GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Umranz
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Umranz
Model Tree
Qwen/Qwen2.5-1.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Aetheris-Small-1.5b is a specialized, zero-shot "FrankenMoE" created by merging three distinct Qwen2.5-1.5B models using MergeKit.
By leveraging a Shared Expert for base language understanding alongside three highly distinct Routed Experts, Aetheris dynamically routes inputs to the most capable subset of parameters. This provides broad capabilities across code, mathematics, and general chat without the memory footprint of a massive dense model.
This MoE model comprises 3 Routed Experts and 1 Shared Expert, with a routing protocol of 2 active experts per token:
| Expert Domain | Source Model | Routing Focus |
|---|---|---|
| 💻 Code & Logic | qwen2.5-Coder-1.5B | Python, C++, SQL, debugging, algorithms, syntax. |
| 🔢 Mathematics | qwen2.5-math-1.5b | Calculus, algebra, matrix operations, step-by-step proofs. |
| 💬 Chat & Instruct | qwen2.5-1.5b-instruct | General conversational AI, translation, summarization. |
| 🌐 Shared Expert | qwen2.5-1.5b-instruct | Always-active base layer for foundational language structure. |
Aetheris is lightweight enough to run entirely in float16 on free-tier GPUs (like the NVIDIA T4 in Google Colab).
python
!pip install -U transformers accelerate import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM model_id = "Umranz/Aetheris-Small-1.5b" tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype=torch.float16, device_map="auto", trust_remote_code=True) prompt = "Write a python function to compute the factorial of a number."messages = [{"role": "user", "content": prompt}]text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer([text], return_tensors="pt").to("cuda") with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=150, do_sample=True, temperature=0.7 ) response = tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True)print(response)!pip install -U transformers accelerate import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM model_id = "Umranz/Aetheris-Small-1.5b" tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype=torch.float16, device_map="auto", trust_remote_code=True) prompt = "Write a python function to compute the factorial of a number."messages = [{"role": "user", "content": prompt}]text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer([text], return_tensors="pt").to("cuda") with torch.no_grad(): outputs = model.generate( **inputs, max_new_tokens=150, do_sample=True, temperature=0.7 ) response = tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True)print(response)