feat(integrations): add Axolotl plugin for Dreadnode Strikes logging

[SJCaldwell]

Summary

• Adds an Axolotl plugin that enables automatic logging of training metrics to Dreadnode Strikes
• Reuses the existing DreadnodeCallback from the transformers integration
• Supports configuration via Axolotl YAML: dreadnode_project, dreadnode_workspace, dreadnode_run_name, dreadnode_tags

Usage

In the config you supply to axolotl, for example test_dreadnode_config.yaml

plugins:
  - dreadnode.integrations.axolotl.DreadnodePlugin

dreadnode_project: "my-finetuning-project"
dreadnode_workspace: "my-workspace"        # optional
dreadnode_run_name: "llama-finetune-v1"    # optional
dreadnode_tags:                            # optional
  - "experiment-1"

Run with axolotl like axolotl train test_dreadnode_config.yaml

Test plan

• Plugin imports correctly
• Edge case testing (missing config, non-main process)
• Full training run with metrics logged to Strikes

🤖 Generated with Claude Code

Generated Summary:

• Introduced the DreadnodePlugin for integrating Dreadnode logging with Axolotl's training metrics.
• Created a new file __init__.py that contains the primary plugin implementation.
• Implemented essential methods for registration, model loading, and training callbacks:
  • register: Logs project registration.
  • pre_model_load: Validates project configuration before model loading.
  • add_callbacks_post_trainer: Injects DreadnodeCallback after trainer creation, essential for logging metrics.
• Added a new file args.py for Pydantic schema validation of the integration configuration.
• Included fields for:
  • dreadnode_project: Defines the Dreadnode project.
  • dreadnode_workspace: Specifies workspace configuration.
  • dreadnode_run_name: Name of the training run.
  • dreadnode_tags: Tags to associate with Strikes logging.
• Potential impact includes improved logging for Axolotl training processes and better tracking of experiments in Dreadnode Strikes.

This summary was generated with ❤️ by rigging