trollek/danube2-1.8b-Synthia-v1.3

h2o-danube2 with ChatML template

This model was first fine-tuned with BAdam on migtissera/Synthia-v1.3 using LLama-Factory.

Quants

Big ups, mradermacher!

• Weighted - mradermacher/danube2-1.8b-Synthia-v1.3-i1-GGUF
• Static - mradermacher/danube2-1.8b-Synthia-v1.3-GGUF

Template

<|im_start|>system
{{system}}<|im_end|>
<|im_start|>user
{{instruction}}<|im_end|>
<|im_start|>assistant
{{response}}<|im_end|>

BAdam config

### model
model_name_or_path: danube2-base-chatml

### method
stage: sft
do_train: true
finetuning_type: full
use_badam: true
badam_switch_mode: ascending
badam_switch_interval: 50
badam_verbose: 1
badam_start_block: 16
seed: 1080

### dataset
dataset: synthia
template: hermes_chatml
cutoff_len: 8192
overwrite_cache: false
preprocessing_num_workers: 12

### output
output_dir: synthia-chatml-badam
logging_steps: 5
save_steps: 1
save_strategy: epoch
plot_loss: true
overwrite_output_dir: false

### train
per_device_train_batch_size: 2
gradient_accumulation_steps: 8
learning_rate: 0.00001
num_train_epochs: 1
lr_scheduler_type: constant_with_warmup
warmup_ratio: 0.01
bf16: true
flash_attn: fa2

### eval
val_size: 0.01
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 1000

BAdam training results

Training Loss	Epoch	Step	Validation Loss
0.7895	0.1360	1000	0.7205
0.7153	0.2720	2000	0.7015
0.7024	0.4080	3000	0.7024
0.752	0.5440	4000	0.6977
0.727	0.6800	5000	0.6922
0.6616	0.8160	6000	0.6922
0.7628	0.9519	7000	0.6913