Instructions to use ufakai/ufakzeka-1-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ufakai/ufakzeka-1-base with Transformers:

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

pipe = pipeline("text-generation", model="ufakai/ufakzeka-1-base")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ufakai/ufakzeka-1-base")
model = AutoModelForCausalLM.from_pretrained("ufakai/ufakzeka-1-base", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ufakai/ufakzeka-1-base with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/ufakai/ufakzeka-1-base

SGLang

How to use ufakai/ufakzeka-1-base 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 "ufakai/ufakzeka-1-base" \
    --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": "ufakai/ufakzeka-1-base",
		"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 "ufakai/ufakzeka-1-base" \
        --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": "ufakai/ufakzeka-1-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ufakai/ufakzeka-1-base with Docker Model Runner:
```
docker model run hf.co/ufakai/ufakzeka-1-base
```

ufakzeka-1-base / ATTRIBUTION.md

stfurkan

ufakzeka-1 release: weights, card and attribution

f9e11ee 22 days ago

preview code

Raw History Blame Contribute Delete

2.89 kB

Attribution

ufakzeka-1 is a produced work derived from the following datasets. Each is used under its licence; the ODC-By 1.0 datasets require this notice.

dataset	authors	licence
FineWeb2-HQ (tur_Latn)	EPFL Machine Learning and Optimization Lab (Messmer, Ali, Jaggi), built on FineWeb-2 by Hugging Face (Penedo, Kydlicek et al.)	ODC-By 1.0, Common Crawl terms of use
mogan-turkish-web	Mogan AI	ODC-By 1.0, Common Crawl terms of use
FinePDFs-edu (tur_Latn)	Hugging Face (Kydlicek et al.)	ODC-By 1.0, Common Crawl terms of use
FineMath (finemath-4plus)	Hugging Face (Allal, Lozhkov et al.)	ODC-By 1.0, Common Crawl terms of use
FineWiki (tr)	Hugging Face, from Wikipedia	CC-BY-SA 4.0 and GFDL
BILGE-Synthetic-Stories, -Web, -Math	TUBITAK BILGEM	Apache-2.0
COSMOS-Sentetic-Turkish-Corpus-2GB	Berkesule	Apache-2.0
Turkce-Atlas-Instruct, Turkish-SFT-Dataset-v1.0	Alican Kiraz	MIT
Aya Dataset (tur)	Cohere Labs and Aya contributors	Apache-2.0
everyday-conversations-tur	SoAp9035	Apache-2.0
diyalog-dataset	Ali Bayram	Apache-2.0
WikiRAG-TR	Metin	Apache-2.0
InstructPapers-TR	selimc	Apache-2.0
gsm8k_tr	YTU COSMOS	Apache-2.0
Turkish Wikipedia dump, August 2026 (recency tier; lead sentences rendered as question and answer pairs for post-training)	Wikimedia contributors	CC-BY-SA 4.0
Vikikaynak (Turkish Wikisource): public-domain folk poems and türkü lyrics, Nasreddin Hoca and Karadeniz fıkras, folk riddles and tales (post-training)	authors dead more than 70 years, transcribed by Wikisource contributors	texts public domain; the transcriptions CC-BY-SA 4.0
Common Crawl CC-MAIN-2026-30 and 2026-34, Turkish pages (recency tier)	Common Crawl	Common Crawl terms of use
orpo-dpo-mix-TR-20k	selimc	Apache-2.0; used only in preference training experiments that are not part of the released weights

Generated data, written for this project: about 18,000 multi-turn Turkish dialogues in several sets, about 7,500 short Turkish stories in the TinyStories style, 600 short original poems, acrostics for Turkish first names, 3,500 warm conversations, revision follow-ups, free-conversation sets and 5,000 boundary cases between answerable and unknowable questions, all written by various large language models, each set validated by rules and scored by an LLM judge before use; and templated identity, arithmetic, percentage, correction, fact, memory, abstention and safety conversations produced by code in this repository. None of the generated data is redistributed in this release; the weights are the derived work.

Evaluation sets (never trained on): TurkishMMLU (Yuksel et al.), TurBLiMP (Basar et al., CC-BY 4.0), XCOPA, Belebele (Meta, CC-BY-SA 4.0), hellaswag_tr and arc-tr (malhajar), and the Cetvel suite (KUIS-AI).