llm_learning

all about LLM (large language model)

1. model basic and config

• clone llm bin files and utils code
• set run at gpu num

2. chatglm2 inference

• load base model for inference

3. chatglm2 ptuning

• ptuning steps
• 1GPU.fp16
• 1GPU.int4
• 3GPU.fp16

4. chatglm2 load ptuning model for inference

• load ptuning model checkpoint for inference

5. LLaMA-Factory

• steps to use LLaMA-Factory fine tuning framework
• fine tuning chatglm2.lora.1gpu.fp16
• load model cli demo chatglm2.lora.1gpu
• support multiple dataset
• mulitple GPU on one node by accelerate

6. Prompt tech

• persona assignment

7. datasketch: lsh for data process, dup-remove

• Keep Chinese, English and underline, no punctuation
• minhash lsh for similarity
• lsh cluster

8. fastparquet

• read parquet file

9. dataset construct

• prompt prefix for sft dataset

10. FlagEmbedding

• BGE-M3 embedding for text

11. BERT

• bert for long text classification

12. datasets

• load and show data structure