NavitraTechnologies01/tn_01_base_coder

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 13, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Alpha-Coder-14B by Jay2003Bhatt is a 14 billion parameter language model, fine-tuned via LoRA from Qwen2.5-Coder-14B-Instruct, specifically optimized for generating typed and tested Python code. This model demonstrates improved MMLU performance and maintained GSM8K scores post-fine-tuning, making it suitable for code generation tasks. It is available in fp16 safetensors and a Q4_K_M GGUF quantization for local inference.

Loading preview...

Alpha-Coder-14B: Python Code Generation Specialist

Alpha-Coder-14B is a specialized language model developed by Jay2003Bhatt, fine-tuned from the Qwen/Qwen2.5-Coder-14B-Instruct base model. Its primary focus is on generating high-quality, typed, and tested Python code. The fine-tuning process utilized LoRA (Low-Rank Adaptation) on Apple Silicon hardware, resulting in a model that retains the strong capabilities of its base while enhancing specific coding performance.

Key Capabilities

  • Python Code Generation: Optimized to produce functional, typed, and tested Python code.
  • Performance Improvement: Achieved a 5-point improvement on the MMLU benchmark (from 72% to 77%) compared to its base model.
  • Knowledge Retention: Maintained strong performance on the GSM8K benchmark (~93%), indicating no measurable forgetting of general reasoning abilities during fine-tuning.
  • Flexible Deployment: Provided as full-precision fp16 safetensors for transformers and a Q4_K_M GGUF quantization (~8.4GB) for efficient local inference with tools like llama.cpp and Ollama.

Good for

  • Coding Assistance: Ideal for developers seeking an AI assistant to generate Python code snippets or functions.
  • Local Development: The availability of a quantized GGUF version makes it suitable for running on consumer-grade hardware.
  • Educational Purposes: Can be used to explore code generation and fine-tuning techniques on a specialized model.

Users should always evaluate the generated code for correctness and security, especially in production environments.