v1.0.11

What's Changed

• refactor: remove unused variables. by @chungongyu in #114
• fix: .contiguous() makes O(N) iterative copy to O(1) memory copy. by @chungongyu in #115
• feat: fuse char encoder/decoder features in distogram head by @chungongyu in #116
• feat: apply bidirectional refiner for classification tasks. by @chungongyu in #118
• feat: add tools to 1. predict contact by ESM2-style models; 2. draw scater graphs; 3. tokenizer depp analysis; ProteinGym eval by @chungongyu in #104
• fix: attention_mask -> char_attention_mask by @chungongyu in #121

Full Changelog: v1.0.10...v1.0.11

v1.0.10

What's Changed

• fix: global shuffle to restore IID assumption by @chungongyu in #113
• refactor: code format to pass pylint check. by @chungongyu in #106

Full Changelog: v1.0.9...v1.0.10

v1.0.9

What's Changed

• fix: make sure that there is no identical sequences in each epoch. by @chungongyu in #107
• refactor: wrap dataset loading from data files to data_utils by @chungongyu in #108
• refactor: remove the sort function -> tokenize_dataset by @chungongyu in #109
• fix: variable shadowing (the env -> ctx rename) by @chungongyu in #110
• fix: precision in exception handling (The ImportError Fix) by @chungongyu in #111
• fix: typo by @chungongyu in #112

Full Changelog: v1.0.8...v1.0.9

v1.0.8

What's Changed

• feat: reweighting by inserse_log or log by @chungongyu in #103

Full Changelog: v1.0.7...v1.0.8

What's Changed

• feat: reweighting by inserse_log or log by @chungongyu in #103
• refactor: apply pow(cluster_weight, power) to balance abundance and diversity. by @chungongyu in #105

Full Changelog: v1.0.7...v1.0.8

v1.0.7

What's Changed

• refactor: rename contact_range names by @chungongyu in #97
• refactor: make deduplicate function reusable. by @chungongyu in #98
• feat: add PPL to classification.py; add --ignore_data_skip to train.py by @chungongyu in #99
• fix: eps=1e-6 for training stability. by @chungongyu in #100
• fix: apply ignore_index to CLE token if Ca is not available. by @chungongyu in #101
• fix: apply Tanh Soft-Cap to distoram logits. by @chungongyu in #102

Full Changelog: v1.0.6...v1.0.7

v1.0.6

What's Changed

• refactor: code format by @chungongyu in #87
• fix: disabled squared partition, OOM when sampling size is large. by @chungongyu in #88
• fix: *kwargs -> **kwargs by @chungongyu in #89
• refactor: classify tokens sampled from a clustered dataset by @chungongyu in #91
• docs: add install_env.sh by @chungongyu in #92
• fix: install_env.sh transformers==5.12.0 by @chungongyu in #93
• refactor: mask coords by pLDDT by @chungongyu in #94
• refactor: residue_idx -> char_position_ids by @chungongyu in #90
• fix: gly_idx differs in pdb and foldcomp dataset by @chungongyu in #95
• feat: dropout residue spans based on missing coordinates. by @chungongyu in #96

Full Changelog: v1.0.5...v1.0.6

v1.0.5

What's Changed

• fix: swap CB and O atoms at the pdb_transform by @chungongyu in #85
• refactor: to(device) is not required because of accelerator.prepare by @chungongyu in #86

Full Changelog: v1.0.4...v1.0.5

v1.0.4

What's Changed

• feat: add tools to make evaluation easy. by @chungongyu in #81
• fix: compute valid length from distogram_labels by @chungongyu in #82
• fix: cb_idx=3 -> la-proteina, cb_idx=4 -> FoldCompDataset by @chungongyu in #83
• refactor: to(device) is not required because of accelerator.prepare by @chungongyu in #84

Full Changelog: v1.0.3...v1.0.4

v1.0.3

What's Changed

• docs: architecure.svg by @chungongyu in #77
• fix: add --contact_ranges by @chungongyu in #78
• refactor: opener file for mode='w' by @chungongyu in #79
• fix: classify dedupe by id, contact precision F import and cutoff off-by-one by @chungongyu in #80

Full Changelog: v1.0.2...v1.0.3

v1.0.2

What's Changed

• feat: distributed generation and classification by @chungongyu in #74
• fix: truncate the padded items to match your exact original dataset size by @chungongyu in #75
• fix: compute protein length by char_attention_mask by @chungongyu in #76

Full Changelog: v1.0.1...v1.0.2