Update llm-lora-ddp-gpus

benchmarks/llm/configs/llama3_8B_lora_single_device.yaml

@@ -1,83 +1,97 @@
 # Config for single device LoRA finetuning in lora_finetune_single_device.py
-# using a Llama3 8B Instruct model
+# using a Llama3.1 8B Instruct model
 #
 # This config assumes that you've run the following command before launching
 # this run:
-#   tune download meta-llama/Meta-Llama-3-8B-Instruct --output-dir /tmp/Meta-Llama-3-8B-Instruct --hf-token <HF_TOKEN>
+#   tune download meta-llama/Meta-Llama-3.1-8B-Instruct --output-dir /tmp/Meta-Llama-3.1-8B-Instruct --ignore-patterns "original/consolidated.00.pth"
 #
 # To launch on a single device, run the following command from root:
-#   tune run lora_finetune_single_device --config llama3/8B_lora_single_device
+#   tune run lora_finetune_single_device --config llama3_1/8B_lora_single_device
 #
 # You can add specific overrides through the command line. For example
 # to override the checkpointer directory while launching training
 # you can run:
-#   tune run lora_finetune_single_device --config llama3/8B_lora_single_device checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
+#   tune run lora_finetune_single_device --config llama3_1/8B_lora_single_device checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
 #
 # This config works only for training on single device.
 
 
+output_dir: /tmp/torchtune/llama3_1_8B/lora_single_device # /tmp may be deleted by your system. Change it to your preference.
+
 # Model Arguments
 model:
   _component_: torchtune.models.llama3_1.lora_llama3_1_8b
-  lora_attn_modules: ['q_proj', 'v_proj']
-  apply_lora_to_mlp: False
+  lora_attn_modules: ['q_proj', 'v_proj', 'output_proj']
+  apply_lora_to_mlp: True
   apply_lora_to_output: False
-  lora_rank: 8
-  lora_alpha: 16
+  lora_rank: 8  # higher increases accuracy and memory
+  lora_alpha: 16  # usually alpha=2*rank
+  lora_dropout: 0.0
 
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.llama3.llama3_tokenizer
-  path: /tmp/Meta-Llama-3-8B-Instruct/original/tokenizer.model
+  path: /tmp/Meta-Llama-3.1-8B-Instruct/original/tokenizer.model
+  max_seq_len: null
 
 checkpointer:
-  _component_: torchtune.training.FullModelMetaCheckpointer
-  checkpoint_dir: /tmp/Meta-Llama-3-8B-Instruct/original/
+  _component_: torchtune.training.FullModelHFCheckpointer
+  checkpoint_dir: /tmp/Meta-Llama-3.1-8B-Instruct/
   checkpoint_files: [
-    consolidated.00.pth
+    model-00001-of-00004.safetensors,
+    model-00002-of-00004.safetensors,
+    model-00003-of-00004.safetensors,
+    model-00004-of-00004.safetensors
   ]
   recipe_checkpoint: null
-  output_dir: /tmp/Meta-Llama-3-8B-Instruct/
+  output_dir: ${output_dir}
   model_type: LLAMA3
 resume_from_checkpoint: False
+save_adapter_weights_only: False
 
 # Dataset and Sampler
 dataset:
   _component_: torchtune.datasets.alpaca_cleaned_dataset
+  packed: False  # True increases speed
 seed: null
 shuffle: True
 batch_size: 2
 
 # Optimizer and Scheduler
 optimizer:
   _component_: torch.optim.AdamW
+  fused: True
   weight_decay: 0.01
   lr: 3e-4
 lr_scheduler:
-  _component_: torchtune.modules.get_cosine_schedule_with_warmup
+  _component_: torchtune.training.lr_schedulers.get_cosine_schedule_with_warmup
   num_warmup_steps: 100
 
 loss:
-  _component_: torch.nn.CrossEntropyLoss
+  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
 
 # Training
 epochs: 1
 max_steps_per_epoch: null
-gradient_accumulation_steps: 64
-compile: False
+gradient_accumulation_steps: 8  # Use to increase effective batch size
+clip_grad_norm: null
+compile: False  # torch.compile the model + loss, True increases speed + decreases memory
 
 # Logging
-output_dir: /tmp/lora_finetune_output
 metric_logger:
   _component_: torchtune.training.metric_logging.DiskLogger
-  log_dir: ${output_dir}
+  log_dir: ${output_dir}/logs
 log_every_n_steps: 1
-log_peak_memory_stats: False
+log_peak_memory_stats: True
 
 # Environment
 device: cuda
 dtype: bf16
-enable_activation_checkpointing: True
+
+# Activations Memory
+enable_activation_checkpointing: True  # True reduces memory
+enable_activation_offloading: False  # True reduces memory
+
 
 # Profiler (disabled)
 profiler:
@@ -100,6 +114,6 @@ profiler:
   # `torch.profiler.schedule` options:
   # wait_steps -> wait, warmup_steps -> warmup, active_steps -> active, num_cycles -> repeat
   wait_steps: 5
-  warmup_steps: 5
+  warmup_steps: 3
   active_steps: 2
   num_cycles: 1