v2.0.1

This release adds JSONL tuning optimizations as well as additional configuration arguments.

Changes

• handle chat templates for tuning models with JSONL
• add --max-seq-length CLI argument
• update FAQ

v2.0.0

This release adds support for most major LLM types with the addition of the 'generic' llm-type as well as various minor optimizations.

Changes

• add support for most LLM types with the 'generic'(default value) LLM type
• minor code optimizations
• update title & description
• update documentation
• optimize LangChain agent vocabulary tokens

v1.6.0

This release adds CPU based tuning, along with instruction tuning optimizations and additional CLI arguments.

Changes

• add CPU based tuning
• add instruction model tuning option
• fix load JSONL training datasets
• add option to add additional vocabulary tokens

v1.5.2

This is a hotfix for additional vocabulary tokens.

Changes

• fix additional-vocabulary-tokens default argument value
• fix merge models with additional vocab tokens

v1.5.1

This release optimizes the process of fine-tuning models for use with LangChain ReAct agents.

Changes

• optimize LangChain ReAct Agent fine-tuning
• establish pattern to easily add support for other LLM types(modules)
• add ability to tune LLM models with custom vocabulary tokens

1.5.0

This release brings embedding tuning optimizations. This release BREAKS output directories.

Changes

• update docs
• optimize embedding training
• add addition tuner arguments
• refactor adapter/model directories

1.4.5

This is a hotfix for JSONL tuning.

Changes

• fix load JSONL training data

1.4.4

This change adds the ability to provide stop sequences to completions.

Changes

• add stop sequences to generation
• optimize agent tokens

1.4.3

This release fixes merging agent adapters

Changes

• fix merge LangChain agent adapter

1.4.2

Hot fix for broken LangChain agent tokens

Changes

• fix broken LangChain agent token feature
• update docs