Upload new file
Browse files- config.yaml +91 -0
config.yaml
ADDED
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
wandb:
|
2 |
+
log: True
|
3 |
+
entity: "weblab-geniac-leaderboard"
|
4 |
+
project: "leaderboard"
|
5 |
+
run_name: 'weblab-geniac2/exp_012_dMoE_8B_finetune_all_v6_epoch2_v0.1' # 変更可:weblab-geniac7/exp_{num}_{model_name}にします
|
6 |
+
|
7 |
+
github_version: v2.0.0 #for recording
|
8 |
+
|
9 |
+
testmode: false
|
10 |
+
|
11 |
+
# if you don't use api, please set "api" as "false"
|
12 |
+
# if you use api, please select from "openai", "anthoropic", "google", "cohere"
|
13 |
+
api: false
|
14 |
+
|
15 |
+
model:
|
16 |
+
use_wandb_artifacts: false
|
17 |
+
artifacts_path: ""
|
18 |
+
pretrained_model_name_or_path: "geniacllm/dMoE_8B_finetune_all_v6_epoch2_v0.1" # 変更可
|
19 |
+
trust_remote_code: true
|
20 |
+
device_map: "auto"
|
21 |
+
load_in_8bit: false
|
22 |
+
load_in_4bit: false
|
23 |
+
|
24 |
+
generator:
|
25 |
+
do_sample: false
|
26 |
+
num_beams: 1 # https://huggingface.co/docs/transformers/v4.40.2/en/main_classes/text_generation
|
27 |
+
top_p: 1.0
|
28 |
+
top_k: 0
|
29 |
+
temperature: 0.1
|
30 |
+
repetition_penalty: 1.0
|
31 |
+
|
32 |
+
tokenizer:
|
33 |
+
use_wandb_artifacts: false
|
34 |
+
artifacts_path: ""
|
35 |
+
pretrained_model_name_or_path: "geniacllm/dMoE_8B_finetune_all_v6_epoch2_v0.1" # 変更可
|
36 |
+
use_fast: false
|
37 |
+
|
38 |
+
# for llm-jp-eval
|
39 |
+
max_seq_length: 2048 # 変更可
|
40 |
+
dataset_artifact: "wandb-japan/llm-leaderboard/jaster:v11" #if you use artifacts, please fill here (if not, fill null)
|
41 |
+
dataset_dir: "/jaster/1.2.6/evaluation/test"
|
42 |
+
target_dataset: "all" # {all, jamp, janli, jcommonsenseqa, jemhopqa, jnli, jsem, jsick, jsquad, jsts, niilc, chabsa}
|
43 |
+
log_dir: "./logs"
|
44 |
+
torch_dtype: "bf16" # {fp16, bf16, fp32}
|
45 |
+
custom_prompt_template: "<s>以下は、タスクを説明する指示と、文脈のある入力の組み合わせです。要求を適切に満たす応答を書きなさい。\n\n### 指示:\n{instruction}\n\n### 入力:\n{input}\n\n### 応答:\n" # 変更可
|
46 |
+
|
47 |
+
custom_fewshots_template: null
|
48 |
+
# Please include {input} and {output} as variables
|
49 |
+
# example of fewshots template
|
50 |
+
# "\n### 入力:\n{input}\n### 回答:\n{output}"
|
51 |
+
|
52 |
+
metainfo:
|
53 |
+
basemodel_name: "geniacllm/dMoE_8B_finetune_all_v6_epoch2_v0.1" # 変更可
|
54 |
+
model_type: "open llm" # {open llm, commercial api}
|
55 |
+
instruction_tuning_method: "None" # {"None", "Full", "LoRA", ...}
|
56 |
+
instruction_tuning_data: ["None"] # {"None", "jaster", "dolly_ja", "oasst_ja", ...}
|
57 |
+
num_few_shots: 0
|
58 |
+
llm-jp-eval-version: "1.1.0"
|
59 |
+
|
60 |
+
# for mtbench
|
61 |
+
mtbench:
|
62 |
+
question_artifacts_path: "wandb-japan/llm-leaderboard/mtbench_ja_question:v0" # if testmode is true, small dataset will be used
|
63 |
+
referenceanswer_artifacts_path: "wandb-japan/llm-leaderboard/mtbench_ja_referenceanswer:v0" # if testmode is true, small dataset will be used
|
64 |
+
judge_prompt_artifacts_path: "wandb-japan/llm-leaderboard/mtbench_ja_prompt:v1"
|
65 |
+
bench_name: "japanese_mt_bench"
|
66 |
+
model_id: null # cannot use '<', '>', ':', '"', '/', '\\', '|', '?', '*', '.'
|
67 |
+
question_begin: null
|
68 |
+
question_end: null
|
69 |
+
max_new_token: 1024 # 変更可
|
70 |
+
num_choices: 1
|
71 |
+
num_gpus_per_model: 1 # 変更可
|
72 |
+
num_gpus_total: 1 # 変更可
|
73 |
+
max_gpu_memory: null
|
74 |
+
dtype: bfloat16 # None or float32 or float16 or bfloat16
|
75 |
+
# for gen_judgment
|
76 |
+
judge_model: "gpt-4"
|
77 |
+
mode: "single"
|
78 |
+
baseline_model: null
|
79 |
+
parallel: 1
|
80 |
+
first_n: null
|
81 |
+
# for conv template # added
|
82 |
+
custom_conv_template: true # 変更可
|
83 |
+
# the following variables will be used when custom_conv_template is set as true
|
84 |
+
conv_name: "custom"
|
85 |
+
conv_system_message: "<s>以下は、タスクを説明する指示と、文脈のある入力の組み合わせです。要求を適切に満たす応答を書きなさい。" # 変更可
|
86 |
+
conv_roles: "('指示', '応答')" # 変更可
|
87 |
+
conv_sep: "\n\n### " # 変更可
|
88 |
+
conv_stop_token_ids: "[2]"
|
89 |
+
conv_stop_str: "###"
|
90 |
+
conv_role_message_separator: ":\n" # 変更可(llm-jp-evalに合わせて改行前のスペースは削除した)
|
91 |
+
conv_role_only_separator: ":\n" # 変更可(llm-jp-evalに合わせて改行前のスペースは削除した)
|