v25.2.1

25.2.1 - 2025-02-26

Added

• Config: prompt_template_file option added. This allows defining a prompt template from a file. If relative path, first the CWD is checked, then the path relative to the config.json.
• Config: prompt_template option added. This is a unified prompt template for both training and inference. Replaces training_prompt and inference_prompt.
• APP: chat_evaluate_extract method added to chathpc to evaluate and extract the answer portion in one call.
• APP: save_readme function to save a starter readme for models.
• Train: Now saves a template readme in the output folders.
• CLI app: log_level is now a command line argument.
• Interactive: Added optional ability to extract answer from response with --extract.
• Interactive: Gracefully handle EOF.
• Interactive & Template: Make context optional.
• Interactive: Allow context to be added inline with /context <context>.
• Interactive: A blank context, unsets the context.
• Ollama: Now tests both generate and chat API. It is recommended to use the chat API to be more compatible with OpenAI's api.
• CLI & Method: Verify subcommand to verify the model against the training dataset.
• Examples: Added Marimo example using ChatHPC.
• Full Test: Top level scripts/full_test.sh to run all the tests in order.
• Data Explore: Data explore script scripts/tests/data_expore.py to print out an inspect the training dataset.
• Added version to package for easy query of ChatHPC version.

Changed

• Template: Switch from format string to Jinja for the templates. The main change required here, is to use {{}} for variable names instead of {}.
• Template: Now only one template is used for both training and inference.
• Template: Can either be set by a file path or a string.
• datastore: Removed dependency on datastore, by coping datastore into utilities.
• verify_app.py: Refactored to remove duplicate code.

Removed

• Removed training_prompt. Replaced by prompt_template.
• Removed inference_prompt. Replaced by prompt_template.