Instructions to use karakuri-ai/karakuri-lm-32b-thinking-2501-exp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use karakuri-ai/karakuri-lm-32b-thinking-2501-exp with Transformers:

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

pipe = pipeline("text-generation", model="karakuri-ai/karakuri-lm-32b-thinking-2501-exp")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("karakuri-ai/karakuri-lm-32b-thinking-2501-exp")
model = AutoModelForCausalLM.from_pretrained("karakuri-ai/karakuri-lm-32b-thinking-2501-exp")
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]:]))

Inference
Local Apps Settings

vLLM

How to use karakuri-ai/karakuri-lm-32b-thinking-2501-exp with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "karakuri-ai/karakuri-lm-32b-thinking-2501-exp"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "karakuri-ai/karakuri-lm-32b-thinking-2501-exp",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/karakuri-ai/karakuri-lm-32b-thinking-2501-exp

SGLang

How to use karakuri-ai/karakuri-lm-32b-thinking-2501-exp 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 "karakuri-ai/karakuri-lm-32b-thinking-2501-exp" \
    --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": "karakuri-ai/karakuri-lm-32b-thinking-2501-exp",
		"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 "karakuri-ai/karakuri-lm-32b-thinking-2501-exp" \
        --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": "karakuri-ai/karakuri-lm-32b-thinking-2501-exp",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use karakuri-ai/karakuri-lm-32b-thinking-2501-exp with Docker Model Runner:
```
docker model run hf.co/karakuri-ai/karakuri-lm-32b-thinking-2501-exp
```

karakuri-lm-32b-thinking-2501-exp / README.md

ynakashima

Update README.md

e77c18c verified over 1 year ago

2.16 kB

library_name: transformers
license: apache-2.0
language:
  - en
  - ja
base_model: Qwen/QwQ-32B-Preview

KARAKURI LM 32B Thinking 2501 Experimental

Model Details

Model Description

Developed by: KARAKURI Inc.
Model type: Causal Language Models
Languages: Japanese
License: Apache 2.0
Finetuned from model: Qwen/QwQ-32B-Preview
Contact: For questions and comments about the model, please email karakuri-rd@karakuri.ai
Demo: https://lm.karakuri.cc/

Usage

Run the model

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "karakuri-ai/karakuri-lm-32b-thinking-2501-exp"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [
    {"role": "user", "content": "こんにちは。"}
]
input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)
outputs = model.generate(input_ids, max_new_tokens=512)
tokenizer.decode(outputs[0][input_ids.shape[-1]:])

Training Details

Training Infrastructure

Hardware: The model was trained on 16 nodes of an Amazon EC2 trn1.32xlarge instance.
Software: We use code based on neuronx-nemo-megatron.

Acknowledgments

This work was supported by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO) through the Generative AI Accelerator Challenge (GENIAC).

Citation

@misc{karakuri_lm_32b_thinking_2501_exp,
    author       = { {KARAKURI} {I}nc. },
    title        = { {KARAKURI} {LM} 32{B} {T}hinking 2501 {E}xperimental },
    year         = { 2025 },
    url          = { https://huggingface.co/karakuri-ai/karakuri-lm-32b-thinking-2501-exp },
    publisher    = { Hugging Face },
    journal      = { Hugging Face repository }
}