v1.0.7

We add CollieDatasetForTemplatedMultiTurnChat for multi-turn chat training in #162 and fix some issues.

What's Changed

• add load Internlm2 with safetensors by @KaiLv69 in #160
• feat(dataset): add multi-turn dataset with template by @KaiLv69 in #162
• fix(adalomo): fix adalomo without zero3 by @KaiLv69 in #164
• Update README.md by @xpqiu in #165
• fix(trainer): fix peft config save by @KaiLv69 in #166
• fix: ColliePipelineEngine has no attribute using_bf16_optimizer by @KYLN24 in #168
• perf(monitor): add init mode for monitor by @KaiLv69 in #170
• perf(config): set ds bs for hf models by @KaiLv69 in #175
• fix:CollieDatasetForTraining class not support list when loading data by @Zhyw0 in #173
• Add an example for chat fine-tuning by @KYLN24 in #169
• refactor(example): refactor examples by @KaiLv69 in #176

New Contributors

• @Zhyw0 made their first contribution in #173

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

Patch release: v1.0.6

A patch release mainly adds InternLM2 model.

Patch release: v1.0.5

A patch release mainly to resolve the following issues:

• Add AdaLomo optimizer
• Fix generation with collie model, now it's faster
• Fix model saving
• Fix checkpoint resuming
• Add more tutorials