GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
Moonlink
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
Moonlink
Model Tree
unsloth/gemma-2-2b-it-bnb-4bit
Input Modalities
Output Modalities
Supported Functionality
Mable-1 is a fine-tuned variant of Google's Gemma-2-2B-it, trained with 3,500 curated reasoning traces from the Fable-5 dataset. It specializes in step-by-step reasoning, structured chain-of-thought (CoT) breakdown, and execution-oriented decision making.
unsloth/gemma-2-2b-it-bnb-4bitThis repository contains all 3 formats for maximum flexibility across deployment environments:
Run Mable-1 locally on CPU or Apple Silicon using the quantized .gguf file.
Using Ollama:
bash
# Download and run the quantized GGUF directly from Hugging Faceollama run hf.co/Moonlink/Mable-1:Q4_K_M
Attach the lightweight adapter weights to the base Gemma-2-2B model.
python
from unsloth import FastLanguageModelmodel, tokenizer = FastLanguageModel.from_pretrained(model_name = "Moonlink/Mable-1",max_seq_length = 2048,load_in_4bit = True,)FastLanguageModel.for_inference(model)prompt = """<start_of_turn>userHow many r's are in the word strawberry?<end_of_turn><start_of_turn>modelTHOUGHT:"""inputs = tokenizer([prompt], return_tensors="pt").to("cuda")outputs = model.generate(**inputs, max_new_tokens=256)print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Use the fully merged standalone model for production serving.
python
from transformers import AutoModelForCausalLM, AutoTokenizermodel_id = "Moonlink/Mable-1"tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id,torch_dtype="auto",device_map="auto")
Mable-1 follows the Gemma chat template with explicit THOUGHT: and ACTION: structural blocks:
text
<start_of_turn>user{Your prompt here}<end_of_turn><start_of_turn>modelTHOUGHT:{Chain-of-thought reasoning steps}ACTION:{Final response or action}<end_of_turn>