-
Notifications
You must be signed in to change notification settings - Fork 1
/
test_llama_vllm.sh
116 lines (103 loc) · 3.73 KB
/
test_llama_vllm.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
#!/bin/bash
#SBATCH --job-name=test_llama
#SBATCH --gres=gpu:a6000:1
#SBATCH --time=24:00:00
#SBATCH --output=../../jobs/test_llama/%j.out
nvidia-smi
## Note
# please download the data in the working directory as indicated in the Data Preparation section in the read me
# quick command: gdown https://drive.google.com/uc?id=1DAjvig-A_57CuBvENLg8A2PycOaz9ZkT
##
model_name="meta-llama/Llama-2-7b-hf"
outputs_dir=""
# outputs_dir="../../outputs"
checkpoint_name="ref"
# checkpoint_path="${outputs_dir}/${model_name}/${checkpoint_name}/checkpoint-532"
checkpoint_path="TIGER-Lab/TIGERScore-13B"
human_score_names="gpt_rank_score"
data_path="../../data/evaluation/lfqa/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task "long-form QA" \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
task="instruction-following"
human_score_names="gpt_rank_score"
data_path="../../data/evaluation/instruct/just-eval-instruct/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
task="mathQA"
human_score_names="accuracy"
data_path="../../data/evaluation/mathqa/gsm8k/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
# mtme test mqm
task="translation"
human_score_names="mqm"
data_path="../../data/evaluation/translation/wmt22/zh-en/eval_data.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
# sum test relevance
task="summarization"
human_score_names="coherence,consistency,fluency,relevance"
data_path="../../data/evaluation/summarization/summeval/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
# d2t test Correctness
task="data2text"
human_score_names="Correctness,DataCoverage,Fluency,Relevance,TextStructure"
data_path="../../data/evaluation/d2t/webnlg_2020/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite
# storygen test human
task="storygen"
human_score_names="human"
data_path="../../data/evaluation/storygen/test_data_prepared.json"
output_path="${data_path}.llama_2_7b_${checkpoint_name}_test.output"
python test_llama_vllm.py \
--model_name_or_path ${checkpoint_path} \
--task ${task} \
--data_path ${data_path} \
--output_path ${output_path} \
--batch_size 60 \
--human_score_names ${human_score_names} \
--overwrite