poonehmousavi
commited on
Commit
•
88b58c6
1
Parent(s):
175ea6f
Create hyperparams.yaml
Browse files- hyperparams.yaml +44 -0
hyperparams.yaml
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# ################################
|
2 |
+
# Model: Llama2 dModel + NLL
|
3 |
+
# Authors:
|
4 |
+
# Pooneh Mousavi 2023
|
5 |
+
# ################################
|
6 |
+
|
7 |
+
|
8 |
+
# URL for the gpt2 model
|
9 |
+
model_hub: meta-llama/Llama-2-7b-chat-hf
|
10 |
+
llama2_folder: recipes/MultiWOZ/response_generation/llama2/results/train_with_llama2/1995/save/llama2_checkpoint/
|
11 |
+
|
12 |
+
|
13 |
+
# history_window, i.e. how many user-system exchanges consider as context.
|
14 |
+
max_history: 2
|
15 |
+
|
16 |
+
# decoder setting
|
17 |
+
freeze_model: True
|
18 |
+
num_beams: 8
|
19 |
+
max_new_tokens: 50
|
20 |
+
top_k: 45
|
21 |
+
top_p: 0.9
|
22 |
+
|
23 |
+
#LLAMA2 model
|
24 |
+
model: !new:custom.LLAMA2_expanded
|
25 |
+
source: !ref <model_hub>
|
26 |
+
freeze: !ref <freeze_model>
|
27 |
+
save_path: !ref <llama2_folder>
|
28 |
+
max_new_tokens: !ref <max_new_tokens>
|
29 |
+
num_beams: !ref <num_beams>
|
30 |
+
top_k: !ref <top_k>
|
31 |
+
top_p: !ref <top_p>
|
32 |
+
with_peft: True
|
33 |
+
|
34 |
+
|
35 |
+
# Masks
|
36 |
+
padding_mask: !name:speechbrain.lobes.models.transformer.Transformer.get_key_padding_mask
|
37 |
+
|
38 |
+
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
39 |
+
loadables:
|
40 |
+
model: !ref <model>
|
41 |
+
|
42 |
+
modules:
|
43 |
+
model: !ref <model>
|
44 |
+
|