merged full weights after 1 ep training

configs/experiment.yml

@@ -0,0 +1,65 @@
+base_model: wcde/llama-7b-4bit-gr128
+base_model_config: wcde/llama-7b-4bit-gr128
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: true
+load_4bit: true
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter: lora
+lora_model_dir:
+sequence_len: 512
+max_packed_sequence_len: 
+lora_r: 16
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+lora_fan_in_fan_out:
+wandb_project: lora-experiment
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./lora-experiment
+batch_size: 128
+micro_batch_size: 8
+num_epochs: 4
+optimizer: adamw_torch
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00001
+train_on_inputs: false
+group_by_length: true
+bf16: true
+tf32: true
+gradient_checkpointing:
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize: 128
+gptq_model_v1: false
+warmup_steps: 10
+eval_steps: 10
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
+

configs/experiment2.yml

@@ -0,0 +1,68 @@
+base_model: ./lora-experiment/merged
+base_model_config: huggyllama/llama-7b 
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+load_4bit: false
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 
+lora_r: 16
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+lora_fan_in_fan_out:
+wandb_project: lora-experiment
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./lora-experiment
+batch_size: 8
+micro_batch_size: 4
+num_epochs: 4
+optimizer: adamw_torch
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: true
+bf16: true
+tf32: true
+gradient_checkpointing:
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 50
+eval_steps: 134
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
+

pytorch_model-00001-of-00002.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81b5054f1e8cd5d83fdb298b665041e19323261b00588855a2f0059db10d6517
-size 9976637950
+oid sha256:da8de3751aa79df207365b6546fbbc8ab8aeb8a5bbe411624953a0bec10a2810
+size 9976642558

pytorch_model-00002-of-00002.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc45dda38a2cdfa6f63b897a7250326951cccf192c7fc77af616667155b4d190
-size 3500316627
+oid sha256:08bfb0bb717720500f2947009dcfe8f152a5150c9834bd1b858d83e7ff8cfbec
+size 3500318291