Xangel0s/ozyvler-Ozygram-Neural-7B-Coder-MCTS

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 25, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Ozygram Neural 7B is a 7.6 billion parameter neuro-symbolic code intelligence model developed by Xangel0s, built upon Qwen2.5-Coder-7B-Instruct. It features a dual-phase architecture with Dream-RSI (Monte Carlo Tree Search for reasoning) and Kev Action Engine (structured JSON for AST mutations). This model specializes in local, autonomous code self-healing, bug localization, and transactional repair, optimized for edge workstations with 8GB VRAM.

Loading preview...

Ozygram Neural 7B: Autonomous Code Self-Healing

Ozygram Neural 7B is a specialized, open-weights code intelligence model designed for local self-healing codebases, automated bug localization, and high-concurrency transactional repair. Built on the Qwen2.5-Coder-7B-Instruct base, it integrates a unique dual-phase neuro-symbolic reasoning architecture.

Key Capabilities

  • Dream-RSI (Reasoning via Self-Improvement / MCTS): An internal reasoning chain that simulates AST execution graphs, evaluates blast radius, and detects runtime pitfalls (e.g., deadlocks, zero-division, memory leaks) before generating code. This MCTS-guided approach enhances predictive debugging.
  • Kev Action Engine: A strict, structured JSON output protocol for atomic AST mutations (ast_mutation, apply_patch). This ensures safe and precise application of patches by IDEs and autonomous agents, avoiding common LLM prose hallucinations.
  • Autonomous Bug Localization & Repair: Proven in real-world stress benchmarks to diagnose and fix complex issues like zero-division errors, worker concurrency deadlocks, and corrupted parsing expressions with high accuracy.
  • Optimized for Edge Hardware: Tuned to run efficiently on single-GPU developer laptops and mobile workstations, requiring as little as ~5.2 GB VRAM for Q4_K_M quantization.

Good For

  • Developers and teams requiring local, autonomous code repair capabilities.
  • Integrating self-healing mechanisms into IDEs or CI/CD pipelines.
  • Use cases demanding deterministic, structured code modifications without prose-based ambiguity.
  • Environments with limited GPU resources (e.g., 8GB VRAM workstations) needing advanced code intelligence.

Finetunes
1 models