mirror of
https://github.com/hanzoai/gym.git
synced 2026-09-30 22:37:59 +00:00
zeekayandHanzo Dev
680408a1ef
Verified on Metal: loss 0.155 → 0.052 over three steps, PEFT adapter written. Co-authored-by: Hanzo Dev <[email protected]>
21 lines
406 B
YAML
21 lines
406 B
YAML
base_model: Qwen/Qwen3-0.6B
|
|
chat_template: qwen3
|
|
datasets:
|
|
- path: examples/add.jsonl
|
|
type: chat_template
|
|
val_set_size: 4
|
|
output_dir: outputs/qwen3-lora
|
|
sequence_len: 64
|
|
adapter: lora
|
|
lora_r: 8
|
|
lora_alpha: 16
|
|
lora_target_modules: [q_proj, v_proj]
|
|
micro_batch_size: 4
|
|
gradient_accumulation_steps: 1
|
|
num_epochs: 1
|
|
max_steps: 20
|
|
learning_rate: 1.0e-4
|
|
lr_scheduler: constant
|
|
logging_steps: 1
|
|
bf16: false
|