Skip to content

PDO Strategy Not Working: Config Strategy Setting Ignored in Favor of Auto-Detection #60

Description

@culturalLife

What happened?

Description
When explicitly configuring strategy: "pdo" in the optimization config, the system ignores this setting and auto-detects/forces a different strategy based on the model name, preventing PDO from running.

How to reproduce

config.yaml
optimization:
strategy: "pdo"
total_rounds: 10
num_duels_per_round: 3
num_eval_examples_per_duel: 5
num_initial_instructions: 2
thompson_alpha: 2.0
num_top_prompts_to_combine: 2
gen_new_prompt_round_frequency: 2
num_threads: 5
verbose: true
max_errors: 5
max_concurrent_threads: 5

model:
name: "meta/llama-4-scout-17b-16e-instruct"
api_base: "https://integrate.api.nvidia.com/v1"
temperature: 0.5

Environment

Package: llama-prompt-ops / prompt-ops

Model: meta/llama-4-scout-17b-16e-instruct

Python Version: 3.10+

Installation: pip install from source

Error logs (if any)

Auto-detected LlamaStrategy for model: llama-4-scout-17b-16e-instruct
2026-01-05 13:13:07,428 - root - WARNING - Model does not appear to be a Llama model
Applying LlamaStrategy to optimize prompt

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions