Instructions to use ZEROLABS1/zero-loop with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ZEROLABS1/zero-loop with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ZEROLABS1/zero-loop:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ZEROLABS1/zero-loop:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ZEROLABS1/zero-loop:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ZEROLABS1/zero-loop:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf ZEROLABS1/zero-loop:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf ZEROLABS1/zero-loop:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf ZEROLABS1/zero-loop:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ZEROLABS1/zero-loop:Q4_K_M

Use Docker

docker model run hf.co/ZEROLABS1/zero-loop:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use ZEROLABS1/zero-loop with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ZEROLABS1/zero-loop"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ZEROLABS1/zero-loop",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ZEROLABS1/zero-loop:Q4_K_M

Ollama
How to use ZEROLABS1/zero-loop with Ollama:
```
ollama run hf.co/ZEROLABS1/zero-loop:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use ZEROLABS1/zero-loop with Docker Model Runner:
```
docker model run hf.co/ZEROLABS1/zero-loop:Q4_K_M
```

Lemonade

How to use ZEROLABS1/zero-loop with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ZEROLABS1/zero-loop:Q4_K_M

Run and chat with the model

lemonade run user.zero-loop-Q4_K_M

List all available models

lemonade list

Atomic Chat

ZEROLABS1 commited on 25 days ago

Commit

ff9f179

verified ·

1 Parent(s): 20c8aea

Update README.md

Browse files

Files changed (1) hide show

README.md +4 -5

README.md CHANGED Viewed

@@ -20,7 +20,7 @@ tags:
 # ZERO LOOP
 **ZERO LOOP is a tiny foundation language model from Zero Labs, trained from scratch.**
-At 19.9M parameters it is built to run fully on-device: it thinks before it answers, can call a calculator and Wikipedia search, and is small enough to fine-tune on a free GPU in minutes.
 ## Highlights
@@ -54,7 +54,7 @@ At 19.9M parameters it is built to run fully on-device: it thinks before it answ
 |---|---|
 | Developer | Zero Labs |
 | Parameters | 19.9M |
-| Architecture | Decoder-only transformer, Qwen3 architecture in Transformers (grouped-query attention, RoPE, RMSNorm, SwiGLU, QK-norm, tied embeddings) |
 | Layers / hidden size / FFN size | 16 / 256 / 1024 |
 | Attention heads (query / key-value) | 4 / 2, head dimension 64 |
 | Vocabulary | 16,384 tokens, byte-level BPE, trimmed from the Qwen2/3 vocabulary to the tokens that matter for this model |
@@ -119,7 +119,7 @@ Available tools: `search` (argument `query`, Wikipedia) and `calculator` (argume
 ## Fine-tuning
-ZERO LOOP is small enough for full fine-tuning (no LoRA needed) on a single consumer GPU or a free Colab/Kaggle T4. Start from `base/` for new domains or languages, or from the chat weights to adapt behaviour.
 ```python
 from datasets import Dataset
@@ -150,7 +150,7 @@ To make your own GGUF, use llama.cpp `convert_hf_to_gguf.py`. The tokenizer uses
 ## Training
-**Tokenizer.** A byte-level BPE vocabulary trimmed from the Qwen2/3 vocabulary to the 16,384 tokens most used in the training text, plus chat and tool markers.
 **Pretraining (from random initialisation).** Next-token prediction on educational web text and synthetic textbook-style text (FineWeb-Edu and Cosmopedia from the SmolLM corpus), in several phases with cosine learning-rate schedules. Part of pretraining also used logit distillation from Qwen/Qwen3-1.7B-Base (soft targets restricted to this model's vocabulary). A final continued-pretraining phase mixes in conversational, reasoning and tool-call text so the base already knows the format. The checkpoint was only kept if held-out validation loss did not get worse.
@@ -163,7 +163,6 @@ To make your own GGUF, use llama.cpp `convert_hf_to_gguf.py`. The tokenizer uses
 The best checkpoint on a held-out validation set was kept, and it was published only if it beat the previous release on that set.
-**Compute.** Trained on free-tier Kaggle GPUs (2x T4) over several sessions.
 ## Evaluation

 # ZERO LOOP
 **ZERO LOOP is a tiny foundation language model from Zero Labs, trained from scratch.**
+At 19.9M parameters it is built to run fully on-device: it thinks before it answers, can call a calculator and Wikipedia search, and is small enough to fine-tune on a CPU
 ## Highlights
 |---|---|
 | Developer | Zero Labs |
 | Parameters | 19.9M |
+| Architecture | Decoder-only transformer,Architecture in Transformers (grouped-query attention, RoPE, RMSNorm, SwiGLU, QK-norm, tied embeddings) |
 | Layers / hidden size / FFN size | 16 / 256 / 1024 |
 | Attention heads (query / key-value) | 4 / 2, head dimension 64 |
 | Vocabulary | 16,384 tokens, byte-level BPE, trimmed from the Qwen2/3 vocabulary to the tokens that matter for this model |
 ## Fine-tuning
+ZERO LOOP is small enough for full fine-tuning (no LoRA needed) on a single consumer GPU  Start from `base/` for new domains or languages, or from the chat weights to adapt behaviour.
 ```python
 from datasets import Dataset
 ## Training
+**Tokenizer.** A byte-level BPE  vocabulary to the 16,384 tokens most used in the training text, plus chat and tool markers.
 **Pretraining (from random initialisation).** Next-token prediction on educational web text and synthetic textbook-style text (FineWeb-Edu and Cosmopedia from the SmolLM corpus), in several phases with cosine learning-rate schedules. Part of pretraining also used logit distillation from Qwen/Qwen3-1.7B-Base (soft targets restricted to this model's vocabulary). A final continued-pretraining phase mixes in conversational, reasoning and tool-call text so the base already knows the format. The checkpoint was only kept if held-out validation loss did not get worse.
 The best checkpoint on a held-out validation set was kept, and it was published only if it beat the previous release on that set.
 ## Evaluation