wave-on-discord/llama-3-70b-no-robots-adapter

See axolotl config

axolotl version: 0.4.0

base_model: NousResearch/Meta-Llama-3-70B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer  # PreTrainedTokenizerFast

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: Doctor-Shotgun/no-robots-sharegpt
    type: sharegpt
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./out/qlora-llama3-70b

adapter: qlora
lora_model_dir:

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

lora_r: 64
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
  fsdp_limit_all_gathers: true
  fsdp_sync_module_states: true
  fsdp_offload_params: false
  fsdp_use_orig_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD
special_tokens:
  pad_token: <|end_of_text|>

out/qlora-llama3-70b

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

• Loss: 1.5377

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• 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
1.7144	0.02	1	1.8096
1.6886	0.25	12	1.5367
1.6174	0.49	24	1.5176
1.5848	0.74	36	1.5054
1.6542	0.98	48	1.5018
1.572	1.21	60	1.4993
1.5966	1.45	72	1.5007
1.5643	1.7	84	1.4981
1.6312	1.94	96	1.4980
1.5311	2.16	108	1.5027
1.519	2.41	120	1.5109
1.4034	2.65	132	1.5165
1.4658	2.9	144	1.5187
1.5434	3.11	156	1.5264
1.4608	3.35	168	1.5364
1.4529	3.6	180	1.5377
1.3893	3.85	192	1.5377

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0.dev0
• Pytorch 2.2.1
• Datasets 2.15.0
• Tokenizers 0.15.0