Instructions to use henryhale/Kintu-2.0-30M-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use henryhale/Kintu-2.0-30M-Base with Transformers:

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

pipe = pipeline("text-generation", model="henryhale/Kintu-2.0-30M-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("henryhale/Kintu-2.0-30M-Base")
model = AutoModelForCausalLM.from_pretrained("henryhale/Kintu-2.0-30M-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 henryhale/Kintu-2.0-30M-Base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "henryhale/Kintu-2.0-30M-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": "henryhale/Kintu-2.0-30M-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/henryhale/Kintu-2.0-30M-Base

SGLang

How to use henryhale/Kintu-2.0-30M-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 "henryhale/Kintu-2.0-30M-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": "henryhale/Kintu-2.0-30M-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 "henryhale/Kintu-2.0-30M-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": "henryhale/Kintu-2.0-30M-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use henryhale/Kintu-2.0-30M-Base with Docker Model Runner:
```
docker model run hf.co/henryhale/Kintu-2.0-30M-Base
```

Kintu-2.0-30M-Base

A 30M-parameter Luganda language model, pretrained from scratch with the MiniMind recipe. It's a base model: it continues text and doesn't chat. The chat model is Kintu-2.0-30M.


Architecture	Qwen3-style decoder (`Qwen3ForCausalLM`): RMSNorm, SwiGLU, RoPE (θ = 1e6), QK-norm, grouped-query attention (8 query / 4 KV heads), tied embeddings
Size	30.0M parameters: 8 layers, hidden size 512, MLP 1,664
Tokenizer	Byte-level BPE, 6,400 tokens, trained on the pretraining data
Context	512 tokens

Training

273M tokens of henryhale/luganda-pretrain2, packed into sequences of up to 512 tokens, for 2 epochs: 256 sequences per step, AdamW, learning rate 5e-4 with cosine decay to 5e-5, fp16, on 2× T4 in 2.1 hours. Final training loss ≈ 2.49.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("henryhale/Kintu-2.0-30M-Base")
model = AutoModelForCausalLM.from_pretrained("henryhale/Kintu-2.0-30M-Base")
enc = tok(tok.bos_token + "Olunaku lumu,", return_tensors="pt", return_token_type_ids=False)  # pretraining texts started with <|im_start|>
out = model.generate(**enc, max_new_tokens=80)
print(tok.decode(out[0], skip_special_tokens=True))

Samples

Sampled with temperature 0.85 and top-p 0.95:

💬 Oli otya?
🧠 Oyinza otya okugamba nti olaba nga okimanyi nti Katonda akulaba? (Zabbuli 67:9) Tetuyinza kutya Katonda nga tusinza ebifaananyi. N'olwekyo, tetujja kwesiga nsibuko z'obunnabbi bwaffe ez'edda.—2 Abakkolinso 6:17; Matayo 6:38.
Yakuwa Katonda mu ngeri ey

💬 Wasuze otya nno?
🧠 Kuba ntandise bulungi,
Njagala okakkaabya
Olwaleero ngenda kuddamu kusisinkana bakulu,
Nali nebaza mukama wange, nansaba nkubirayo
Okusinziira ku bakulu, nkubira abatujju
Abatujju bonna abakozesa ennyimba ne Bannansi ba Misiri n'abajaasi
Nti ntuuse mu kibuga
Nti

💬 Osiibye otyanno ssebo?
🧠 Munnange mumanyi bulungi. Nze maama, nnafuuka kiggala. Wosomera mu luyimba lwa Golden Band, n'okulutuga wano mu Buganda ne mu Buganda. Wulira obuyonjo mu bitundu by'obubaga bw'abawala bunyuma, basomba emmere, embuga ezitaggwa, abalambuzi, Abaana abasomyeko

💬 Ndi musanyufu nnyo okulaba.
🧠 - "Nnyinza okuteeka omupiira mu kasanduuko k'eby'okusangamu. "
Omutwazi bwe yawulira ekyo, yamwenyumira era n'amwenya ng'agamba nti, "Ekyo kirungi nnyo.
Yali akimanyi nti wadde nga kyali kizibu nnyo, yali asobola okukikola.
Omusajja omukulu yamwenya n'amugamba nti, "Lwaki tojja kwegatta?
Maama wa John yamwenya n'amugamba nti, "Totya.

💬 Weebale nnyo okuyamba bulungi.
🧠 - Yamutunuulira n'amwenya, ng'asanyuka nnyo okuba nti yali azzeemu okubeerako mu kisiikirize.
Oluvannyuma lw'ekiseera kitono, mukwano gwe ow'okubiri n'ajja n'amugamba nti, "Totya, nnina ekintu ekikulu kye njagala okukuwa".
Bwe yali anaatera okutuuka ku mulyango, empewo yatandika okubuuka era omuguwa ne gutandika okugwa.
John yaseka nnyo bwe yawulira ebigambo ebyo nga bigenda

Limitations

At 30M parameters it writes fluent Luganda but is often off-topic and gets facts wrong, so don't rely on it for information. It was trained on sequences of up to 512 tokens, and TinyStories translations are 68.5% of its pretraining tokens, so it leans towards simple stories.

Downloads last month: 6

Safetensors

Model size

30M params

Tensor type

F16

Model tree for henryhale/Kintu-2.0-30M-Base

Finetunes

1 model

Dataset used to train henryhale/Kintu-2.0-30M-Base

Collection including henryhale/Kintu-2.0-30M-Base

Kintu - 2.x

Collection

Luganda language models pretrained from scratch with the MiniMind recipe on 2× T4 GPUs • 3 items • Updated about 18 hours ago