fix lora target module, require explicit flash attention, fix min logging steps, don't use adam8bit for int4, hash prepared datasets, support hf hub datasets

configs/llama_65B_alpaca.yml

@@ -21,7 +21,7 @@ lora_alpha: 16
 lora_dropout: 0.05
 lora_target_modules:
   - q_proj
-  - w_proj
+  - v_proj
 lora_fan_in_fan_out: false
 wandb_project: llama-65b-lora
 wandb_watch:

configs/llama_7B_4bit.yml

@@ -0,0 +1,41 @@
+base_model: decapoda-research/llama-7b-hf-int4
+base_model_config: decapoda-research/llama-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: true
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: data/last_run_prepared
+val_set_size: 0.04
+adapter: lora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 1024
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+#  - k_proj
+#  - o_proj
+lora_fan_in_fan_out: false
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model: checkpoint
+output_dir: ./lora-test
+batch_size: 8
+micro_batch_size: 2
+num_epochs: 3
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+gradient_checkpointing: false
+early_stopping_patience: 3
+resume_from_checkpoint:
+local_rank:
+load_4bit: true

configs/llama_7B_alpaca.yml

@@ -21,7 +21,7 @@ lora_alpha: 16
 lora_dropout: 0.05
 lora_target_modules:
   - q_proj
-  - w_proj
+  - v_proj
 lora_fan_in_fan_out: false
 wandb_project: llama-7b-lora
 wandb_watch: