v0.5.0

What's Changed

• dpp custom dataset now uses directory by @JRosenkranz in #163
• fix for #164 paged.py get head_size from model.config head_dim when defined by @rzbhatti in #165
• Add non-AIU utility tests as part of github action by @Abhishek-TAMU in #155
• Force static kv cache for DPP+fp8+chunked prefill by @ani300 in #167
• Add padding to chunk size and update env vars by @ani300 in #160
• Chunked debug by @Ssukriti in #176
• Allow any multiple of 64 during chunked prefill by @JRosenkranz in #169
• Fix IndexError: list index out of range when no prompt is found by @Abhishek-TAMU in #177

New Contributors

• @rzbhatti made their first contribution in #165
• @Ssukriti made their first contribution in #176

Full Changelog: v0.4.4...v0.5.0

v0.4.4

What's Changed

• Chunked prefill support for paged attention by @ani300 in #122
• fixed bug with loading validation info outputs by @JRosenkranz in #161

Full Changelog: v0.4.3...v0.4.4

v0.4.3

What's Changed

• fixed test_scripts program assertions by @JRosenkranz in #151
• Refactor Decoder Tests by @alex-jw-brooks in #93
• 🍱 include scripts with AFTU installs by @joerunde in #140
• Extend transformers support by @joerunde in #159

New Contributors

• @alex-jw-brooks made their first contribution in #93

Full Changelog: v0.4.2...v0.4.3

v0.4.2

What's Changed

• Prefix rewrite by @kcirred in #148

Full Changelog: v0.4.1...v0.4.2

v0.4.1

What's Changed

• [utils] decrement_version fix by @kcirred in #146
• DPP: check valid prompt shapes until one is available in the dataset by @lupalby in #145
• added dpp to test_scripts by @JRosenkranz in #147
• [dpp] store enforce_sizes in log name and added generic kwargs to get_default_validation_prefix by @kcirred in #136
• update llama model expectations tests by @JRosenkranz in #150

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

• Drive program script enhancements last n tokens by @JRosenkranz in #133

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• Test decoder long ctx by @kcirred in #117
• Get criteria from DT artifact by @lupalby in #118
• open gpt.json read-only to support parallel reading. by @gpaulsen in #126
• Drive Paged Program Script enhancements by @JRosenkranz in #128
• [dpp] eliminated pad_token_id from print by @kcirred in #130
• Add the ability to enforce homogeneous program ids in prefill in DPP script by @JRosenkranz in #131
• update test scripts to work with 4 layer micro model by @JRosenkranz in #134
• fixed inference.py for batch size 1 symbolic sdpa by @JRosenkranz in #135
• Make limits more flexible by @ani300 in #138
• Allow specific user prompts in DPP script by @JRosenkranz in #139
• Fix warmup to match vllm by @JRosenkranz in #141
• Add ability in DPP script to select one or many programs that satisfy min batch and min sequence requirements by @JRosenkranz in #137
• Fix paged generate with too much padding by @ani300 in #142
• clean_up_tokenization_spaces=True (default) causes incorrect number of tokens after sampling by @JRosenkranz in #143
• fixed issue where program_id was int when should have been string by @JRosenkranz in #144

New Contributors

• @lupalby made their first contribution in #118
• @gpaulsen made their first contribution in #126

Full Changelog: v0.2.3...v0.3.0

v0.2.3

What's Changed

• Add Env var to control cumulative test tokens generated per sequence by @Abhishek-TAMU in #125
• Fix model metric thresholds print out by @flaviabeo in #116
• ⬆️ bump transformers to 4.55.2 by @prashantgupta24 in #127

New Contributors

• @prashantgupta24 made their first contribution in #127

Full Changelog: v0.2.2...v0.2.3

v0.2.2

What's Changed

• modification to enforce_size behavior to start accepting samples even before enforcing sizes when there is sufficient space by @kcirred in #121

Full Changelog: v0.2.1...v0.2.2

v0.2.1

What's Changed

• Add a script which drives particular programs during decode when using paged model by @JRosenkranz in #109
• Add model testing results to pytest xml report by @pavi2707 in #114
• Include version in validation info outputs by @JRosenkranz in #119
• [test_decoder] add aftu version to validation log name and search for closest version log name by @kcirred in #115
• add truncation option to enforce_sizes to truncate prompts from a larger length to meet sizes that may not be available by @kcirred in #113
• Update transformers by @joerunde in #120

New Contributors

• @pavi2707 made their first contribution in #114
• @joerunde made their first contribution in #120

Full Changelog: v0.2.0...v0.2.1