---
license: other
license_name: "inherits-base-model-and-dataset-terms"
base_model: "reallexi/lexi-coder-v4.3"
library_name: transformers
pipeline_tag: "text-generation"
tags:
- "ai-model-builder"
- "fine-tuned"
- lora
- reallexi
- "text-generation"
---

# lexi-coder-v5.1

**lexi-coder-v5.1** by Reallexi LLC AI Model Builder — [llm.reallexi.io](https://llm.reallexi.io)

Copyright (c) 2026 Reallexi LLC. All rights reserved.

A standalone model of 3.86B parameters, derived from [`reallexi/lexi-coder-v4.3`](https://huggingface.co/reallexi/lexi-coder-v4.3).

The adapter has been merged into the base weights, so no PEFT adapter is needed at runtime.


## Size and requirements

| | |
|---|---|
| Parameters | 3,859,090,432 (3.86B) |
| Weights on disk | 7.15 GB |
| Trained context length | 1,024 tokens |
| Base model | `reallexi/lexi-coder-v4.3` |


Approximate memory to hold the weights. Add context and runtime overhead on top.

| Precision | Weights |
|---|---|
| FP16 / BF16 | 7.19 GB |
| 8-bit (Q8_0) | 3.59 GB |
| 4-bit (Q4_K_M) | 1.98 GB |


## Training

| | |
|---|---|
| Strategy | lora |
| Adapter | Auto LoRA |
| LoRA rank / alpha | 16 / 32 |
| Dataset | `reallexi/lexi-coder-v3-datasest` |
| Samples learned | 110,000 (through phase 28 of 258) |
| Training steps | 15,000 |
| Epochs | 3 |


## Before / after samples

The same prompts, drawn from the training data, run through the base model before training and the finished model after. This shows what the run changed on representative prompts -- it is not a benchmark. Full outputs are in `SAMPLES.md` and `samples.json` alongside this file.

**Prompt:** def partition(p, r): i = p for j in range(p, r): if A[r][1] >= A[j][1]: A[i], A[j] = A[j], A[i]

- Before: i += 1 return partition(p, r) if partition(p, r) == partition(p, r): return partition(p, r) else: return partition(p, r)

n = int(input())
A = [tuple(map(int, input().split())) for _ in range(n)]
print(*partition(0, n)[1:]) for i in range(n) if partition(0,
- After: i += 1 return i
def quicksort(p, r): if p < r: i = partition(p, r) quicksort(p, i-1) quicksort(i, r)
def mergeSort(p, r): if p < r: q = partition(p, r) mergeSort(p, q) mergeSort(q, r)
def printList(): print(" ", end =

**Prompt:** {i : Node(None, None, None) for i in range(n)} # 情報を入れるdictをNodeクラスで作成 for _ in range(n): # 変数名を「 _ 」にすることによって、「その変数を使っていない」ことを表現している（Pythonの習慣） tmp

- Before: = input().split() # 変数名と値を入れる tmp[0] = tmp[0].strip() # 変数名を入れる tmp[1] = tmp[1].strip() # 値を入れる tmp[2] = tmp[2].strip() # 値を入れる tmp[3] = tmp[3].strip() #
- After: = input().split() # 入れられた値を変数に入れる tmp[0] = int(tmp[0]) # 変数の値をintに変換 tmp[1] = int(tmp[1]) # 変数の値をintに変換 tmp[2] = int(tmp[2]) # 変数の値をintに変

**Prompt:** rmat(node_id), end = '') _pre_walk(self.nodes[node_id].left_child) _pre_walk(self.nodes[node_id].right_child) _pre_walk(self.root_id) print('') def inorder_walk…

- Before: _in_walk(self.nodes[node_id].left_child) print(str(self.nodes[node_id].value)) _in_walk(self.nodes[node_id].right_child) def _post_walk(self, node_id, end = ''): if node_id != -1: _post_walk(self.nodes[node_id].left_child) print(str(self.nodes[node_id].value)) _post_walk(self.nodes[node_id].right
- After: print(' ', end = '') print(node_id, end = '') _in_walk(self.nodes[node_id].left_child) _in_walk(self.nodes[node_id].right_child) _in_walk(self.nodes[node_id].parent) def post_walk(self): print('Post') root_node = self.nodes[self.root_id] def _post_walk(node_id): if node_id != -1: _post_walk


## Training curve

![Training loss](training_curve.png)


## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("lexi-coder-v5.1")
tokenizer = AutoTokenizer.from_pretrained("lexi-coder-v5.1")
```


## License and attribution

The effective terms are inherited from the base model and the training data, which are not necessarily the same as this project's own license. Review both before redistributing.

- Base model: [`reallexi/lexi-coder-v4.3`](https://huggingface.co/reallexi/lexi-coder-v4.3)

- Training data: `reallexi/lexi-coder-v3-datasest`


Copyright (c) 2026 Reallexi LLC. All rights reserved.


Produced by Reallexi LLC AI Model Builder from training job #161.
Core: https://llm.reallexi.io

## Who, where, and what platform trained this?

Produced by Reallexi LLC on [Reallexi AI Model Builder](https://llm.reallexi.io), a local-first training platform (https://llm.reallexi.io). Hugging Face repository: [reallexi/lexi-coder-v5.1](https://huggingface.co/reallexi/lexi-coder-v5.1).
Copyright (c) 2026 Reallexi LLC. All rights reserved.