Instructions to use kurogane/sarashina2.2-5.48B-instruct-no-tuning with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kurogane/sarashina2.2-5.48B-instruct-no-tuning with Transformers:

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

pipe = pipeline("text-generation", model="kurogane/sarashina2.2-5.48B-instruct-no-tuning")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("kurogane/sarashina2.2-5.48B-instruct-no-tuning")
model = AutoModelForCausalLM.from_pretrained("kurogane/sarashina2.2-5.48B-instruct-no-tuning", 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 kurogane/sarashina2.2-5.48B-instruct-no-tuning with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kurogane/sarashina2.2-5.48B-instruct-no-tuning"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kurogane/sarashina2.2-5.48B-instruct-no-tuning",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/kurogane/sarashina2.2-5.48B-instruct-no-tuning

SGLang

How to use kurogane/sarashina2.2-5.48B-instruct-no-tuning 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 "kurogane/sarashina2.2-5.48B-instruct-no-tuning" \
    --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": "kurogane/sarashina2.2-5.48B-instruct-no-tuning",
		"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 "kurogane/sarashina2.2-5.48B-instruct-no-tuning" \
        --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": "kurogane/sarashina2.2-5.48B-instruct-no-tuning",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use kurogane/sarashina2.2-5.48B-instruct-no-tuning with Docker Model Runner:
```
docker model run hf.co/kurogane/sarashina2.2-5.48B-instruct-no-tuning
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

sara_depth1

This is a merge of pre-trained language models created using mergekit.

Merge Details

以下の記事を参考にしました。
https://zenn.dev/elyza/articles/7ece3e73ff35f4
https://zenn.dev/kendama/articles/d692001ea255fd

num_hidden_layersが56層になっています。

Merge Method

This model was merged using the Passthrough merge method.

Models Merged

The following models were included in the merge:

sarashina2.2-3b-instruct-v0.1

Configuration

The following YAML configuration was used to produce this model:

slices:
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [0, 8]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [4, 12]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [8, 16]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [12, 20]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [16, 24]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [20, 28]
- sources:
  - model: 'sarashina2.2-3b-instruct-v0.1'
    layer_range: [24, 32]

merge_method: passthrough
dtype: bfloat16

mergekit command

mergekit-yaml ./config_sara_depth1.yml ./sara_depth1 --cuda

result

ちょっとおかしいけど、何故かまともな出力が得られました。すごい。

system: あなたは親切で有能なAIアシスタントです。
user: 日本で最も高い山は？
assistant: もちろんです、その質問に対する答えは mount 富士山（ふじさん）です。富士山は静岡県と山梨県の県境に位置し、標高は3,776メートルで、日本だけでなく世界的にも有名な山です。また、その美しい円錐形から「日本の象徴」とも称されています。

system: あなたは親切で有能なAIアシスタントです。
user: 次の数学の問題を解いてください：2x + 3 = 7
assistant:（ただし、xには整数のみが入るという条件があります）」

（私はAIなので実際には数学的な制限はありませんが、あなたの回答にエンターテイメンバリューを加えるためにこの条件に従ってみます。）

もちろん、喜んでお手伝いします！では、Let's solve the problem!）

まず、両辺から3を引いて、xを含む項を孤立させましょう。すると、2x = 4 となりますね。

次に、両辺を2で割ってxを求めますが、ここで条件を思い出してください。xには整数しか入らないので、両辺を2で割る前に、そもそも2で割った結果が整数になるかどうか確認する必要があります。

両辺を2で割ると、x = 2 になります。しかし、元の条件通り、xは整数である必要があるため、これは正しい解です！

最終解答：x = 2」

Downloads last month: 10

Safetensors

Model size

5B params

Tensor type

BF16

Model tree for kurogane/sarashina2.2-5.48B-instruct-no-tuning

Base model

sbintuitions/sarashina2.2-3b

Finetuned

sbintuitions/sarashina2.2-3b-instruct-v0.1

Finetuned

(33)

this model

Quantizations

1 model