ToT_Optimizer_for_LLMSum/config.yaml at main · Siki-cloud/ToT_Optimizer_for_LLMSum · GitHub

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
# config.yaml
project:
  name: "ToT-PromptOptimizer"
  version: "1.0.0"
  description: "Tree of Thoughts Prompt Optimization Framework - English Version"

models:
  target_model:
    path: "Qwen/Qwen3-1.7B"  # 目标LLM路径
    device: "cuda:2"
    tensor_parallel_size: 1
    dtype: "bfloat16"
    max_model_len: 4096

  evaluator_model:
    path: "Qwen/Qwen3-8B"  # 评估LLM路径
    device: "cuda:3"
    tensor_parallel_size: 1
    dtype: "bfloat16"
    max_model_len: 4096

optimization:
  max_iterations: 30
  beam_width: 3
  exploration_weight: 1.41
  max_depth: 5
  max_tokens: 200
  task_description: "Summarize this technical paper"  # 改为英文

evaluation:
  train_samples_per_eval: 10    # 训练评估样本数，-1 表示全部
  val_samples_per_eval: 2      # 验证评估样本数， -1 表示全部
  test_samples_per_eval: 5     # 测试评估样本数， -1 表示全部
  metrics:                     # 评估指标及权重
    faithfulness: 0.25
    conciseness: 0.20
    completeness: 0.25
    readability: 0.15
    insightfulness: 0.15

data:
  train_size: 50
  val_size: 15
  test_size: 10
  cache_dir: "./data"
  results_dir: "./results"
  components_file: "component_library.json"
  train_file: "train_papers.json"
  test_file: "test_papers.json"
  val_file: "val_papers.json"

logging:
  level: "INFO"
  file: "./results/logs/optimization.log"
  console: true