Instructions to use 0labs-in/Sky-v2.0-11B-INT4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use 0labs-in/Sky-v2.0-11B-INT4 with Transformers:

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

pipe = pipeline("text-generation", model="0labs-in/Sky-v2.0-11B-INT4", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("0labs-in/Sky-v2.0-11B-INT4", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use 0labs-in/Sky-v2.0-11B-INT4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "0labs-in/Sky-v2.0-11B-INT4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "0labs-in/Sky-v2.0-11B-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/0labs-in/Sky-v2.0-11B-INT4

SGLang

How to use 0labs-in/Sky-v2.0-11B-INT4 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 "0labs-in/Sky-v2.0-11B-INT4" \
    --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": "0labs-in/Sky-v2.0-11B-INT4",
		"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 "0labs-in/Sky-v2.0-11B-INT4" \
        --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": "0labs-in/Sky-v2.0-11B-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use 0labs-in/Sky-v2.0-11B-INT4 with Docker Model Runner:
```
docker model run hf.co/0labs-in/Sky-v2.0-11B-INT4
```

Sky v2.0-11B-INT4 — Full CREST, Memory-Efficient

Built by 0labs — Atharvsinh Jadav, Gujarat, India
Full adaptive-depth architecture. Fits on consumer GPUs.

What is This Model?

Sky v2.0-11B-INT4 is the complete 11.2 billion parameter CREST model with all K=4 adaptive-depth steps preserved. The weights are stored in BFloat16 on disk and can be loaded with 4-bit quantization at inference time, reducing VRAM usage from 22GB to **6GB**.

This gives you the full power of CREST — all 4 computational steps, all halting gates, complete adaptive-depth behavior — on hardware as modest as an NVIDIA RTX 3060.

How CREST Works

Standard transformers give every token the same amount of computation. CREST changes this by replacing each FFN with 4 independent MLPs and a learned halting gate:

Easy token ("the")           → 1 step   (fast exit)
Medium token ("fibonacci")   → 2 steps  (moderate processing)
Hard token (algorithm logic)  → 3-4 steps (deep reasoning)

The model learns which tokens need more thought. This is analogous to Kahneman's System 1 (fast, automatic) vs System 2 (slow, deliberate) thinking.

Model Details

Property	Value
Architecture	CREST (0labs proprietary)
Total Parameters	11.2B
CREST Steps (K)	4 (all preserved)
Hidden Dimension	2,560
Layers	32
Attention Heads	32
Context Window	32,768 tokens
Disk Size	~22GB (BFloat16)
VRAM (BF16)	~22GB
VRAM (INT4)	~6GB
Vocabulary Size	248,320
License	Apache 2.0

Quick Start

Standard Loading (needs ~22GB VRAM)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "0labs-in/Sky-v2.0-11B-INT4",
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("0labs-in/Sky-v2.0-11B-INT4")

4-Bit Loading (needs only ~6GB VRAM) ⭐

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_quant_type="nf4",
)

model = AutoModelForCausalLM.from_pretrained(
    "0labs-in/Sky-v2.0-11B-INT4",
    trust_remote_code=True,
    quantization_config=quantization_config,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("0labs-in/Sky-v2.0-11B-INT4")

Generation

prompt = "Implement a binary search algorithm in Python with detailed comments."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=1024,
        temperature=0.7,
        top_p=0.9,
        do_sample=True,
    )

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Note: trust_remote_code=True is required — CREST uses custom architecture code included in this repository.

Hardware Requirements

Setup	Loading Mode	VRAM	Works?
RTX 4090 (24GB)	BF16	~22GB	✅
RTX 3090 (24GB)	BF16	~22GB	✅
A100 (40/80GB)	BF16	~22GB	✅
MI300X (192GB)	BF16	~22GB	✅
RTX 3060 (12GB)	INT4	~6GB	✅
RTX 4060 (8GB)	INT4	~6GB	✅
Google Colab T4 (15GB)	INT4	~6GB	✅
Apple M1/M2 (8GB+)	INT4	~6GB	✅

Architecture: Full CREST (K=4)

Input from Attention
       ↓
   ┌──────────┐
   │   MLP₁   │  ← Step 1: Core knowledge retrieval
   └──────────┘
       ↓
   Halting Gate → h₁ ≈ 0.99? → HALT (easy tokens stop here)
       ↓
   ┌──────────┐
   │   MLP₂   │  ← Step 2: Verification & refinement
   └──────────┘
       ↓
   Halting Gate → h₂ high? → HALT (medium tokens stop here)
       ↓
   ┌──────────┐
   │   MLP₃   │  ← Step 3: Logical reasoning
   └──────────┘
       ↓
   Halting Gate → h₃ high? → HALT
       ↓
   ┌──────────┐
   │   MLP₄   │  ← Step 4: Deep computation (hardest tokens only)
   └──────────┘
       ↓
   Output = p₁·s₁ + p₂·s₂ + p₃·s₃ + p₄·s₄

All 4 steps are preserved in this model. The halting gates dynamically allocate computation where it's needed.

Model Family

Model	Parameters	CREST K	Best For
Sky v2.0-11B	11.2B	4	Maximum capability
Sky v2.0-6.5B	6.47B	2	Balanced performance
Sky v2.0-11B-INT4	11.2B	4	Full CREST on consumer GPUs
Sky v2.0-Lite	6.47B	2	Laptops & edge devices

Citation

@article{jadav2026crest,
  title={CREST: Cognitively Recurrent Estimation of Step Termination for Adaptive-Depth Language Modeling},
  author={Jadav, Atharvsinh},
  year={2026},
  url={https://huggingface.co/0labs-in/Sky-v2.0-11B}
}

About 0labs

0labs is an independent AI research lab founded by Atharvsinh Jadav in Gujarat, India. We build adaptive-depth LLMs that think harder on hard problems — trained on a single GPU.

🌐 Website: 0labs.in
🤗 HuggingFace: huggingface.co/0labs-in

Downloads last month: 32

Safetensors

Model size

11B params

Tensor type

BF16

Model tree for 0labs-in/Sky-v2.0-11B-INT4

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Finetuned

0labs-in/Sky-v2.0-11B

Finetuned

(3)

this model