Instructions to use protoLabsAI/Agents-A1-MTP-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use protoLabsAI/Agents-A1-MTP-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="protoLabsAI/Agents-A1-MTP-GGUF",
	filename="Agents-A1-MTP-NVFP4.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use protoLabsAI/Agents-A1-MTP-GGUF 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 protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
llama cli -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
llama cli -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

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 protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
./llama-cli -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

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 protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
# Run inference directly in the terminal:
./build/bin/llama-cli -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Use Docker

docker model run hf.co/protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

LM Studio
Jan

vLLM

How to use protoLabsAI/Agents-A1-MTP-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "protoLabsAI/Agents-A1-MTP-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "protoLabsAI/Agents-A1-MTP-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Ollama
How to use protoLabsAI/Agents-A1-MTP-GGUF with Ollama:
```
ollama run hf.co/protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
```

Unsloth Studio

How to use protoLabsAI/Agents-A1-MTP-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for protoLabsAI/Agents-A1-MTP-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for protoLabsAI/Agents-A1-MTP-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for protoLabsAI/Agents-A1-MTP-GGUF to start chatting

How to use protoLabsAI/Agents-A1-MTP-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "protoLabsAI/Agents-A1-MTP-GGUF:NVFP4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use protoLabsAI/Agents-A1-MTP-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use protoLabsAI/Agents-A1-MTP-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "protoLabsAI/Agents-A1-MTP-GGUF:NVFP4" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use protoLabsAI/Agents-A1-MTP-GGUF with Docker Model Runner:
```
docker model run hf.co/protoLabsAI/Agents-A1-MTP-GGUF:NVFP4
```

Lemonade

How to use protoLabsAI/Agents-A1-MTP-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull protoLabsAI/Agents-A1-MTP-GGUF:NVFP4

Run and chat with the model

lemonade run user.Agents-A1-MTP-GGUF-NVFP4

List all available models

lemonade list

Agents-A1 — GGUF with MTP speculative decoding

llama.cpp builds of InternScience/Agents-A1 with an MTP draft head grafted in — A1 shipped without the mtp.* tensors its Qwen3.5-35B-A3B base carries, so no other A1 GGUF can do --spec-type draft-mtp. This one can: +46% measured, no separate draft model.

Files

file                        size     notes
--------------------------  -------  ------------------------------------------
Agents-A1-MTP-NVFP4.gguf    20.8 GB  NVFP4 experts/attention, Q8_0 trunk, MTP head
Agents-A1-MTP-Q8_0.gguf     37.8 GB  reference quality, MTP head

Measured (RTX A6000 48 GB, `-n 150` greedy)

config                       gen tok/s
---------------------------  ---------
NVFP4, no spec               127.7
NVFP4, --spec-type draft-mtp 187.0   (+46%, draft acceptance 0.62)

A 35B-class agentic MoE at 187 tok/s on a prosumer card in ~21 GB. The MTP head was grafted from the base Qwen3.5-35B-A3B — never re-trained on A1 — and acceptance holds anyway.

Blackwell (RTX PRO 6000 / RTX 50xx-class), MTP on, mean of 6 diverse prompts:

NVFP4 + draft-mtp:  305 tok/s  (287–336)

A 35B-A3B at the same speed our 9B runs — the NVFP4×MTP multiplication (verify-step batching feeds the FP4 tensor cores) reproduces on MoE. Finding writeup on the protoLabs Ornith cards.

Runtime compatibility

llama.cpp (spring-2026+), LM Studio, and recent Ollama (~0.31+): ✅. Older Ollama fails with "layer N missing attn_qkv" — update and re-pull. (Verified on the 9B sibling; if this 35B-MoE build misbehaves on your runtime, open a discussion with the error.)

Usage

llama-server -hf protoLabsAI/Agents-A1-MTP-GGUF:NVFP4 --spec-type draft-mtp -ngl 99

Requires llama.cpp with NVFP4 (type 40) + MTP support (both merged spring 2026). The quant tag resolves (:NVFP4 / :Q8_0) — filenames follow the standard convention.

Provenance & honesty notes

Converted from bf16 + grafted MTP tensors; NVFP4 layers quantized by llama-quantize (llama.cpp b9829) with the trunk (DeltaNet/router/norms) held at Q8_0.
The vLLM sibling Agents-A1-NVFP4 carries the full paired quality gate vs the official FP8 (matches or beats it on all four axes). This GGUF's own gate (paired vs official FP8, same judge/harness): FC 88.9% (identical to FP8) · coherence clean to 28K · reasoning 0.87 vs 0.84 at a 16K thinking budget · claw agentic paired-84: 0.60 vs 0.63.
Two honest caveats. (1) This build reasons noticeably longer than the vLLM artifact (different quant scales → different reasoning paths) — at an 8K generation budget it exhausts thinking before answering on some tasks; give it ≥16K for reasoning-heavy work or disable thinking for quick lookups. (2) One solver-graded sequence task regresses at any budget, and Chinese-language agentic tasks run ~0.03 softer — if your workload is zh-heavy agentic, prefer the vLLM build.
Not a vision build — text-only.

Need a different quant?

Open a Community discussion — requests usually ship within 48h. Rows: protoLabsAI/lab-benchmarks · protolabs.studio/lab.

Downloads last month: 1,987

GGUF

Model size

36B params

Architecture

qwen35moe

Hardware compatibility

4-bit

8-bit

Model tree for protoLabsAI/Agents-A1-MTP-GGUF

Base model

InternScience/Agents-A1

Quantized

(53)

this model

Quantizations

1 model

Agents-A1 — GGUF with MTP speculative decoding

Files

Measured (RTX A6000 48 GB, -n 150 greedy)

Runtime compatibility

Usage

Provenance & honesty notes

Need a different quant?

Model tree for protoLabsAI/Agents-A1-MTP-GGUF

Measured (RTX A6000 48 GB, `-n 150` greedy)