omega_a2a_test / output /training_config.yml
salmanshahid's picture
Upload output/training_config.yml with huggingface_hub
a5b53b4 verified
model:
_component_: models.lora_mmllama3_8b
lora_attn_modules:
- q_proj
- v_proj
apply_lora_to_mlp: false
apply_lora_to_output: false
lora_rank: 8
lora_alpha: 16
perception_tokens: 2
tokenizer:
_component_: models.a2a_tokenizer
path: checkpoints/Meta-Llama-3-8B-Instruct/tokenizer.model
checkpointer:
_component_: torchtune.utils.FullModelMetaCheckpointer
checkpoint_dir: checkpoints/Meta-Llama-3-8B-Instruct/
checkpoint_files:
- consolidated.00.pth
adapter_checkpoint: null
recipe_checkpoint: null
output_dir: output
model_type: LLAMA3
resume_from_checkpoint: false
dataset:
_component_: ds.EvenBatcher
dataset:
_component_: ds.RoundRobinDataset
datasets:
- _component_: ds.LlavaInstructDataset
ib_embed_path: ds/coco_llava_instruct/*.ib_embed.pt
clip_embed_path: ds/coco_llava_instruct/*.clip_embed.pt
caption_path: ds/coco_llava_instruct/*.caption.pt
train_on_input: false
- _component_: ds.LlavaInstructDataset
ib_embed_path: ds/vision_flan/*.ib_embed.pt
clip_embed_path: ds/vision_flan/*.clip_embed.pt
caption_path: ds/vision_flan/*.caption.pt
train_on_input: false
- _component_: ds.CaptionInstructDataset
ib_embed_path: ds/sam_llava/00.ib_embed.pt
clip_embed_path: ds/sam_llava/00.clip_embed.pt
caption_path: ds/sam_llava/00.caption.pt
train_on_input: false
seed: null
shuffle: true
batch_size: 4
optimizer:
_component_: torch.optim.AdamW
weight_decay: 0.01
lr: 0.0003
lr_scheduler:
_component_: torchtune.modules.get_cosine_schedule_with_warmup
num_warmup_steps: 100
loss:
_component_: torch.nn.CrossEntropyLoss
epochs: 1
max_steps_per_epoch: null
gradient_accumulation_steps: 64
compile: false
output_dir: /tmp/lora_finetune_output
metric_logger:
_component_: torchtune.utils.metric_logging.DiskLogger
log_dir: ${output_dir}
log_every_n_steps: null
device: cuda
dtype: bf16
enable_activation_checkpointing: false
profiler:
_component_: torchtune.utils.profiler
enabled: false