diff options
| author | Marcin Junczys-Dowmunt <marcinjd@microsoft.com> | 2022-02-06 23:33:58 +0300 |
|---|---|---|
| committer | Marcin Junczys-Dowmunt <marcinjd@microsoft.com> | 2022-02-06 23:33:58 +0300 |
| commit | 3cf9e83bac033a80c7b27912435917d65e40f095 (patch) | |
| tree | 78ff005463dc4334854d7b4980260b345777b9f6 | |
| parent | 8da539e835e8661d00697c8bd01164e64ab9ce62 (diff) | |
| parent | e8ea37cd5b85e3df817b9ced68bef9cc64b45d16 (diff) | |
resolve conflicts
| -rw-r--r-- | CHANGELOG.md | 6 | ||||
| -rw-r--r-- | VERSION | 3 | ||||
| -rw-r--r-- | src/data/corpus_base.cpp | 20 |
3 files changed, 8 insertions, 21 deletions
diff --git a/CHANGELOG.md b/CHANGELOG.md index d42c652e..cf0f3844 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -9,7 +9,11 @@ and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0. ## [Unreleased] ### Added -- Adds option --add-lsh to marian-conv which allows the LSH to be memory-mapped. +- Parallelized data reading with e.g. `--data-threads 8` +- Top-k sampling during decoding with e.g. `--output-sampling topk 10` +- Improved mixed precision training with `--fp16` +- Set FFN width in decoder independently from encoder with e.g. `--transformer-dim-ffn 4096 --transformer-decoder-dim-ffn 2048` +- Adds option `--add-lsh` to marian-conv which allows the LSH to be memory-mapped. - Early stopping based on first, all, or any validation metrics via `--early-stopping-on` - Compute 8.6 support if using CUDA>=11.1 - Support for RMSNorm as drop-in replace for LayerNorm from `Biao Zhang; Rico Sennrich (2019). Root Mean Square Layer Normalization`. Enabled in Transformer model via `--transformer-postprocess dar` instead of `dan`. @@ -1,2 +1 @@ -v1.10.28 - +v1.10.42 diff --git a/src/data/corpus_base.cpp b/src/data/corpus_base.cpp index 20301103..636752c9 100644 --- a/src/data/corpus_base.cpp +++ b/src/data/corpus_base.cpp @@ -566,23 +566,7 @@ size_t CorpusBase::getNumberOfTSVInputFields(Ptr<Options> options) { return 0; } -<<<<<<< HEAD -void SentenceTuple::setWeights(const std::vector<float>& weights) { - if(weights.size() != 1) { // this assumes a single sentence-level weight is always fine - ABORT_IF(empty(), "Source and target sequences should be added to a tuple before data weights"); - auto numWeights = weights.size(); - auto numTrgWords = back().size(); - // word-level weights may or may not contain a weight for EOS tokens - if(numWeights != numTrgWords && numWeights != numTrgWords - 1) - LOG(warn, - "[warn] " - "Number of weights ({}) does not match the number of target words ({}) in line #{}", - numWeights, - numTrgWords, - id_); - } - weights_ = weights; -======= +#if 0 // experimental: hide inline-fix source tokens from cross attention std::vector<float> SubBatch::crossMaskWithInlineFixSourceSuppressed() const { @@ -630,8 +614,8 @@ std::vector<float> SubBatch::crossMaskWithInlineFixSourceSuppressed() const ABORT_IF(batchWords() != 0/*n/a*/ && numWords != batchWords(), "batchWords() inconsistency??"); } return m; ->>>>>>> master } +#endif } // namespace data } // namespace marian |