Model Overview

The uukuguy/speechless-coding-7b-16k-tora is a 7 billion parameter model, fine-tuned from llm_agents/tora-code-7b-v1.0, designed to enhance reasoning, planning, and coding capabilities. It features an extended context window of 16,384 tokens and accepts the Alpaca instruction format for prompting.

Key Capabilities & Performance

This model excels in code generation and problem-solving, as evidenced by its benchmark results:

• HumanEval-Python: Achieves a score of 52.44, performing comparably to larger CodeLlama-34B variants.
• MultiPL-E: Demonstrates proficiency across various programming languages, including Python (55.96), JavaScript (46.93), and C++ (37.48).

Training Details

The model was fine-tuned on a diverse dataset of 177,333 samples, totaling 316 MB, specifically curated for coding, reasoning, and planning tasks. The training data includes filtered categories from:

• jondurbin/airoboros-2.2 (coding, reasoning, planning)
• Open-Orca/OpenOrca (COT category from GPT-4 data)
• garage-bAInd/Open-Platypus
• WizardLM/WizardLM_evol_instruct_V2_196k (coding conversation)
• TokenBender/python_eval_instruct_51k (Python-specific outputs)

Use Cases

This model is particularly well-suited for applications requiring:

• Code generation and completion
• Programming problem-solving
• Reasoning and planning in technical contexts
• Instruction-following for coding tasks