Code

[DRAFT] s-quark support#1154

Code

Open

calad0i wants to merge 69 commits intofastmachinelearning:mainfastmachinelearning/hls4ml:mainfrom calad0i:s-quark-prcalad0i/hls4ml:s-quark-prCopy head branch name to clipboard

+4,824-496

Commits

Commits on Dec 15, 2024

• import converter dependencies lazily

  calad0i

  committed
  ef20647

  Copy full SHA for ef20647

• make tf and qkeras optionl, stop assuming keras is tf.keras

  calad0i

  committed
  028b4d0

  Copy full SHA for 028b4d0

• less mandatory dependency

  calad0i

  committed
  72eb053

  Copy full SHA for 72eb053

• fix dsp_aware_pruning test import path

  calad0i

  committed
  63af2ac

  Copy full SHA for 63af2ac

• fix broken setup.cfg after rebase, rm pyparsing

  calad0i

  committed
  c11dddb

  Copy full SHA for c11dddb

• purge qkeras workaround

  calad0i

  committed
  d9aaa1a

  Copy full SHA for d9aaa1a

Commits on Dec 16, 2024

• switch to pyproject.toml

  Show description for 4854423
  calad0i

  committed
  4854423

  Copy full SHA for 4854423

• format

  calad0i

  committed
  06f9cda

  Copy full SHA for 06f9cda

• rm useless flake8 config in pyprject.toml

  calad0i

  committed
  014c1db

  Copy full SHA for 014c1db

• Add hint on import failure

  calad0i

  committed
  d3c8881

  Copy full SHA for d3c8881

• leftover

  calad0i

  committed
  738e5b0

  Copy full SHA for 738e5b0

• rm setup.py from manifest

  calad0i

  committed
  bc7778b

  Copy full SHA for bc7778b

• manifest fix 2

  calad0i

  committed
  b76b5cb

  Copy full SHA for b76b5cb

• keras v3 object based parser

  calad0i

  committed
  b7f60f5

  Copy full SHA for b7f60f5

• sequential and i/o tensor name parsing fix

  calad0i

  committed
  a7206b4

  Copy full SHA for a7206b4

• support activation layers

  calad0i

  committed
  1605f96

  Copy full SHA for 1605f96

• consistent v2 weight reader behavior

  calad0i

  committed
  a8aa489

  Copy full SHA for a8aa489

• add v3 conv handlers

  calad0i

  committed
  eafe8b9

  Copy full SHA for eafe8b9

• add test

  calad0i

  committed
  6b8a44c

  Copy full SHA for 6b8a44c

Commits on Dec 17, 2024

• pre-commit fix

  calad0i

  committed
  3f8acb5

  Copy full SHA for 3f8acb5

• revert keras v2 converter

  calad0i

  committed
  d2ccfb4

  Copy full SHA for d2ccfb4

• make reshape handler compatiable with keras v3

  calad0i

  committed
  0334960

  Copy full SHA for 0334960

• add general transpose for vivado/vitis

  calad0i

  committed
  074b4b6

  Copy full SHA for 074b4b6

• general einsum support for io_parallel and latency

  calad0i

  committed
  29674db

  Copy full SHA for 29674db

• add tests for einsumdense

  calad0i

  committed
  1fb23b9

  Copy full SHA for 1fb23b9

• keras v3 converter clean-up

  calad0i

  committed
  5489803

  Copy full SHA for 5489803

• add symbolic quantized interval

  calad0i

  committed
  5e18781

  Copy full SHA for 5e18781

• preliminary bit-exact precision derivation opt pass

  calad0i

  committed
  02ff0c3

  Copy full SHA for 02ff0c3

• squark layer support start

  calad0i

  committed
  7c47be9

  Copy full SHA for 7c47be9

• fix einsum_dense precision computation

  calad0i

  committed
  43847c4

  Copy full SHA for 43847c4

• add leftover

  calad0i

  committed
  afdaf21

  Copy full SHA for afdaf21

• qdense fix

  calad0i

  committed
  0da5cd0

  Copy full SHA for 0da5cd0

• support batch_norm

  calad0i

  committed
  6b73774

  Copy full SHA for 6b73774

• support merge layers

  calad0i

  committed
  93043de

  Copy full SHA for 93043de

• support bit-exact q_einsum and fix precision trace for multi inp layers

  calad0i

  committed
  d8708f5

  Copy full SHA for d8708f5

• add einsum test

  calad0i

  committed
  cba1411

  Copy full SHA for cba1411

• declare all softmax attrs in layer class

  calad0i

  committed
  f8ae929

  Copy full SHA for f8ae929

• fix lazy import in handler

  calad0i

  committed
  9326ad5

  Copy full SHA for 9326ad5

• cleanup einsum handler

  calad0i

  committed
  0cde312

  Copy full SHA for 0cde312

• cleanup einsum handler

  calad0i

  committed
  b97d01e

  Copy full SHA for b97d01e

• more granular control over softmax for vivado

  calad0i

  committed
  c34abbe

  Copy full SHA for c34abbe

• properly propagate inv/exp_table_size

  calad0i

  committed
  7ea6310

  Copy full SHA for 7ea6310

• support bit-exact softmax for stable impl

  calad0i

  committed
  0ecd12e

  Copy full SHA for 0ecd12e

• bit-exact softmax fix and leftovers

  calad0i

  committed
  fdfaac5

  Copy full SHA for fdfaac5

• softmax table fixer update

  calad0i

  committed
  3f4c642

  Copy full SHA for 3f4c642

• support input scaler in softmax

  calad0i

  committed
  bf99e83

  Copy full SHA for bf99e83

• support multidim parallel softmax

  calad0i

  committed
  b925bc8

  Copy full SHA for b925bc8

• fuse quantizer when possible

  calad0i

  committed
  c611c77

  Copy full SHA for c611c77

• partial activation, fix input precision in SAT mode

  calad0i

  committed
  b7975fa

  Copy full SHA for b7975fa

• fix padded convXd precition derivation rule

  calad0i

  committed
  3d1431e

  Copy full SHA for 3d1431e

• add unary lut support

  calad0i

  committed
  f97d4d8

  Copy full SHA for f97d4d8

• fix bit-exact corner case introduced by reverse flow

  calad0i

  committed
  61e76a2

  Copy full SHA for 61e76a2

• general data_t inference

  calad0i

  committed
  e50e731

  Copy full SHA for e50e731

• softmax compatbility

  calad0i

  committed
  4a6b0b5

  Copy full SHA for 4a6b0b5

• fix typo in einsum handler

  calad0i

  committed
  a6128ae

  Copy full SHA for a6128ae

• fix more typos

  calad0i

  committed
  5190c33

  Copy full SHA for 5190c33

• MHA 🎉

  calad0i

  committed
  9cdb67c

  Copy full SHA for 9cdb67c

• fix einsum and softmax template typos

  calad0i

  committed
  5bcae96

  Copy full SHA for 5bcae96

• assert einsum ops doesnot include direct sum operation

  calad0i

  committed
  d780de2

  Copy full SHA for d780de2

• style

  calad0i

  committed
  e3cef20

  Copy full SHA for e3cef20

• fix mha layer indexing

  calad0i

  committed
  2bcf9e7

  Copy full SHA for 2bcf9e7

• switch to model opt

  calad0i

  committed
  c426ddc

  Copy full SHA for c426ddc

• pooling layers

  calad0i

  committed
  a749c27

  Copy full SHA for a749c27

• handle stray inputs

  calad0i

  committed
  0317b5b

  Copy full SHA for 0317b5b

• fix pooling layer accum_t

  calad0i

  committed
  b38420d

  Copy full SHA for b38420d

• bit-exact concatenate

  calad0i

  committed
  a2d6e1a

  Copy full SHA for a2d6e1a

Commits on Jan 17, 2025

• rm np.float_ in favor of numpy >=2.0

  calad0i

  committed
  af5c798

  Copy full SHA for af5c798

Commits on Jan 18, 2025

• add comments

  calad0i

  committed
  c32df4b

  Copy full SHA for c32df4b

• skip non-bit-exact compatiable softmax in bit-exact pass

  calad0i

  committed
  fe0ff2f

  Copy full SHA for fe0ff2f