Instructions to use prithivMLmods/Castula-U2-QwenRe-1.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use prithivMLmods/Castula-U2-QwenRe-1.5B with Transformers:

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

pipe = pipeline("text-generation", model="prithivMLmods/Castula-U2-QwenRe-1.5B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("prithivMLmods/Castula-U2-QwenRe-1.5B")
model = AutoModelForCausalLM.from_pretrained("prithivMLmods/Castula-U2-QwenRe-1.5B")
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

vLLM

How to use prithivMLmods/Castula-U2-QwenRe-1.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "prithivMLmods/Castula-U2-QwenRe-1.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prithivMLmods/Castula-U2-QwenRe-1.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/prithivMLmods/Castula-U2-QwenRe-1.5B

SGLang

How to use prithivMLmods/Castula-U2-QwenRe-1.5B 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 "prithivMLmods/Castula-U2-QwenRe-1.5B" \
    --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": "prithivMLmods/Castula-U2-QwenRe-1.5B",
		"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 "prithivMLmods/Castula-U2-QwenRe-1.5B" \
        --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": "prithivMLmods/Castula-U2-QwenRe-1.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use prithivMLmods/Castula-U2-QwenRe-1.5B with Docker Model Runner:
```
docker model run hf.co/prithivMLmods/Castula-U2-QwenRe-1.5B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Castula-U2-QwenRe-1.5B

Castula-U2-QwenRe-1.5B is a compact, multilingual reasoning model fine-tuned from Qwen-1.5B, excelling in mathematical problem solving, logical reasoning, code generation, and general-purpose tasks. Its step-by-step reasoning and bilingual fluency make it ideal for educational systems, coding assistants, and lightweight reasoning applications.

Key Features

Advanced Step-by-Step Reasoning
Fine-tuned to produce intermediate steps for math, logic, and code problems, offering transparency and interpretability crucial for education, coding help, and diagnostics.
Multilingual Proficiency (English + Chinese)
Understands and solves problems in both English and Simplified Chinese, making it accessible in diverse learning and working environments.
Compact Yet Versatile (1.5B Parameters)
Small enough for low-resource environments, yet capable of math, logical puzzles, basic coding tasks, and general comprehension, balancing performance and efficiency.
Structured Computation & Problem Solving
Mirrors human-like multi-step problem-solving, making solutions easy to follow, debug, or verify.

Quickstart with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Castula-U2-QwenRe-1.5B"

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

prompt = "Solve: A train travels 180 km in 3 hours. What is its average speed?"
messages = [
    {"role": "system", "content": "You are a helpful tutor skilled in solving math, logic, and code problems with step-by-step explanations."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Intended Use

Math & Logic Tutoring: Solves problems with explanations ideal for students and educators.
Code Assistant: Helps with beginner-to-intermediate code generation and understanding.
Bilingual Apps: Educational tools in English and Chinese for a global audience.
Lightweight Reasoning Systems: Deployable in mobile apps, browser extensions, and edge devices.

Limitations

Domain Specialization:
Best in math, logic, and code. Performance may degrade in highly creative or abstract language tasks.
Compact Scale:
While efficient, may underperform larger models in deeply complex reasoning or long-context tasks.
Inherited Bias:
May reflect biases from the base model (Qwen-1.5B); outputs should be verified for sensitive or critical uses.
Prompt Sensitivity:
Structured, clearly stated inputs produce significantly better outputs.