reallexi/lexi-coder-v5.1

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Aug 13, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

reallexi/lexi-coder-v5.1 is a 3.86 billion parameter causal language model developed by Reallexi LLC, derived from reallexi/lexi-coder-v4.3. This model is specifically fine-tuned for code generation and completion tasks, with its adapter merged directly into the base weights for streamlined runtime. It features a trained context length of 1,024 tokens and is optimized for efficient deployment across various precision levels.

Loading preview...

Model Overview

reallexi/lexi-coder-v5.1 is a 3.86 billion parameter language model developed by Reallexi LLC, specifically designed for code-related tasks. It is an evolution of the reallexi/lexi-coder-v4.3 base model, with its PEFT adapter fully merged into the base weights, eliminating the need for separate adapter loading at runtime.

Key Characteristics

  • Parameter Count: 3.86 billion parameters.
  • Memory Footprint: Weights on disk are 7.15 GB, with approximate memory requirements of 7.19 GB for FP16/BF16, 3.59 GB for 8-bit (Q8_0), and 1.98 GB for 4-bit (Q4_K_M).
  • Context Length: Trained with a context length of 1,024 tokens.
  • Training: Utilizes a LoRA strategy with a rank of 16 and alpha of 32, trained on the reallexi/lexi-coder-v3-datasest over 15,000 steps and 3 epochs.

Use Cases

This model is particularly well-suited for:

  • Code Completion: Demonstrates improved code completion capabilities compared to its base model, as shown in provided before/after samples.
  • Code Generation: Designed to assist in generating code snippets and functions.
  • Integration: Easy to integrate into projects using the Hugging Face transformers library, requiring only AutoModelForCausalLM and AutoTokenizer.

Finetunes
1 models