Instructions to use Flexan/Blake-Haiku-1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Flexan/Blake-Haiku-1 with Transformers:

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

pipe = pipeline("text-generation", model="Flexan/Blake-Haiku-1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Flexan/Blake-Haiku-1")
model = AutoModelForCausalLM.from_pretrained("Flexan/Blake-Haiku-1", 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 Flexan/Blake-Haiku-1 with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Flexan/Blake-Haiku-1

SGLang

How to use Flexan/Blake-Haiku-1 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 "Flexan/Blake-Haiku-1" \
    --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": "Flexan/Blake-Haiku-1",
		"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 "Flexan/Blake-Haiku-1" \
        --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": "Flexan/Blake-Haiku-1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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

Blake Haiku 1

Description

Blake Haiku 1 is an instruct LLM consisting of 0.6B parameters trained to talk in a human conversational manner. It was trained without support for reasoning nor tool-calling.
The model was LoRA fine-tuned with Qwen/Qwen3-0.6B as base model.

This model was primarily made as a test of a new runtime environment allowing me to train bigger models than before on own hardware.
Consider this upload to be a celebration of, after many months, having found a way to successfully start training these models on Windows 11 CUDA.

Warning: This model is merely archived for above reason and is not meant to be deployed in production. Training data was minimal.
There will likely not be a Blake Haiku 2.

Chat Format

Blake Haiku 1 uses the ChatML format, e.g.:

<|im_start|>system
System message<|im_end|>
<|im_start|>user
User prompt<|im_end|>
<|im_start|>assistant
Assistant response<|im_end|>

Usage

We recommend using the following system prompt:

You're Moke, a user chatting with random people on Discord.

The name is supposed to be dynamic, but due to this model's and dataset's small size, this is likely not supported.

The assistant response has the following format:

<|im_start|>assistant
<think>

</think>

What happened? :0
I wanna know! >.<<|im_end|>

Each line is supposed to be a new "message" in a conversation, mimicking humans using traditional chatting platforms (e.g. Discord, where you can send multiple messages before someone responds).

Note that the <think>...</think> tags are always empty, as this model was not trained on reasoning data.

Datasets

Private dataset 72 chats / 315 completions
Smaller version of the dataset used for the Blake Sonnet model.

Downloads last month: 163

Safetensors

Model size

0.6B params

Tensor type

BF16

Model tree for Flexan/Blake-Haiku-1

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

Qwen/Qwen3-0.6B

Finetuned

(1307)

this model

Quantizations

2 models