mirror of
https://github.com/zenlm/zen-family.git
synced 2026-11-02 17:32:06 +00:00
28 lines
413 B
YAML
28 lines
413 B
YAML
base_model: Qwen/Qwen3-4B-Instruct-2507
|
|
|
|
sequence_len: 2048
|
|
sample_packing: true
|
|
load_in_4bit: true
|
|
|
|
datasets:
|
|
- path: zen_nano_instruct_training_data_fixed_3.jsonl
|
|
type: instruction
|
|
|
|
num_epochs: 3
|
|
|
|
adapter: qlora
|
|
lora_r: 32
|
|
lora_alpha: 64
|
|
lora_dropout: 0.05
|
|
|
|
|
|
output_dir: ./zen-nano-finetuned-axolotl
|
|
|
|
|
|
gradient_accumulation_steps: 2
|
|
micro_batch_size: 2
|
|
learning_rate: 0.0001
|
|
optimizer: paged_adamw_8bit
|
|
|
|
|