Upload README.md
Browse files
README.md
CHANGED
@@ -21,7 +21,28 @@ Instruction-tuning with [PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0](https://huggin
|
|
21 |
datasets: [kyujinpy/KOR-OpenOrca-Platypus-v3](https://huggingface.co/datasets/kyujinpy/KOR-OpenOrca-Platypus-v3).
|
22 |
|
23 |
**Hyperparameters**
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
# **Model Benchmark**
|
27 |
|
|
|
21 |
datasets: [kyujinpy/KOR-OpenOrca-Platypus-v3](https://huggingface.co/datasets/kyujinpy/KOR-OpenOrca-Platypus-v3).
|
22 |
|
23 |
**Hyperparameters**
|
24 |
+
```python
|
25 |
+
python finetune.py \
|
26 |
+
--base_model PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0 \
|
27 |
+
--data-path kyujinpy/KOR-OpenOrca-Platypus-v3 \
|
28 |
+
--output_dir ./SOLAR-tail-10.7B-instruct \
|
29 |
+
--batch_size 64 \
|
30 |
+
--micro_batch_size 1 \
|
31 |
+
--num_epochs 1 \
|
32 |
+
--learning_rate 3e-5 \
|
33 |
+
--cutoff_len 4096 \
|
34 |
+
--val_set_size 0 \
|
35 |
+
--lora_r 16 \
|
36 |
+
--lora_alpha 16 \
|
37 |
+
--lora_dropout 0.05 \
|
38 |
+
--lora_target_modules '[q_proj, k_proj, v_proj, o_proj, gate_proj, down_proj, up_proj, lm_head]' \
|
39 |
+
--train_on_inputs False \
|
40 |
+
--add_eos_token False \
|
41 |
+
--group_by_length False \
|
42 |
+
--prompt_template_name user_prompt \
|
43 |
+
--lr_scheduler 'cosine' \
|
44 |
+
```
|
45 |
+
> Platypus repo.
|
46 |
|
47 |
# **Model Benchmark**
|
48 |
|