# Minicpm5 Finetune Unsloth

Fine-tune MiniCPM5-1B or MiniCPM5-2B with unsloth for tight-VRAM single-GPU LoRA / QLoRA.

- Repository: https://github.com/OpenBMB/MiniCPM (11776 stars)
- Category: Engineering
- License: Apache-2.0
- Last updated: 2026-09-21
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add OpenBMB/MiniCPM --skill minicpm5-finetune-unsloth -g -a claude-code -y`
Codex: `npx skills add OpenBMB/MiniCPM --skill minicpm5-finetune-unsloth -g -a codex -y`
Cursor: `npx skills add OpenBMB/MiniCPM --skill minicpm5-finetune-unsloth -g -a cursor -y`
Gemini CLI: `npx skills add OpenBMB/MiniCPM --skill minicpm5-finetune-unsloth -g -a gemini-cli -y`

## Description

Fine-tune MiniCPM5-1B or MiniCPM5-2B with unsloth for tight-VRAM single-GPU LoRA / QLoRA. Use when the user wants "unsloth", "FastLanguageModel", QLoRA on a 24 GB consumer GPU, or asks for the smallest VRAM footprint.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Fine-tune MiniCPM5-1B and MiniCPM5-2B with unsloth

Single-GPU LoRA / QLoRA. Heavy custom kernels for memory savings (~2× reduction at 4-bit).

> ⚠️ **Two install-time pins (read this BEFORE running)**:
> 1. **`transformers==4.57.3`**, required if vLLM is in the same env (unsloth's vLLM coexistence patch only handles transformers ≤ 4.57.x).
> 2. **`torch==2.7.1` + `torchvision==0.22.1`**, `pip install unsloth` may pull a cu13 torch wheel that fails on cu12.x drivers (`cuda.is_available()` returns False).
>
> ```bash
> pip install --force-reinstall "torch==2.7.1" "torchvision==0.22.1" "transformers==4.57.3"
> ```

## Required input

| Var | Example | Default |
| --- | --- | --- |
| `BASE_MODEL` | `openbmb/MiniCPM5-2B` | required; `openbmb/MiniCPM5-1B` also works |
| `DATA` | path to messages-format jsonl | required |
| `OUTPUT_DIR` | `./runs/minicpm5_unsloth` | required |
| `LOAD_IN_4BIT` | `True` (QLoRA, lowest VRAM) / `False` (LoRA bf16) | `False` |

## Steps

### 1. Install (once), pin transformers to 4.57.3

```bash
pip install "unsloth>=2026.5"
pip install --force-reinstall "transformers==4.57.3"
```

> 🔑 **`transformers==4.57.3` is required** if vLLM is in the same env. unsloth's vLLM coexistence patch only handles transformers ≤ 4.57.x; if you skip this pin you'll get a `dataclass` error at import time.

### 2. Train, save as `train_unsloth.py`

```python
import json, os, torch
os.environ.setdefault("CUDA_VISIBLE_DEVICES", "0")

from unsloth import FastLanguageModel
from datasets import Dataset
from trl import SFTTrainer, SFTConfig

BASE = os.environ["BASE_MODEL"]
DATA = os.environ["DATA"]
OUT  = os.environ["OUTPUT_DIR"]

# 1. Load with unsloth's fast wrapper
model, tok = FastLanguageModel.from_pretrained(
    model_name=BASE,
    max_seq_length=2048,
    dtype=torch.bfloat16,
    load_in_4bit=os.environ.get("LOAD_IN_4BIT", "False") == "True",
    full_finetuning=False,
)

The rest: https://github.com/OpenBMB/MiniCPM/blob/HEAD/skills/minicpm5-finetune-unsloth/SKILL.md