v0.1.4: Dataset Streaming

• Support dataset streaming
• Fix LLaMA-2 #268
• Fix DeepSpeed ZeRO-3 model save #274
• Fix #242 #284

v0.1.3: Patch release

release v0.1.3

Former-commit-id: 62c68bcbf591516e8f90b47810bea6f710fd23f6

v0.1.2: LLaMA-2 Models

• Support LLaMA-2 (good issue #202 )
• Advanced configurations in Web UI
• Fix API (downgrade pydantic<2.0.0)
• Fix baichuan lora hparam #194 #212
• Fix padding #196
• Fix ZeRO-3 #199
• Allow pass args to app #213
• Code simplification
• Add ShareGPT dataset

v0.1.1

• Web UI: source_prefix, max_length, dev set
• Bug fix: reward token #179
• Update template #171 #177
• Bug fix: replace the Literal type with Enum for pydantic [1] #176
• Add Web demo #180

[1] pydantic/pydantic#5821, fastapi/sqlmodel#67

v0.1.0: All-in-one Web UI

• Fix gradient accumulation in PPO Trainer hiyouga/ChatGLM-Efficient-Tuning#299
• All-in-one Web UI by @hiyouga , @KanadeSiina and @codemayq

v0.0.9

add custom baichuan-13B code supports left-padding

Former-commit-id: 130ae986d13e4d03ae04d767a80c8863aee33fea