See axolotl config

axolotl version: 0.4.0

adapter: qlora
base_model: meta-llama/Meta-Llama-3-70B-Instruct
bf16: auto
datasets:
- conversation: llama-3
  path: a265546be8c24d59bfdc6ba69431b635/./data/with_function_response/original_clean/function_used_training_shuffled.jsonl
  type: sharegpt
- conversation: llama-3
  path: a265546be8c24d59bfdc6ba69431b635/./data/with_function_response/original_clean/function_not_used_training.jsonl
  type: sharegpt
- conversation: llama-3
  path: a265546be8c24d59bfdc6ba69431b635/./data/with_function_response/parallel_call/parallel_data_training.jsonl
  type: sharegpt
debug: null
deepspeed: null
early_stopping_patience: null
eval_table_size: null
evals_per_epoch: 4
flash_attention: true
fp16: null
fsdp:
- full_shard
- auto_wrap
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_cpu_ram_efficient_loading: true
  fsdp_limit_all_gathers: true
  fsdp_offload_params: true
  fsdp_sharding_strategy: FULL_SHARD
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sync_module_states: true
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
  fsdp_use_orig_params: false
gradient_accumulation_steps: 2
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
group_by_length: false
hub_model_id: liuylhf/empower-functions-llama3-70b-parallel-all-linear
learning_rate: 0.0002
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lora_target_modules: null
lr_scheduler: cosine
micro_batch_size: 4
model_type: LlamaForCausalLM
num_epochs: 4
optimizer: adamw_torch
output_dir: a265546be8c24d59bfdc6ba69431b635/model
pad_to_sequence_len: true
resume_from_checkpoint: null
sample_packing: true
saves_per_epoch: 10
sequence_len: 4096
special_tokens:
  pad_token: <|end_of_text|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
val_set_size: 0.05
wandb_entity: null
wandb_log_model: null
wandb_name: null
wandb_project: null
wandb_watch: null
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

empower-functions-llama3-70b-parallel-all-linear

This model is a fine-tuned version of meta-llama/Meta-Llama-3-70B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0436

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
2.0962	0.0067	1	2.0635
0.0715	0.2492	37	0.0770
0.0556	0.4983	74	0.0600
0.0559	0.7475	111	0.0549
0.0542	0.9966	148	0.0523
0.0439	1.2256	185	0.0505
0.0484	1.4747	222	0.0496
0.043	1.7239	259	0.0477
0.0467	1.9731	296	0.0464
0.0406	2.2020	333	0.0462
0.0424	2.4512	370	0.0453
0.0378	2.7003	407	0.0443
0.0382	2.9495	444	0.0435
0.0352	3.1785	481	0.0439
0.0328	3.4276	518	0.0438
0.0329	3.6768	555	0.0437
0.0378	3.9259	592	0.0436

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0
• Pytorch 2.1.2+cu121
• Datasets 2.15.0
• Tokenizers 0.19.1