Mipro Optimization

1
from opik_optimizer import MiproOptimizer
2
3
optimizer = MiproOptimizer(
4
    model="openai/gpt-4",  # or "azure/gpt-4"
5
    project_name="my-project",
6
    temperature=0.1,
7
    max_tokens=5000,
8
    num_threads=8,
9
    seed=42
10
)

1
optimizer = MiproOptimizer(
2
    model="openai/gpt-4",
3
    project_name="my-project",
4
    temperature=0.1,
5
    max_tokens=5000,
6
    num_threads=8,
7
    seed=42,
8
    num_agents=3,           # Number of optimization agents
9
    interaction_rounds=5,   # Number of interaction rounds
10
    exploration_rate=0.3,   # Exploration vs exploitation balance
11
    feedback_weight=0.7,    # Weight of feedback in optimization
12
    convergence_threshold=0.01  # Optimization convergence threshold
13
)

1
from opik_optimizer import MiproOptimizer
2
from opik.evaluation.metrics import LevenshteinRatio
3
from opik_optimizer import (
4
    MetricConfig,
5
    TaskConfig,
6
    from_llm_response_text,
7
    from_dataset_field,
8
)
9
from opik_optimizer.demo import get_or_create_dataset
10
11
# Initialize optimizer
12
optimizer = MiproOptimizer(
13
    model="openai/gpt-4",
14
    temperature=0.1,
15
    max_tokens=5000
16
)
17
18
# Prepare dataset
19
dataset = get_or_create_dataset("hotpot-300")
20
21
# Define metric and task configuration (see docs for more options)
22
metric_config = MetricConfig(
23
    metric=LevenshteinRatio(),
24
    inputs={
25
        "output": from_llm_response_text(),  # Model's output
26
        "reference": from_dataset_field(name="answer"),  # Ground truth
27
    }
28
)
29
task_config = TaskConfig(
30
    type="text_generation",
31
    instruction_prompt="Provide an answer to the question.",
32
    input_dataset_fields=["question"],
33
    output_dataset_field="answer"
34
)
35
36
# Run optimization
37
results = optimizer.optimize_prompt(
38
    dataset=dataset,
39
    num_trials=10,
40
    metric_config=metric_config,
41
    task_config=task_config
42
)
43
44
# Access results
45
print(f"Best prompt: {results.best_prompt}")
46
print(f"Improvement: {results.improvement_percentage}%")

1
optimizer = MiproOptimizer(
2
    model="anthropic/claude-3-opus",  # or any LiteLLM supported model
3
    project_name="my-project",
4
    temperature=0.1,
5
    max_tokens=5000
6
)