Text Generation
Transformers
Safetensors
English
lumees
reasoning
thinking
conversational
custom_code
Instructions to use hasankursun/Lumees-3.8B-Reasoning with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use hasankursun/Lumees-3.8B-Reasoning with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="hasankursun/Lumees-3.8B-Reasoning", trust_remote_code=True) messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("hasankursun/Lumees-3.8B-Reasoning", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use hasankursun/Lumees-3.8B-Reasoning with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "hasankursun/Lumees-3.8B-Reasoning" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hasankursun/Lumees-3.8B-Reasoning", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/hasankursun/Lumees-3.8B-Reasoning
- SGLang
How to use hasankursun/Lumees-3.8B-Reasoning 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 "hasankursun/Lumees-3.8B-Reasoning" \ --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": "hasankursun/Lumees-3.8B-Reasoning", "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 "hasankursun/Lumees-3.8B-Reasoning" \ --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": "hasankursun/Lumees-3.8B-Reasoning", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use hasankursun/Lumees-3.8B-Reasoning with Docker Model Runner:
docker model run hf.co/hasankursun/Lumees-3.8B-Reasoning
Lumees-3.8B-Reasoning
Lumees is a specialized reasoning model distilled from Phi-3 Mini. It features an internal "Thinking Process" enabled by the <|thought|> token, allowing it to break down complex logic before answering.
Architecture
This model uses a custom architecture (LumeesForCausalLM) derived from Phi-3, optimized for structural reasoning chains.
How to Use
You must use trust_remote_code=True to load this model.
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
import torch
# Path to your locally merged model
model_id = "lumees/Lumees-3.8B-Reasoning"
print(f"--- Loading {model_id} ---")
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_id,
trust_remote_code=True,
device_map="auto",
torch_dtype=torch.bfloat16
)
# 1. The Question
messages = [{"role": "user", "content": "What is 25*25-15?"}]
# 2. Apply Template (Standard)
input_ids = tokenizer.apply_chat_template(
messages,
add_generation_prompt=True,
return_tensors="pt"
).to(model.device)
# 3.Force the Thinking Token
# We append the <|thought|> token ID (32002) to the end of the prompt.
# This leaves the model no choice but to start generating inside the thought block.
thought_token_id = tokenizer.convert_tokens_to_ids("<|thought|>")
# Safety check to ensure the token exists
if thought_token_id is None:
print("Warning: <|thought|> token not found. Using raw ID 32002.")
thought_token_id = 32002
input_ids = torch.cat([
input_ids,
torch.tensor([[thought_token_id]], device=model.device)
], dim=1)
print("\n--- Lumees Reasoning ---")
# 4. Stream the Output
# We use TextStreamer so you can watch it think in real-time
streamer = TextStreamer(tokenizer, skip_prompt=True)
outputs = model.generate(
input_ids,
max_new_tokens=1024,
temperature=0.6, # Lower temperature is better for logic
top_p=0.9,
pad_token_id=tokenizer.eos_token_id,
streamer=streamer
)
Prompt: "Solve this: 25 * 25 - 15"
Inference Result (Lumees Output):
<|user|> Solve this: 25 * 25 - 15<|end|>
<|assistant|> <|thought|>
Okay, let's see. The problem is 25 multiplied by 25, then subtract 15. Hmm, let me break it down step by step.
First, I need to do the multiplication part. So, 25 times 25. I remember that 25 squared is 625. So 25 * 25 is 625.
Now, the next step is to subtract 15 from that result. So 625 minus 15. Let me do that subtraction. 625 minus 15. Let's subtract the units first. 5 minus 5 is 0. Then 2 minus 1 is 1. The hundreds place: 6 minus nothing is 6. So the result is 610.
Wait, let me double-check that. 625 minus 15. Yes, 5 minus 5 is 0, 2 minus 1 is 1, 6 minus nothing is 6. So 610. That seems right.
</|thought|>
Therefore, the final answer is 610.<|end|><|endoftext|>
- Downloads last month
- 5
Model tree for hasankursun/Lumees-3.8B-Reasoning
Base model
microsoft/Phi-3-mini-4k-instruct