Instructions to use thealper2/MiniCPM5-1B-Turkish with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use thealper2/MiniCPM5-1B-Turkish with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="thealper2/MiniCPM5-1B-Turkish")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("thealper2/MiniCPM5-1B-Turkish")
model = AutoModelForCausalLM.from_pretrained("thealper2/MiniCPM5-1B-Turkish", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use thealper2/MiniCPM5-1B-Turkish with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Use Docker

docker model run hf.co/thealper2/MiniCPM5-1B-Turkish:Q4_K_M

LM Studio
Jan

vLLM

How to use thealper2/MiniCPM5-1B-Turkish with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "thealper2/MiniCPM5-1B-Turkish"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thealper2/MiniCPM5-1B-Turkish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/thealper2/MiniCPM5-1B-Turkish:Q4_K_M

SGLang

How to use thealper2/MiniCPM5-1B-Turkish with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "thealper2/MiniCPM5-1B-Turkish" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thealper2/MiniCPM5-1B-Turkish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "thealper2/MiniCPM5-1B-Turkish" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thealper2/MiniCPM5-1B-Turkish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use thealper2/MiniCPM5-1B-Turkish with Ollama:
```
ollama run hf.co/thealper2/MiniCPM5-1B-Turkish:Q4_K_M
```

Unsloth Studio

How to use thealper2/MiniCPM5-1B-Turkish with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for thealper2/MiniCPM5-1B-Turkish to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for thealper2/MiniCPM5-1B-Turkish to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for thealper2/MiniCPM5-1B-Turkish to start chatting

How to use thealper2/MiniCPM5-1B-Turkish with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "thealper2/MiniCPM5-1B-Turkish:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use thealper2/MiniCPM5-1B-Turkish with Docker Model Runner:
```
docker model run hf.co/thealper2/MiniCPM5-1B-Turkish:Q4_K_M
```

Lemonade

How to use thealper2/MiniCPM5-1B-Turkish with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Run and chat with the model

lemonade run user.MiniCPM5-1B-Turkish-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use thealper2/MiniCPM5-1B-Turkish with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use thealper2/MiniCPM5-1B-Turkish with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf thealper2/MiniCPM5-1B-Turkish:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "thealper2/MiniCPM5-1B-Turkish:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MiniCPM5-1B-Turkish / README.md

thealper2

Update README.md

8db58fb verified 4 days ago

preview code

Raw

History Blame Contribute Delete

7.39 kB

	---
	language:
	- tr
	- en
	license: apache-2.0
	library_name: transformers
	pipeline_tag: text-generation
	base_model: openbmb/MiniCPM5-1B
	tags:
	- turkish
	- türkçe
	- instruction-tuning
	- sft
	- full-finetune
	- code
	- code-generation
	- minicpm
	- conversational
	- gguf
	- edge-ai
	datasets:
	- AlicanKiraz0/Turkce-Atlas-Instruct
	- tascib/turkish-instruction
	- sixfingerdev/turkish-qa-multi-dialog-dataset
	- berhaan/Turkish-CodeAlpaca-20k
	- alztrk/turkish-code-instructions
	- bysismo/Turkish-Python-instruction-500k
	---

	# MiniCPM5-1B-Turkish

	`openbmb/MiniCPM5-1B` tabanlı, Türkçe talimat takibi ve kod üretimi için tam ince ayar (full fine-tuning) yapılmış bir sohbet modeli.

	A fully fine-tuned (not LoRA/QLoRA) chat model derived from `openbmb/MiniCPM5-1B`, targeting Turkish instruction following and Turkish-instructed code generation.

	## Özet / At a glance

	\| \| \|
	\|---\|---\|
	\| Taban model / Base model \| `openbmb/MiniCPM5-1B` \|
	\| Mimari / Architecture \| LlamaForCausalLM (MiniCPM5) \|
	\| Parametre / Parameters \| 1,080,632,832 \|
	\| Eğitilen parametre / Trained \| 1,080,632,832 (100.0%) \|
	\| Yöntem / Method \| Supervised fine-tuning (SFT), full-parameter \|
	\| Diller / Languages \| Türkçe (birincil), İngilizce (korunmuş) \|
	\| Bağlam / Context \| eğitim 2048 token (taban model 131k) \|
	\| Precision \| torch.bfloat16 \|
	\| Lisans / License \| apache-2.0 \|

	## Kullanım / Usage

	### Transformers

	```python
	import torch
	from transformers import AutoModelForCausalLM, AutoTokenizer

	model_id = "thealper2/MiniCPM5-1B-Turkish"
	tokenizer = AutoTokenizer.from_pretrained(model_id)
	model = AutoModelForCausalLM.from_pretrained(
	model_id, dtype=torch.bfloat16, device_map="auto"
	)

	messages = [
	{"role": "user", "content": "Python'da bir CSV dosyasını okuyup eksik değerleri temizleyen bir fonksiyon yaz."}
	]
	inputs = tokenizer.apply_chat_template(
	messages, add_generation_prompt=True, return_tensors="pt"
	).to(model.device)

	# do_sample=True sarttir - asagidaki nota bakin / see the decoding note below
	out = model.generate(inputs, max_new_tokens=512, do_sample=True, temperature=0.7, top_p=0.95)
	print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))
	```

	> Sohbet şablonu taban modelin ChatML türevidir (`<\|im_start\|>role\n...<\|im_end\|>`). Her zaman `apply_chat_template` kullanın; prompt'u elle kurmayın.

	Model, yanıtlarına boş bir düşünme bloğu (`<think>\n\n</think>`) ile başlayacak şekilde eğitildi; bu, taban modelin `enable_thinking=False` biçimiyle uyumludur. Uzun zincirleme akıl yürütme (CoT) verisiyle eğitilmedi.

	### llama.cpp / GGUF

	```bash
	llama-cli -hf thealper2/MiniCPM5-1B-Turkish -p "Merhaba, kendini tanıt."
	# veya yerel dosyayla:
	llama-cli -m MiniCPM5-1B-Turkish-Q8_0.gguf -cnv
	```

	\| dosya / file \| tür / type \| boyut / size \|
	\|---\|---\|---\|
	\| `MiniCPM5-1B-Turkish-BF16.gguf` \| BF16 \| 2066 MB \|
	\| `MiniCPM5-1B-Turkish-Q4_K_M.gguf` \| Q4_K_M \| 656 MB \|
	\| `MiniCPM5-1B-Turkish-Q5_K_M.gguf` \| Q5_K_M \| 750 MB \|
	\| `MiniCPM5-1B-Turkish-Q8_0.gguf` \| Q8_0 \| 1100 MB \|

	## Eğitim verisi / Training data

	Tüm kaynaklar tek bir konuşma şemasına (`messages`) normalize edildi; ardından temizleme, tüm veri setleri arasında birebir tekrar temizliği (exact dedup) ve kategori bazlı ağırlıklı örnekleme uygulandı. Veri setleri körlemesine birleştirilmedi.

	\| kaynak / source \| kategori \| ham satır \| temizlik sonrası \| seçilen \|
	\|---\|---\|---\|---\|---\|
	\| `AlicanKiraz0/Turkce-Atlas-Instruct` \| general_turkish \| 336,146 \| 336,099 \| 55,000 \|
	\| `tascib/turkish-instruction` \| general_turkish \| 324,080 \| 311,627 \| 55,000 \|
	\| `sixfingerdev/turkish-qa-multi-dialog-dataset` \| qa_dialog \| 21,282 \| 18,790 \| 20,000 \|
	\| `berhaan/Turkish-CodeAlpaca-20k` \| coding \| 19,996 \| 19,565 \| 19,474 \|
	\| `alztrk/turkish-code-instructions` \| coding \| 2,676 \| 1,895 \| 1,895 \|
	\| `bysismo/Turkish-Python-instruction-500k` \| coding \| 335,286 \| 330,668 \| 48,631 \|

	- Birebir tekrar silinen / exact duplicates removed: 27,709
	- Eğitim örneği / train examples: 195,876
	- Doğrulama örneği / validation examples: 3,960
	- Eğitim token'ı / training tokens: 82,165,802 (bunun 42,684,848 tanesi kayıp hesabına giriyor)
	- Hedef dağılım / target mix: general_turkish 55%, coding 35%, qa_dialog 10%

	## Eğitim yöntemi / Training procedure

	\| ayar / setting \| değer / value \|
	\|---\|---\|
	\| Yöntem \| Full fine-tuning (LoRA/QLoRA kullanılmadı) \|
	\| Kayıp / Loss \| Yalnızca asistan token'ları (kullanıcı ve sistem token'ları -100 ile maskelendi) \|
	\| Epoch \| 1 \|
	\| Learning rate \| 2e-05 \|
	\| Scheduler \| cosine (warmup 0.03) \|
	\| Optimizer \| paged_adamw_8bit \|
	\| Weight decay \| 0.1 \|
	\| Grad clipping \| 1.0 \|
	\| Batch \| 1 x 16 accum \|
	\| Sekans uzunluğu \| 2048 \|
	\| Sequence packing \| True \|
	\| Gradient checkpointing \| True \|
	\| Precision \| torch.bfloat16 \|
	\| Donanım / Hardware \| NVIDIA GeForce RTX 5060 Ti \|

	### Eğitim metrikleri / Training metrics

	- İlk kayıp / first loss: `2.0483232498168946`
	- Son kayıp / last loss: `1.3010472297668456`
	- En iyi doğrulama kaybı / best eval loss: `1.276915431022644`
	- Zirve GPU belleği / peak VRAM: `5.435 GB`

	## Değerlendirme / Evaluation

	Değerlendirme, taban model ile ince ayarlı model üzerinde aynı prompt'lar ve aynı çözümleme ayarlarıyla yapıldı. Türkçe kazanımının genel bir iyileşme anlamına gelmediğini görebilmek için İngilizce ve akıl yürütme yetenekleri de ayrıca ölçüldü (regresyon testi).

	\| sonuç / verdict \| prompt \|
	\|---\|---\|
	\| improved \| 9 \|
	\| not auto-scored \| 9 \|
	\| possible regression \| 2 \|
	\| unchanged \| 10 \|

	Olası gerilemeler / possible regressions

	- `code_js_01` (code_generation): 0/1 vs 1/1 checks
	- `en_code_02` (english_coding): 0/2 vs 2/2 checks

	## Sınırlar ve riskler / Limitations

	- 1B parametreli küçük bir modeldir; olgusal doğruluk sınırlıdır ve halüsinasyon görülebilir.
	- Eğitim verisi büyük ölçüde sentetik/derlenmiş Türkçe talimat setlerinden gelir; kaynaklardaki hatalar modele geçmiş olabilir.
	- Kaynaklardan biri (`alztrk/turkish-code-instructions`) Türkçe karakterleri ASCII'ye indirgenmiş metin içerir; bu nedenle katkısı bilinçli olarak düşük tutuldu.
	- Kod çıktıları çalıştırılmadan doğrulanmamıştır; üretime almadan önce test edin.
	- Tam ince ayar yapıldığı için taban modelin bazı yetenekleri (ör. araç çağırma, uzun bağlam, 2048 token üstü davranış) zayıflamış olabilir.
	- Güvenlik hizalaması ayrıca eğitilmedi; taban modelden gelen davranış korunmaya çalışıldı ancak garanti edilmez.
	- Model tıbbi, hukuki veya finansal tavsiye için kullanılmamalıdır.

	## Yeniden üretim / Reproduction

	```bash
	python scripts/inspect_datasets.py
	python scripts/prepare_datasets.py
	python scripts/train.py --dry_run
	python scripts/train.py
	python scripts/compare_models.py --run
	```

	Tüm hiperparametreler `config/training.yaml` içindedir ve eğitim çıktısıyla birlikte `config.yaml`, `dataset_report.json`, `training_summary.json` olarak kaydedilir.

	## Atıf / Citation

	```bibtex
	@misc{minicpm5_turkish,
	title = {MiniCPM5-1B-Turkish},
	note = {Full supervised fine-tune of openbmb/MiniCPM5-1B for Turkish instruction following and coding},
	year = {2026}
	}
	```

	Taban model / base model: [`openbmb/MiniCPM5-1B`](https://huggingface.co/openbmb/MiniCPM5-1B)