See axolotl config

axolotl version: 0.4.1

adapter: qlora
base_model: DewEfresh/neo_7b-slerp
bf16: auto
chat_template: chatml
dataset_prepared_path: null
datasets:
- conversation: chatml
  path: https://huggingface.co/datasets/cognitivecomputations/dolphin-2.9.3/resolve/main/openhermes200k_unfiltered.jsonl
  type: sharegpt
debug: null
deepspeed: null
early_stopping_patience: null
eval_sample_packing: false
evals_per_epoch: 3
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: DewEfresh/neo-7b-slerp-hermes2
learning_rate: 0.0002
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 16
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: 8
num_epochs: 3
optimizer: adamw_bnb_8bit
output_dir: ./outputs/qlora-out
pad_to_sequence_len: true
resume_from_checkpoint: null
sample_packing: true
saves_per_epoch: 1
sequence_len: 4096
special_tokens: null
strict: false
tf32: false
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_log_model: null
wandb_name: null
wandb_project: neo-7b-slerp-hermes2
wandb_watch: null
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

neo-7b-slerp-hermes2

This model is a fine-tuned version of DewEfresh/neo_7b-slerp on the None dataset. It achieves the following results on the evaluation set:

• Loss: 5.5603

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 1024
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
15.1348	0.0521	1	14.5242
9.6676	0.3648	7	8.0894
7.076	0.7296	14	6.8289
6.6836	1.0717	21	6.4673
6.3156	1.4365	28	6.1033
6.0471	1.8013	35	5.8471
5.834	2.1433	42	5.6670
5.7349	2.5081	49	5.5762
5.7014	2.8730	56	5.5603

Framework versions

• PEFT 0.11.1
• Transformers 4.41.1
• Pytorch 2.1.2+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1