Released in 2023, ChatGLM3 is the third in a series of pre-trained dialogue models jointly released by Zhipu AI and Tsinghua KEG. Building off the work in the “GLM: General Language Model Pretraining with Autoregressive Blank Infilling” paper, ChatGLM3-6B is an open-source offering in the ChatGLM3 series. Although it is open open-source it retains many excellent features of the first two generations such as smooth dialogue and easy deployment. The provided documentation works for both ChatGLM3-6B and ChatGLM2-6B
- Data Preparation
- Training with Predefined Configurations
- Checkpoint Conversion
- Model Evaluation
- Parameter Efficient Fine-Tuning (PEFT)
Feature |
Status |
---|---|
Data parallelism | ✓ |
Tensor parallelism | ✓ |
Pipeline parallelism | ✓ |
Interleaved Pipeline Parallelism Sched | N/A |
Sequence parallelism | ✓ |
Selective activation checkpointing | ✓ |
Gradient checkpointing | ✓ |
Partial gradient checkpointing | ✓ |
FP32/TF32 | ✓ |
AMP/FP16 | ✗ |
BF16 | ✓ |
TransformerEngine/FP8 | ✗ |
Multi-GPU | ✓ |
Multi-Node | ✓ |
Inference | N/A |
Slurm | ✓ |
Base Command Manager | ✓ |
Base Command Platform | ✓ |
Distributed data preprcessing | ✓ |
NVfuser | ✗ |
P-Tuning and Prompt Tuning | ✓ |
IA3 and Adapter learning | ✓ |
Distributed Optimizer | ✓ |
Distributed Checkpoint | ✓ |
Fully Shared Data Parallel | N/A |