Integrate new released Falcon3 Family base models

recipes/configs/falcon3/10B_full.yaml

@@ -0,0 +1,107 @@
+# Config for multi-device full finetuning in full_finetune_distributed.py
+# using a Falcon3 10B
+#
+# This config assumes that you've run the following command before launching
+# this run:
+#   tune download tiiuae/Falcon3-10B-Base --output-dir /tmp/Falcon3-10B --ignore-patterns None
+#
+# To launch on 4 devices, run the following command from root:
+#   tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config falcon3/10B_full
+#
+# You can add specific overrides through the command line. For example
+# to override the checkpointer directory while launching training
+# you can run:
+#   tune run --nnodes 1 --nproc_per_node 4 full_finetune_distributed --config falcon3/10B_full checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
+#
+# This config works best when the model is being fine-tuned on 2+ GPUs.
+# Single device full finetuning requires more memory optimizations. It's
+# best to use 10B_full.yaml for those cases
+
+# Tokenizer
+tokenizer:
+  _component_: torchtune.models.falcon3.falcon3_tokenizer
+  path: /tmp/Falcon3-10B-Base/tokenizer.json
+  max_seq_len: null
+
+# Dataset
+dataset:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  packed: False  # True increases speed
+seed: null
+shuffle: True
+
+# Model Arguments
+model:
+  _component_: torchtune.models.falcon3.falcon3_10b
+
+checkpointer:
+  _component_: torchtune.training.FullModelHFCheckpointer
+  checkpoint_dir: /tmp/Falcon3-10B-Base
+  checkpoint_files: [
+    model-00001-of-00005.safetensors,
+    model-00002-of-00005.safetensors,
+    model-00003-of-00005.safetensors,
+    model-00004-of-00005.safetensors,
+    model-00005-of-00005.safetensors
+  ]
+  recipe_checkpoint: null
+  output_dir: /tmp/Falcon3-10B-Base
+  model_type: FALCON3
+resume_from_checkpoint: False
+
+# Fine-tuning arguments
+batch_size: 2
+epochs: 1
+optimizer:
+  _component_: torch.optim.AdamW
+  fused: True
+  lr: 5e-6
+loss:
+  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
+max_steps_per_epoch: null
+gradient_accumulation_steps: 8  # Use to increase virtual batch size
+compile: False  # pytorch compile, set to true for better perf/memory
+optimizer_in_bwd: False  # True saves memory. Requires gradient_accumulation_steps=1
+
+# Training env
+device: cuda
+
+# Memory management
+enable_activation_checkpointing: True  # True reduces memory
+enable_activation_offloading: False  # True reduces memory
+
+# Reduced precision
+dtype: bf16
+
+# Logging
+metric_logger:
+  _component_: torchtune.training.metric_logging.DiskLogger
+  log_dir: ${output_dir}
+output_dir: /tmp/Falcon3-10B-Base-finetune
+log_every_n_steps: 1
+log_peak_memory_stats: True
+
+# Profiler (disabled)
+profiler:
+  _component_: torchtune.training.setup_torch_profiler
+  enabled: False
+
+  #Output directory of trace artifacts
+  output_dir: ${output_dir}/profiling_outputs
+
+  #`torch.profiler.ProfilerActivity` types to trace
+  cpu: True
+  cuda: True
+
+  #trace options passed to `torch.profiler.profile`
+  profile_memory: False
+  with_stack: False
+  record_shapes: True
+  with_flops: False
+
+  # `torch.profiler.schedule` options:
+  # wait_steps -> wait, warmup_steps -> warmup, active_steps -> active, num_cycles -> repeat
+  wait_steps: 5
+  warmup_steps: 3
+  active_steps: 2
+  num_cycles: 1

recipes/configs/falcon3/10B_full_single_device.yaml

@@ -0,0 +1,108 @@
+# Config for single device full finetuning in full_finetune_single_device.py
+# using a Falcon3 10B
+#
+# This config assumes that you've run the following command before launching
+# this run:
+#   tune download tiiuae/Falcon3-10B-Base --output-dir /tmp/Falcon3-10B --ignore-patterns None
+#
+# The default config uses an optimizer from bitsandbytes. If you do not have it installed,
+# you can install it with
+#   pip install bitsandbytes
+#
+# To launch on a single device, run the following command from root:
+#   tune run full_finetune_single_device --config falcon3/10B_full_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 full_finetune_single_device --config falcon3/10B_full_single_device checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
+#
+# This config works only for training on single device.
+
+# Tokenizer
+tokenizer:
+  _component_: torchtune.models.falcon3.falcon3_tokenizer
+  path: /tmp/Falcon3-10B-Base/tokenizer.json
+  max_seq_len: null
+
+# Dataset
+dataset:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  packed: False  # True increases speed
+seed: null
+shuffle: True
+
+# Model Arguments
+model:
+  _component_: torchtune.models.falcon3.falcon3_10b
+
+checkpointer:
+  _component_: torchtune.training.FullModelHFCheckpointer
+  checkpoint_dir: /tmp/Falcon3-10B-Base
+  checkpoint_files: [
+    model-00001-of-00005.safetensors,
+    model-00002-of-00005.safetensors,
+    model-00003-of-00005.safetensors,
+    model-00004-of-00005.safetensors,
+    model-00005-of-00005.safetensors
+  ]
+  recipe_checkpoint: null
+  output_dir: /tmp/Falcon3-10B-Base
+  model_type: FALCON3
+resume_from_checkpoint: False
+
+# Fine-tuning arguments
+batch_size: 1
+epochs: 1
+optimizer:
+  _component_: torch.optim.AdamW
+  lr: 5e-6
+optimizer_in_bwd: True  # True saves memory. Requires gradient_accumulation_steps=1
+loss:
+  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
+max_steps_per_epoch: null
+gradient_accumulation_steps: 1  # Use to increase virtual batch size
+compile: False  # pytorch compile, set to true for better perf/memory
+
+# Training environment
+device: cuda
+
+# Memory management
+enable_activation_checkpointing: True  # True reduces memory
+enable_activation_offloading: False  # True reduces memory
+
+# Reduced precision
+dtype: bf16
+
+# Logging
+metric_logger:
+  _component_: torchtune.training.metric_logging.DiskLogger
+  log_dir: ${output_dir}
+output_dir: /tmp/Falcon3-10B-Base-finetune
+log_every_n_steps: 1
+log_peak_memory_stats: True
+
+# Profiler (disabled)
+profiler:
+  _component_: torchtune.training.setup_torch_profiler
+  enabled: False
+
+  #Output directory of trace artifacts
+  output_dir: ${output_dir}/profiling_outputs
+
+  #`torch.profiler.ProfilerActivity` types to trace
+  cpu: True
+  cuda: True
+
+  #trace options passed to `torch.profiler.profile`
+  profile_memory: False
+  with_stack: False
+  record_shapes: True
+  with_flops: False
+
+  # `torch.profiler.schedule` options:
+  # wait_steps -> wait, warmup_steps -> warmup, active_steps -> active, num_cycles -> repeat
+  wait_steps: 5
+  warmup_steps: 3
+  active_steps: 2
+  num_cycles: 1

recipes/configs/falcon3/10B_lora.yaml

@@ -0,0 +1,117 @@
+# Config for multi-device LoRA finetuning in lora_finetune_distributed.py
+# using a Falcon3 10B
+#
+# This config assumes that you've run the following command before launching
+# this run:
+#   tune download tiiuae/Falcon3-10B-Base --output-dir /tmp/Falcon3-10B --ignore-patterns None
+#
+# To launch on 2 devices, run the following command from root:
+#   tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config falcon3/10B_lora
+#
+# You can add specific overrides through the command line. For example
+# to override the checkpointer directory while launching training
+# you can run:
+#   tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config falcon3/10B_lora checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
+#
+# This config works best when the model is being fine-tuned on 2+ GPUs.
+# For single device LoRA finetuning please use 10B_lora_single_device.yaml
+
+
+# Model Arguments
+model:
+  _component_: torchtune.models.falcon3.lora_falcon3_10b
+  lora_attn_modules: ['q_proj', 'v_proj', 'output_proj']
+  apply_lora_to_mlp: True
+  apply_lora_to_output: False
+  lora_rank: 8  # higher increases accuracy and memory
+  lora_alpha: 16  # usually alpha=2*rank
+  lora_dropout: 0.0
+
+tokenizer:
+  _component_: torchtune.models.falcon3.falcon3_tokenizer
+  path: /tmp/Falcon3-10B-Base/tokenizer.json
+  max_seq_len: null
+
+checkpointer:
+  _component_: torchtune.training.FullModelHFCheckpointer
+  checkpoint_dir: /tmp/Falcon3-10B-Base
+  checkpoint_files: [
+    model-00001-of-00005.safetensors,
+    model-00002-of-00005.safetensors,
+    model-00003-of-00005.safetensors,
+    model-00004-of-00005.safetensors,
+    model-00005-of-00005.safetensors
+  ]
+  recipe_checkpoint: null
+  output_dir: /tmp/Falcon3-10B-Base
+  model_type: FALCON3
+resume_from_checkpoint: False
+
+# Dataset and Sampler
+dataset:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  packed: False  # True increases speed
+seed: null
+shuffle: True
+
+
+# Optimizer and Scheduler
+optimizer:
+  _component_: torch.optim.AdamW
+  fused: True
+  weight_decay: 0.01
+  lr: 3e-4
+lr_scheduler:
+  _component_: torchtune.training.lr_schedulers.get_cosine_schedule_with_warmup
+  num_warmup_steps: 100
+
+loss:
+  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
+
+# Training
+batch_size: 2
+epochs: 1
+max_steps_per_epoch: null
+gradient_accumulation_steps: 8  # Use to increase virtual batch size
+compile: False  # pytorch compile, set to true for better perf/memory
+
+# Logging
+metric_logger:
+  _component_: torchtune.training.metric_logging.DiskLogger
+  log_dir: ${output_dir}
+output_dir: /tmp/Falcon3-10B-Base-finetune
+log_every_n_steps: 1
+log_peak_memory_stats: False
+
+# Environment
+device: cuda
+dtype: bf16
+enable_activation_checkpointing: False  # True reduces memory
+enable_activation_offloading: False  # True reduces memory
+
+# Show case the usage of pytorch profiler
+# Set enabled to False as it's only needed for debugging training
+profiler:
+  _component_: torchtune.training.setup_torch_profiler
+
+  enabled: False
+
+  #Output directory of trace artifacts
+  output_dir: ${output_dir}/profiling_outputs
+
+  #`torch.profiler.ProfilerActivity` types to trace
+  cpu: True
+  cuda: True
+
+  #trace options passed to `torch.profiler.profile`
+  profile_memory: False
+  with_stack: False
+  record_shapes: True
+  with_flops: False
+
+  # `torch.profiler.schedule` options:
+  # wait_steps -> wait, warmup_steps -> warmup, active_steps -> active, num_cycles -> repeat
+  wait_steps: 5
+  warmup_steps: 5
+  active_steps: 2
+  num_cycles: 1