Modelpublic

reallexi/lexi-coder-v4.4

sourceHugging Faceotherupdated 17d agoView on Hugging Face

2likes1.2kdownloads

Model Card

lexi-coder-v4.4

lexi-coder-v4.4 by Reallexi LLC AI Model Builder — llm.reallexi.io

A standalone model of 3.85B parameters, derived from `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	3,847,556,096 (3.85B)
Weights on disk	7.15 GB
Trained context length	2,048 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	7.17 GB
8-bit (Q8_0)	3.58 GB
4-bit (Q4KM)	1.97 GB

Training


Strategy	lora
Adapter	Auto LoRA
LoRA rank / alpha	8 / 16
Dataset	`reallexi/lexi-coder-v3-datasest`
Samples learned	395,000 (through phase 53 of 174)
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: import numpy as np N, D = map(int, input().split()) cnt = 0 for i in range(N): x, y = map(int,

—Before: input().split()) if x2 + y2 <= D**2: cnt += 1 print(cnt) # print(N, D) # print(N, D, cnt) # print(N, D, cnt, cnt) # print(N, D, cnt, cnt, cnt) # print(N, D, cnt, cnt, cnt, cnt)
—After: input().split()) if x2 + y2 <= D2: cnt += 1 print(cnt) # print(x, y) # print(np.sqrt(x2 + y2)) # print(np.sqrt(D2)) # print(np.sqrt(D2 - x2)) # print(np.sqrt(D2 - y2))

Prompt: n, d = map(int, input().split()) ans = 0 for _ in range(n): x, y = map(float, input().split()) z = (x**2

—Before: + y2)0.5 if z <= d else -1 ans += 1 print(ans) # ans = 0 for _ in range(n): x, y = map(int, input().split()) z = (x2 + y2)**0.5 if z <= d else -1 ans += 1 print(ans) # ans = 0 for
—After: + y2)0.5 if z <= d else -1 ans += 1 print(ans) # ans = 0 for _ in range(n): x, y = map(int, input().split()) z = (x2 + y2)**0.5 if z <= d else -1 ans += 1 print(ans) # ans = 0 for

Prompt: #n, m = map(int,input().split()) k = int(input()) #hl = list(map(int,input().split())) s = len(str(k)) for i in range(1,k+1): if int('7'*i) %

—Before: k == 0: print(i) break k = int(input()) n = 7 for i in range(1,k+1): if n % k == 0: print(i) break n = n*10+7 else: print(-1) #k % 7 == 0の時は-1を返す
—After: k == 0: print(i) break k = int(input()) if k % 2 == 0 or k % 5 == 0: print(-1) else: i = 1 while True: if int('7'*i) % k == 0: print(i) break i += 1 if i >

Training curve

[image]

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("lexi-coder-v4.4")
tokenizer = AutoTokenizer.from_pretrained("lexi-coder-v4.4")

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`

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

Produced by Reallexi LLC AI Model Builder from training job #1996. Core: https://llm.reallexi.io

Who, where, and what platform trained this?

Produced by Reallexi LLC on Reallexi AI Model Builder, a local-first training platform (https://llm.reallexi.io). Hugging Face repository: reallexi/lexi-coder-v4.4. Copyright (c) 2026 Reallexi LLC. All rights reserved.