Upload fine_tuning_configuration.yml with huggingface_hub
Browse files
fine_tuning_configuration.yml
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
wandb_config:
|
| 2 |
+
wandb_project: "llm_dialog_summarizer_faster"
|
| 3 |
+
run_name: SmolLM2-360M-Instruct-large-R
|
| 4 |
+
|
| 5 |
+
model_config:
|
| 6 |
+
model_id: "HuggingFaceTB/SmolLM2-360M-Instruct"
|
| 7 |
+
load_in_4bit: False
|
| 8 |
+
max_seq_length: 8192
|
| 9 |
+
|
| 10 |
+
lora_config:
|
| 11 |
+
r: 64
|
| 12 |
+
lora_alpha: 64
|
| 13 |
+
use_rslora: True
|
| 14 |
+
|
| 15 |
+
sft_config:
|
| 16 |
+
learning_rate: 0.0003
|
| 17 |
+
epochs: 2
|
| 18 |
+
optimizer: "adamw_8bit"
|
| 19 |
+
warmup_steps: 100
|
| 20 |
+
weight_decay: 0.01
|
| 21 |
+
lr_scheduler_type: "linear"
|
| 22 |
+
seed: 90201
|
| 23 |
+
dataset_text_field: "text"
|