LLM Consortium

Inspiration

Based on Karpathy's observation:

"I find that recently I end up using all of the models and all the time. One aspect is the curiosity of who gets what, but the other is that for a lot of problems they have this 'NP Complete' nature to them, where coming up with a solution is significantly harder than verifying a candidate solution. So your best performance will come from just asking all the models, and then getting them to come to a consensus."

This plugin for the llm package implements a model consortium system with iterative refinement and response synthesis. It orchestrates multiple language models to collaboratively solve complex problems through structured dialogue, evaluation, and arbitration.

Core Algorithm Flow

flowchart TD
    A[Start] --> B[Get Model Responses]
    B --> C[Synthesize Responses]
    C --> D{Check Confidence}
    D -- Confidence ≥ Threshold --> E[Return Final Result]
    D -- Confidence < Threshold --> F{Max Iterations Reached?}
    F -- No --> G[Prepare Next Iteration]
    G --> B
    F -- Yes --> E

Loading

Features

• Multi-Model Orchestration: Coordinate responses from multiple models in parallel.
• Iterative Refinement: Automatically refine output until a confidence threshold is achieved.
• Advanced Arbitration: Uses a designated arbiter model to synthesize and evaluate responses.
• Database Logging: SQLite-backed logging of all interactions.
• Configurable Parameters: Adjustable confidence thresholds, iteration limits, and model selection.
• Flexible Model Instance Counts: Specify individual instance counts via the syntax model:count.
• Conversation Continuation: Continue previous conversations using the -c or --cid flags, just like with standard llm models. (New in v0.3.2)

New Model Instance Syntax

You can define different numbers of instances per model by appending :count to the model name. For example:

• "o3-mini:1" runs 1 instance of o3-mini.
• "gpt-4o:2" runs 2 instances of gpt-4o.
• "gemini-2:3" runs 3 instances of gemini-2. (If no count is specified, a default instance count (default: 1) is used.)

Installation

First, get llm:

Using uv:

uv tool install llm

Using pipx:

pipx install llm

Then install the consortium plugin:

llm install llm-consortium
# Or to install directly from this repo
# llm install -e .

Command Line Usage

The consortium command now defaults to the run subcommand for concise usage.

Basic usage:

llm consortium "What are the key considerations for AGI safety?"

Or, if you have saved a consortium model (e.g., named my-consortium):

llm -m my-consortium "What are the key considerations for AGI safety?"

This command will:

1. Send your prompt to multiple models in parallel (using the specified instance counts, if provided).
2. Gather responses along with analysis and confidence ratings.
3. Use an arbiter model to synthesize these responses.
4. Iterate to refine the answer until a specified confidence threshold or maximum iteration count is reached.

Conversation Continuation Usage (New in v0.3.2)

After running an initial prompt with a saved consortium model, you can continue the conversation:

To continue the most recent conversation:

# Initial prompt
llm -m my-consortium "Tell me about the planet Mars."
# Follow-up
llm -c "How long does it take to get there?"

To continue a specific conversation:

# Initial prompt (note the conversation ID, e.g., 01jscjy50ty4ycsypbq6h4ywhh)
llm -m my-consortium "Tell me about Jupiter."

# Follow-up using the ID
llm -c --cid 01jscjy50ty4ycsypbq6h4ywhh "What are its major moons?"

Other Options

• -m, --model: Model to include in the consortium for the run command. Use model:count for instance counts (default models used by run: claude-3-opus-20240229, claude-3-sonnet-20240229, gpt-4, gemini-pro).
• --arbiter: The arbiter model (default: claude-3-opus-20240229).
• --confidence-threshold: Minimum required confidence (default: 0.8).
• --max-iterations: Maximum rounds of iterations (default: 3).
• --min-iterations: Minimum iterations to perform (default: 1).
• --system: Custom system prompt.
• --output: Save detailed results to a JSON file.
• --stdin/--no-stdin: Append additional input from stdin (default: enabled).
• --raw: Output raw responses from both the arbiter and individual models (default: enabled).

Advanced example using the run command:

llm consortium "Your complex query" \
  -m o3-mini:1 \
  -m gpt-4o:2 \
  -m gemini-2:3 \
  --arbiter gemini-2 \
  --confidence-threshold 1 \
  --max-iterations 4 \
  --min-iterations 3 \
  --output results.json

Managing Consortium Configurations

You can save a consortium configuration as a model for reuse. This allows you to quickly recall a set of model parameters in subsequent queries.

Saving a Consortium as a Model

llm consortium save my-consortium \
    --model claude-3-opus-20240229 \
    --model gpt-4 \
    --arbiter claude-3-opus-20240229 \
    --confidence-threshold 0.9 \
    --max-iterations 5 \
    --min-iterations 1 \
    --system "Your custom system prompt"

Once saved, you can invoke your custom consortium like this:

llm -m my-consortium "What are the key considerations for AGI safety?"

And continue conversations using -c or --cid as shown above.

Programmatic Usage

Use the create_consortium helper to configure an orchestrator in your Python code. For example:

from llm_consortium import create_consortium

orchestrator = create_consortium(
    models=["o3-mini:1", "gpt-4o:2", "gemini-2:3"],
    confidence_threshold=1,
    max_iterations=4,
    min_iterations=3,
    arbiter="gemini-2",
    raw=True
)

result = orchestrator.orchestrate("Your prompt here")
print(f"Synthesized Response: {result['synthesis']['synthesis']}")

(Note: Programmatic conversation continuation requires manual handling of the conversation object or history.)

License

MIT License

Credits

Developed as part of the LLM ecosystem and inspired by Andrej Karpathy’s insights on collaborative model consensus.

Changelog

• v0.3.2 (Upcoming):
  • Added conversation continuation support using -c and --cid flags.
  • Improved error handling for arbiter response parsing (KeyError: confidence).
  • Updated tests for conversation handling.
• v0.3.1:
  • Introduced the model:count syntax for flexible model instance allocation.
  • Improved confidence calculation and logging.
  • Updated consortium configuration management.