Instructions to use llaa33219/BokantLM0.1-135M-Deepseek with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use llaa33219/BokantLM0.1-135M-Deepseek with Transformers:

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

pipe = pipeline("text-generation", model="llaa33219/BokantLM0.1-135M-Deepseek")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("llaa33219/BokantLM0.1-135M-Deepseek")
model = AutoModelForCausalLM.from_pretrained("llaa33219/BokantLM0.1-135M-Deepseek", 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

vLLM

How to use llaa33219/BokantLM0.1-135M-Deepseek with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "llaa33219/BokantLM0.1-135M-Deepseek"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "llaa33219/BokantLM0.1-135M-Deepseek",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/llaa33219/BokantLM0.1-135M-Deepseek

SGLang

How to use llaa33219/BokantLM0.1-135M-Deepseek 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 "llaa33219/BokantLM0.1-135M-Deepseek" \
    --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": "llaa33219/BokantLM0.1-135M-Deepseek",
		"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 "llaa33219/BokantLM0.1-135M-Deepseek" \
        --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": "llaa33219/BokantLM0.1-135M-Deepseek",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use llaa33219/BokantLM0.1-135M-Deepseek with Docker Model Runner:
```
docker model run hf.co/llaa33219/BokantLM0.1-135M-Deepseek
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

BokantLM0.1–135M-Deepseek

BokantLM – "Small but Supreme in Its Domain"

BokantLM is not a general-purpose model that tries to do everything well.
Instead, it is an ultra-lightweight LLM designed to focus on a single domain, delivering the highest possible efficiency and performance in that area.

Overview

Model Name: BokantLM0.1–135M-Deepseek
Base Model: SmolLM2-135M
Fine-tuning Dataset: paloalma/Reasoning-DeepSeek-R1-Distilled-1.4M-Alpaca-V2

Philosophy

While most LLMs aim for versatility by learning across many fields,
BokantLM is built to achieve top efficiency and performance within a specific domain.

Optimized for DeepSeek R1’s reasoning methodology, This model is fine-tuned to replicate DeepSeek R1’s distinctive reasoning style—delivering the same logical depth and problem-solving approach in a compact, ultra-lightweight architecture.

The reason I created this model

I created this model based on the idea of what would happen if a very small model tried to imitate Deepseek R1.

Future Plans

✅ Imitating Deepseek-R1 model release (current version)
🔄 Mathematics problem-solving specialized version
🔄 Domain-specific ultra-lightweight models for law, medicine, science, etc.
🔄 multilingual support with very small models(1K, 10k, 100K, etc parameters)
🔄 fine-tuning of models with very large maximum tokens but small parameter sizes