Experimental ReLoRA (+qlora) implementation

examples/openllama-3b/relora.yml

@@ -0,0 +1,65 @@
+base_model: /home/charles/.cache/huggingface/hub/models--openlm-research--open_llama_3b/snapshots/8fcddba529aef0eda7293cc9a4171a3994648d2e/
+base_model_config: openlm-research/open_llama_3b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+prompt_format: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.005
+adapter: lora
+lora_model_dir:
+sequence_len: 512
+max_packed_sequence_len: 512
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.001
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+relora_steps: 20
+relora_warmup_steps: 10
+wandb_project: relora
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./lora-out
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
+bf16: false
+fp16: true
+tf32: false
+gradient_checkpointing: false
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+sdp_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 20
+save_steps: 50
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"