Instructions to use TilQazyna/Til-Core-1B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use TilQazyna/Til-Core-1B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="TilQazyna/Til-Core-1B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("TilQazyna/Til-Core-1B-Instruct")
model = AutoModelForCausalLM.from_pretrained("TilQazyna/Til-Core-1B-Instruct")
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

vLLM

How to use TilQazyna/Til-Core-1B-Instruct with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/TilQazyna/Til-Core-1B-Instruct

SGLang

How to use TilQazyna/Til-Core-1B-Instruct 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 "TilQazyna/Til-Core-1B-Instruct" \
    --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": "TilQazyna/Til-Core-1B-Instruct",
		"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 "TilQazyna/Til-Core-1B-Instruct" \
        --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": "TilQazyna/Til-Core-1B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use TilQazyna/Til-Core-1B-Instruct with Docker Model Runner:
```
docker model run hf.co/TilQazyna/Til-Core-1B-Instruct
```

Til Core 1B Instruct

Chat/instruct version of TilQazyna/Til-Core-1B, supervised-fine-tuned on native-Kazakh instruction–response pairs (ChatML format, assistant-only loss). No translated data, no eval-set contamination.

⚠️ Early v1 / research preview. Follows the chat format and answers in Kazakh, but factual accuracy is limited (1.25B params, small SFT set). Not for production or factual reliance.

Details


Base	Til-Core-1B (1.246B, morphbpe-256k)
SFT data	AmanMussa/kazakh-instruction-v2 — 52 173 native-kk Alpaca-style pairs
Format	ChatML (`<
Loss	assistant tokens only
Recipe	3 epochs, LR 1e-5 cosine, bf16, 8×H200 FSDP
Stop token	`<

Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

name = "TilQazyna/Til-Core-1B-Instruct"
tok = AutoTokenizer.from_pretrained(name)
m = AutoModelForCausalLM.from_pretrained(name, dtype=torch.bfloat16).cuda().eval()

msg = [{"role": "user", "content": "Денсаулықты сақтаудың үш кеңесін айт."}]
p = tok.apply_chat_template(msg, tokenize=False, add_generation_prompt=True)
ids = tok(p, add_special_tokens=False, return_tensors="pt").input_ids.cuda()
out = m.generate(ids, max_new_tokens=160, do_sample=True, temperature=0.7,
                 top_p=0.9, repetition_penalty=1.2,
                 eos_token_id=tok.convert_tokens_to_ids("<|im_end|>"))
print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))

Example

User: Қазақстанның астанасы қай қала және ол туралы қысқаша айт. Assistant: Қазақстанның елордасы — Астана қаласы. Ол Есіл өзенінің жағасында орналасқан…

User: Денсаулықты сақтаудың үш кеңесін айт. Assistant: 1. Салауатты өмір салтын ұстану; 2. Дұрыс тамақтану; 3. Тұрақты дене жаттығулары…

Limitations

Small model + small SFT set → weak factual accuracy, occasional topic drift.
No RLHF / safety alignment.
Kazakh-only.

Roadmap

Larger / cleaner SFT set, preference tuning.
A smaller on-device instruct sibling.
Task-specialized variants (e.g. Kazakh grammar correction — see Til-Core experiments).

Downloads last month: 21

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for TilQazyna/Til-Core-1B-Instruct

Base model

TilQazyna/Til-Core-1B

Finetuned

(1)

this model