PocketWeights/PocketWeights-Qwen2.5-14B-Coder-Creative

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

PocketWeights-Qwen2.5-14B-Coder-Creative is a 14.7 billion parameter model from PocketWeights, built on the Qwen2.5 architecture with a 32768 token context length. It is engineered using the DARE-TIES merge algorithm to combine high-density coding logic with an uncensored, abliterated base. This model excels at top-tier code completion and logic reasoning without refusal, making it suitable for unconstrained security testing and technical narrative workflows.

Loading preview...

Model Overview

PocketWeights-Qwen2.5-14B-Coder-Creative is a 14.7 billion parameter model developed by PocketWeights, designed to bridge the gap between specialized coding models and general-purpose, uncensored language models. It addresses the common issue where coding specialists might refuse certain tasks, while general models lack the precision for complex code generation.

Key Capabilities

  • Domain-Fused Architecture: Utilizes the DARE-TIES merge algorithm to integrate the coding prowess of Qwen2.5-Coder-14B-Instruct with the unrestricted nature of Qwen2.5-14B-Instruct-1M-abliterated.
  • Optimal Performance-to-Compute Ratio: At 14.7B parameters, it offers strong performance while being efficient enough to run on hardware with 12GB–16GB VRAM.
  • High-Precision Coding: Retains state-of-the-art syntax and logic capabilities for languages like Python, C++, Rust, and shell scripts.
  • Refusal-Free Operation: Built on an abliterated foundation, it supports unconstrained security testing, scripting, and technical narrative tasks without moralizing safety friction.
  • DARE-TIES Optimization: Redundant parameters were pruned during synthesis (density 0.65) to maintain base model coherence and efficiency.

Ideal Use Cases

  • Code Generation & Completion: Excels in generating and completing code with high syntactic accuracy and logical reasoning.
  • Security Testing & Scripting: Suitable for tasks requiring unrestricted output, such as cybersecurity analysis, red-teaming, and complex scripting.
  • Technical Narrative & Content Creation: Supports technical writing and narrative generation without content restrictions.

Finetunes
2 models