Google Gen AI

1
from google import GenAI
2
import os
3
4
client = GenAI.Client(
5
    api_key=os.environ.get("RESPAN_API_KEY"),
6
    http_options={
7
        "base_url": "https://api.respan.ai/api/google/gemini",
8
    }
9
)

1
response = client.models.generate_content(
2
    model="gemini-2.5-flash",
3
    contents="Hello, world!",
4
)
5
6
print(response.text)

1
response = client.models.generate_content(
2
    model="gemini-2.0-flash-exp",
3
    contents="Tell me a joke.",
4
)

1
from google.GenAI import types
2
3
config = types.GenerateContentConfig(
4
    temperature=0.9,
5
    top_k=1,
6
    top_p=1,
7
    max_output_tokens=2048,
8
)
9
10
response = client.models.generate_content(
11
    model="gemini-2.5-flash",
12
    contents="What is the capital of France?",
13
    config=config,
14
)

1
from google import GenAI
2
from google.GenAI import types
3
import os
4
5
client = GenAI.Client(
6
    api_key=os.environ.get("RESPAN_API_KEY"),
7
    http_options={
8
        "base_url": "https://api.respan.ai/api/google/gemini",
9
    }
10
)
11
12
# Example: Configure tools for grounding
13
grounding_tool = types.Tool(
14
    google_search=types.GoogleSearch()
15
)
16
17
# Example: Comprehensive GenerateContentConfig showcasing various parameters
18
config = types.GenerateContentConfig(
19
    # System instruction to guide the model's behavior
20
    system_instruction="You are a helpful assistant that provides accurate, concise information about sports events.",
21
    
22
    # Sampling parameters
23
    temperature=0.7,  # Controls randomness (0.0-1.0). Lower = more focused, Higher = more creative
24
    top_p=0.95,  # Nucleus sampling. Tokens with cumulative probability up to this value are considered
25
    top_k=40,  # Top-k sampling. Considers this many top tokens at each step
26
    
27
    # Output controls
28
    max_output_tokens=1024,  # Maximum number of tokens in the response
29
    stop_sequences=["\n\n\n"],  # Sequences that will stop generation
30
    
31
    # Tools and function calling
32
    tools=[grounding_tool],  # Enable Google Search grounding
33
    
34
    # Thinking configuration (for models that support it)
35
    thinking_config=types.ThinkingConfig(thinking_budget=0),  # Disables thinking mode
36
    
37
    # Response format options
38
    # response_mime_type="application/json",  # Uncomment for JSON output
39
    # response_schema=types.Schema(  # Uncomment to enforce structured output
40
    #     type=types.Type.OBJECT,
41
    #     properties={
42
    #         "winner": types.Schema(type=types.Type.STRING),
43
    #         "year": types.Schema(type=types.Type.INTEGER)
44
    #     }
45
    # ),
46
    
47
    # Safety settings
48
    safety_settings=[
49
        types.SafetySetting(
50
            category=types.HarmCategory.HARM_CATEGORY_HATE_SPEECH,
51
            threshold=types.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE
52
        ),
53
        types.SafetySetting(
54
            category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
55
            threshold=types.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE
56
        )
57
    ],
58
    
59
    # Diversity controls
60
    presence_penalty=0.0,  # Penalize tokens based on presence in text (-2.0 to 2.0)
61
    frequency_penalty=0.0,  # Penalize tokens based on frequency in text (-2.0 to 2.0)
62
    
63
    # Reproducibility
64
    # seed=42,  # Uncomment to make responses more deterministic
65
    
66
    # Logprobs (for token analysis)
67
    # response_logprobs=True,  # Uncomment to get log probabilities
68
    # logprobs=5,  # Number of top candidate tokens to return logprobs for
69
)
70
71
response = client.models.generate_content(
72
    model="gemini-2.5-flash",
73
    contents="Who won the euro 2024?",
74
    config=config,
75
)
76
77
print(response.text)