mjobe105/qlora-largersequence-dolphindata

See axolotl config

axolotl version: 0.4.0

base_model: meta-llama/Meta-Llama-3-70B-Instruct
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer  # PreTrainedTokenizerFast

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: /workspace/data/dataset/hex_phi_dolphin_responses.jsonl
    ds_type: json
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: /workspace/data/out/qlora

adapter: qlora
lora_model_dir:

sequence_len: 8192
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

lora_r: 32
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_torch
lr_scheduler: cosine
learning_rate: 0.00001

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: 100
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

workspace/data/out/qlora

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: 2.0567

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: 1e-05
• 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: 100
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
1.8335	0.1	1	2.0723
1.7855	0.3	3	2.0737
1.8449	0.6	6	2.0734
1.99	0.9	9	2.0728
1.7771	1.2	12	2.0731
1.8706	1.5	15	2.0735
1.8152	1.8	18	2.0728
1.7398	2.1	21	2.0716
1.7819	2.4	24	2.0715
1.7759	2.7	27	2.0706
1.8823	3.0	30	2.0684
1.7957	3.3	33	2.0649
1.8849	3.6	36	2.0624
1.8567	3.9	39	2.0567

Framework versions

• PEFT 0.10.0
• Transformers 4.40.2
• Pytorch 2.1.2+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1