouson

rouson commented Oct 16, 2024

Copy link
Copy Markdown
ContributorAuthor

Based on correspondences with gfortran developers it appears that the fixes to gfortran that would be required for this repository to continue supporting gfortran are likely to be a ways out. Now that the develop branch also includes a merge of PR #218, delaying merging this PR would also delay the renaming of Inference-Engine to Fiats. The name change is somewhat urgent in the sense that this repository is now being featured in several important documents, including anticipated third-party submissions by outside parties in response to the FASST RFI and the ORNL/DOE "New Frontiers" RFP. It therefore makes sense to merge this PR post haste.

@rouson
rouson marked this pull request as ready for review October 16, 2024 22:21
ktras
ktras reviewed Oct 16, 2024
View reviewed changes
Comment threadREADME.md Outdated
@rouson@ktras
doc(README): fix typo in branch name
dcc407e
Co-authored-by: Katherine Rasmussen <krasmussen@lbl.gov>
ktras
ktras reviewed Oct 16, 2024
View reviewed changes
Comment threadREADME.md Outdated
ktras
ktras reviewed Oct 16, 2024
View reviewed changes
Comment threadREADME.md Outdated
rousonand others added 2 commits October 16, 2024 16:55
@rouson
doc(README): specify compiler issue
55c377b
@rouson@ktras
doc(README): fix CCE compiler status text
53c8550
Co-authored-by: Katherine Rasmussen <krasmussen@lbl.gov>
@rouson
rouson merged commit b2a0663 into mainOct 17, 2024
@rouson
rouson deleted the develop branch October 17, 2024 00:04
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@ktrasktrasktras left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@rouson@ktras