marketplace-eval/configs/sample_simulation.yaml at main · kimdanny/marketplace-eval · 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
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
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
{
  "simulation": {
    "horizon": 1000,
    "total_num_users": 100, # total number of users to be generated/initiated before the marketplace simulation
    "users_per_step": 50,
    "seed": 42
  },
  "users": {
    "taxonomy_base": "datamorgana", # "datamorgana" or "tuna" (datamorgana is used for RAG evaluation, whereas tuna can be used for general usage evaluation)
    "user_data_type": "qa", # "q" (questions only) or "qa" (question-answer pairs); TUNA only supports "q".
    # Note: Whether to use documents to ground questions is now determined by the "needs_documents" field
    #   in each question type definition in the taxonomy config (humans/datamorgana_config.py or humans/tuna_config.py)
    # Path to user data file. If empty or file doesn't exist, data will be generated automatically
    "user_data_source": "/path/to/user_data.json",
    "profiles": []  # Leave profiles empty to use auto-generation, or provide custom profiles to override.
  },
  "judges": [
    {
      "id": "sample_judge",
      "type": "judge",
      "params": {
        "judge_prompt": "sample",
        "provider": "openrouter",
        "model_id": "openai/gpt-4.1",
        "generation_parameters": {
            "temperature": 0.0,
        },
      }
    }
  ],
  "graph": {
    "nodes": [
      {"id": "user_input", "type": "input"},
      {"id": "answer_output", "type": "output"},
      {
        "id": "generator_t5_base",
        "type": "generator",
        "params": {
          "name": "T5-Base",
          "model": {
            "provider": "vllm",
            "model_id": "google-t5/t5-base",
            "sampling_parameters": {
              "temperature": 0.7,
              "max_tokens": 512
            }
          },
          "system_prompt": ""
        }
      },
      {
        "id": "generator_qwen_2_5_7b_instruct",
        "type": "generator",
        "params": {
          "name": "Qwen-2.5-7b-Instruct",
          "model": {
            "provider": "openrouter",
            "model_id": "qwen/qwen-2.5-72b-instruct:free",
            "generation_parameters": {
              "temperature": 0.7,
            }
          },
          "system_prompt": "",
          "retrieval_strategy": {
            "type": "agentic",
            "max_rounds": 2,
            "router_top_k": null,
            "documents_per_retriever": 2,
            "max_context_documents": 4,
            "context_separator": " \n"
          }
        }
      },
      {
        "id": "generator_gpt_oss_20b",
        "type": "generator",
        "params": {
          "name": "GPT-OSS-20B",
          "model": {
            "provider": "openrouter",
            "model_id": "openai/gpt-oss-20b:free",
            "generation_parameters": {
              "temperature": 0.6,
              "top_p": 0.7,
            },
            "headers": {
              "HTTP-Referer": "https://example.com/simulation",
              "X-Title": "My Simulation"
            }
          },
          "system_prompt": "Answer user questions clearly and reference provided context when appropriate.",
          "retrieval_strategy": {
            "type": "naive",
            "router_top_k": 1,
            "documents_per_retriever": 1
          }
        }
      },
      {
        "id": "router_main",
        "type": "router",
        "params": {
          "exploration_prob": 0.2
        }
      },
      {
        "id": "retriever_e5_base",
        "type": "retriever",
        "params": {
          "model_id": "intfloat/e5-base-v2",
          "top_k": 3,
          "corpus_path": "user_data/fineweb_500.jsonl"
        }
      },
      {
        "id": "retriever_e5_small",
        "type": "retriever",
        "params": {
          "model_id": "intfloat/e5-small-v2",
          "top_k": 2,
          "corpus_path": "user_data/fineweb_500.jsonl"
        }
      },
      {
        "id": "retriever_mini",
        "type": "retriever",
        "params": {
          "model_id": "sentence-transformers/all-MiniLM-L12-v2",
          "top_k": 1,
          "corpus_path": "user_data/fineweb_500.jsonl"
        }
      },
      {
        "id": "retriever_google_web",
        "type": "retriever",
        "params": {
          "backend": "google_search",
          "safe": "active",
          "top_k": 3,
          "fetch_full_text": True,
          "max_text_length": 1000,
          "url_timeout": 5.0
        }
      }
    ],
    "edges": [
      {"source": "user_input", "target": "generator_t5_base"},
      {"source": "user_input", "target": "generator_qwen_2_5_7b_instruct"},
      {"source": "user_input", "target": "generator_gpt_oss_20b"},
      {"source": "generator_t5_base", "target": "router_main"},
      {"source": "generator_qwen_2_5_7b_instruct", "target": "router_main"},
      {"source": "generator_gpt_oss_20b", "target": "router_main"},
      {"source": "router_main", "target": "retriever_e5_base"},
      {"source": "router_main", "target": "retriever_e5_small"},
      {"source": "router_main", "target": "retriever_mini"},
      {"source": "generator_t5_base", "target": "answer_output"},
      {"source": "generator_qwen_2_5_7b_instruct", "target": "answer_output"},
      {"source": "generator_gpt_oss_20b", "target": "answer_output"}
    ]
  }
}