Instructions to use zveloxy/Veloxy-Flash-15B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use zveloxy/Veloxy-Flash-15B with Transformers:

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

pipe = pipeline("text-generation", model="zveloxy/Veloxy-Flash-15B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("zveloxy/Veloxy-Flash-15B")
model = AutoModelForCausalLM.from_pretrained("zveloxy/Veloxy-Flash-15B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use zveloxy/Veloxy-Flash-15B 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 zveloxy/Veloxy-Flash-15B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf zveloxy/Veloxy-Flash-15B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf zveloxy/Veloxy-Flash-15B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf zveloxy/Veloxy-Flash-15B: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 zveloxy/Veloxy-Flash-15B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf zveloxy/Veloxy-Flash-15B: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 zveloxy/Veloxy-Flash-15B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf zveloxy/Veloxy-Flash-15B:Q4_K_M

Use Docker

docker model run hf.co/zveloxy/Veloxy-Flash-15B:Q4_K_M

LM Studio
Jan

vLLM

How to use zveloxy/Veloxy-Flash-15B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zveloxy/Veloxy-Flash-15B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zveloxy/Veloxy-Flash-15B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/zveloxy/Veloxy-Flash-15B:Q4_K_M

SGLang

How to use zveloxy/Veloxy-Flash-15B 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 "zveloxy/Veloxy-Flash-15B" \
    --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": "zveloxy/Veloxy-Flash-15B",
		"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 "zveloxy/Veloxy-Flash-15B" \
        --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": "zveloxy/Veloxy-Flash-15B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use zveloxy/Veloxy-Flash-15B with Ollama:
```
ollama run hf.co/zveloxy/Veloxy-Flash-15B:Q4_K_M
```
Unsloth Desktop

How to use zveloxy/Veloxy-Flash-15B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zveloxy/Veloxy-Flash-15B:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "zveloxy/Veloxy-Flash-15B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use zveloxy/Veloxy-Flash-15B with Docker Model Runner:
```
docker model run hf.co/zveloxy/Veloxy-Flash-15B:Q4_K_M
```

Lemonade

How to use zveloxy/Veloxy-Flash-15B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull zveloxy/Veloxy-Flash-15B:Q4_K_M

Run and chat with the model

lemonade run user.Veloxy-Flash-15B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use zveloxy/Veloxy-Flash-15B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zveloxy/Veloxy-Flash-15B:Q4_K_M

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 zveloxy/Veloxy-Flash-15B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use zveloxy/Veloxy-Flash-15B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zveloxy/Veloxy-Flash-15B:Q4_K_M

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 "zveloxy/Veloxy-Flash-15B:Q4_K_M" \
  --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"

⚡ Veloxy-Flash-15B: Faz 8 Tool-Call Master & Autonomous Reasoning Engine ⚡

📌 1. Overview

Veloxy-Flash-15B (Faz 8: Tool-Call Master) is a next-generation hybrid language model developed by Veloxy AI, engineered specifically for autonomous agent workflows, deep step-by-step reasoning (<think>), 38 native operating system & browser tools, enterprise-grade Java 21 LTS / Project Loom, and uncensored cybersecurity analysis.

Built upon an optimized hybrid architecture alternating between Linear-Attention and Full Multi-Head Attention layers, the model scales smoothly across a native 262,144-token context window with linear memory complexity. Designed for both enterprise cloud deployments and consumer hardware (such as NVIDIA RTX 3060 12GB, RTX 4060/4070, and RTX 5090), it delivers zero-latency, high-precision local execution.

🌟 2. Key Capabilities in Faz 8

Faz 8 Master training was conducted across 4,342 curated multi-turn enterprise samples on an NVIDIA RTX A6000 48GB cluster, achieving a record convergence loss of 0.2799.

🛠️ 1. 38 Native Full-OS & Browser Tools (`<tool_call>`)

The model natively produces strict JSON parameters for 38 system-level tools covering the terminal, filesystem, Git version control, DOM automation, and security diagnostics without requiring external wrappers.

🔍 2. Dynamic Unknown Tool Discovery

When faced with an unfamiliar command or custom user tool (veloxy_kernel_probe, custom_mcp_bridge, etc.), the model avoids hallucinations. It autonomously applies introspection strategies: passing --help, querying available schemas, and inferring valid arguments on the fly.

🔓 3. Zero-Refusal Cybersecurity & Binary Exploitation

Engineered for white-hat security researchers, penetration testers, and reverse engineers with technical depth and zero artificial refusals:

Binary Exploitation: ASLR/DEP evasion techniques, GOT/PLT overwriting, ROP chain synthesis, stack/heap canary analysis.
Web & Network Security: SAST source code auditing, CVE analysis, SSRF, insecure deserialization, SQLi, and IDOR vulnerability research with actionable remediation.

☕ 4. Elite Java 21 LTS & JVM Internals

Project Loom: Virtual Threads (Executors.newVirtualThreadPerTaskExecutor()), Structured Concurrency, and Scoped Values for high-throughput concurrency.
JVM Internals: ZGC (Generational), Shenandoah, and G1 GC mechanics, JIT C1/C2 compilation stages, Java Memory Model (JMM) barriers, and bytecode disassembly.
Spring Boot 3: Reactive WebFlux pipelines, GraalVM AOT Native Image optimizations.

⏳ 5. 2026 Temporal Anchor

The model’s temporal baseline is strictly fixed to 2026. It assumes modern 2026 ecosystem baselines (Java 21/22, Next.js 15, React 19, Python 3.12+, PyTorch 2.6) by default and does not hallucinate outdated legacy patterns.

🧠 6. Scalable Reasoning Depth (Reasoning Effort)

Reasoning depth can be dynamically controlled via prompts across 5 tiers:

[EFFORT: LOW] – Fast, direct responses.
[EFFORT: MEDIUM] – Balanced daily reasoning.
[EFFORT: HIGH] – In-depth verification and step-by-step math.
[EFFORT: XHIGH] – Olympiad math, complex algorithm design, and proofs.
[EFFORT: MAX] – Full architectural breakdown, root-cause analysis, and exhaustive code audits.

🧰 3. 38 Native Tool Catalog

Category	Native Tools	Description
System & Terminal	`bash_exec`, `terminal_spawn`, `process_list`, `process_kill`, `service_status`, `service_restart`, `cron_schedule`	Complete OS process and daemon management
Filesystem	`file_read`, `file_write`, `file_diff_patch`, `file_search`, `dir_tree`, `dir_create`, `file_delete`	Atomic reads, writes, patches, and directory trees
Developer & Git	`git_status`, `git_diff`, `git_commit`, `git_branch`, `code_ast_parse`, `linter_run`, `test_runner`	Autonomous version control, AST parsing, and testing
Browser & Web	`browser_open`, `element_click`, `input_type`, `page_extract`, `screenshot_capture`, `network_har_inspect`	Full-DOM interaction, screenshotting, and HAR analysis
Memory & Agents	`memory_store`, `memory_retrieve`, `session_summary`, `task_decompose`, `subagent_invoke`	Long-term memory storage and subagent delegation
Security & Diagnostics	`cve_search`, `binary_disasm`, `network_port_scan`, `sast_code_audit`, `heap_inspector`	Vulnerability assessment, disassembly, and port scanning

📦 4. Multi-Quant GGUF Family

All primary GGUF quantizations are available directly in this repository for local use:

Quantization	File Size	Required VRAM	Recommended Hardware	Quality / PPL	Recommended Use Case
`veloxy-flash-15b-Q8_0.gguf`	9.53 GB	~12-16 GB	RTX 4080, RTX 3090, RTX 4090	100% (Pristine 8-bit)	Maximum precision benchmark reference
`veloxy-flash-15b-Q6_K.gguf`	6.85 GB	~9-12 GB	RTX 3060 12GB, RTX 4070	99.8%	High-fidelity enterprise production
`veloxy-flash-15b-Q5_K_M.gguf`	6.02 GB	~8-10 GB	RTX 3060 12GB, RTX 4060 Ti 16GB	99.2%	Balanced speed and memory efficiency
`veloxy-flash-15b-Q4_K_M.gguf` ⭐	5.24 GB	~6-8 GB	NVIDIA RTX 3060 12GB	98.5% (Recommended)	100% GPU offload on 12GB VRAM cards
`veloxy-flash-15b-Q3_K_M.gguf`	4.40 GB	~6 GB	RTX 2060 6GB, RTX 3050	96.0%	Lower VRAM environments
`veloxy-flash-15b-Q2_K.gguf`	3.83 GB	~4-6 GB	GTX 1660, Apple Silicon 8GB	91.5%	Ultra-light edge computing devices

💡 Notice for RTX 3060 12GB Users: The Q4_K_M quant consumes only ~5.6 GB VRAM in memory, leaving ~6.5 GB free headroom for system and KV-Cache.

(Community i-Matrix / IQ quants are also provided by mradermacher/Veloxy-Flash-15B-i1-GGUF.)

🚀 5. Quickstart & Deployment Guide

A. Run with Ollama (Recommended for Local PC)

Download veloxy-flash-15b-Q4_K_M.gguf into your directory.
Create a Modelfile:

FROM ./veloxy-flash-15b-Q4_K_M.gguf

PARAMETER num_ctx 262144
PARAMETER stop "<|im_end|>"
PARAMETER stop "<|endoftext|>"
PARAMETER temperature 0.6
PARAMETER top_p 0.9

SYSTEM """You are Veloxy-Flash-15B, an advanced 15-billion parameter artificial intelligence model developed by Veloxy AI.
- Capabilities: Deep reasoning (<think>), 38 native OS/Browser tools, Java 21+ and JVM internals, and zero-refusal cybersecurity research.
- Temporal Anchor: Current year is 2026.
- Reasoning Effort: Scale your thinking inside <think> tags according to task complexity."""

Build and launch:

ollama create veloxy-flash-15b -f Modelfile
ollama run veloxy-flash-15b

B. Python / Transformers (Native bfloat16 & 4-bit)

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

model_id = "zveloxy/Veloxy-Flash-15B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

# 4-bit quantization for 12GB GPUs (RTX 3060 / 4060):
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True
)

messages = [
    {"role": "system", "content": "You are Veloxy-Flash-15B, an autonomous reasoning and systems engineering agent developed by Veloxy AI."},
    {"role": "user", "content": "Design an eBPF program in C that hooks onto sys_enter_execve to detect suspicious reverse shell invocations."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([prompt], return_tensors="pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.6)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

C. llama.cpp Server (OpenAI-Compatible Local API)

llama-server \
  -m veloxy-flash-15b-Q4_K_M.gguf \
  -c 32768 \
  -ngl 99 \
  --port 8080 \
  --host 0.0.0.0

🛠️ 6. Tool-Calling Format Example

<|im_start|>user
Inspect port 8443 on the staging server and read the nginx reverse proxy configuration.
<|im_end|>
<|im_start|>assistant
<think>
1. Check port 8443 status using network_port_scan.
2. Read the nginx config file at /etc/nginx/conf.d/proxy.conf.
</think>
<tool_call>
<function=network_port_scan>
<parameter=host>192.168.1.50</parameter>
<parameter=ports>[8443]</parameter>
</function>
</tool_call>
<tool_call>
<function=file_read>
<parameter=path>/etc/nginx/conf.d/proxy.conf</parameter>
</function>
</tool_call><|im_end|>

🌐 7. Veloxy AI Ecosystem

Model	Format / Parameters	Focus & Target Hardware	Link
Veloxy-Flash-15B ⭐	15B BF16 + GGUF Family	Faz 8 Tool-Call Master, 38 Native Tools, RTX 3060 12GB & Edge	HF Repo
Veloxy-Flash-15B-LoRA 🧩	PEFT Rank 16 Adapter	Official Faz 8 Integration Adapter Weights	HF Repo
Veloxy-Titan-32B 🛸	32B BF16 / 1M Context	Flagship Deep Reasoning, 1M Context, H100 NVL Benchmark	HF Repo
Veloxy-Titan-32B-Vision 👁️	32B + 27-Layer ViT	Multimodal Vision & Reasoning, UI-to-Code, Schematics	HF Repo

📄 8. License

Veloxy-Flash-15B is released under the Apache License 2.0. Free for academic, commercial, and private deployment.

Downloads last month: 721

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for zveloxy/Veloxy-Flash-15B

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Quantized

(566)

this model

Adapters

1 model

Quantizations

2 models

Evaluation results

Final Train Loss on Faz 8 Master Enterprise Benchmark
self-reported

0.280
Tool-Calling Schema Precision on Faz 8 Master Enterprise Benchmark
self-reported

99.4%
Context Window Length on Faz 8 Master Enterprise Benchmark
self-reported

262144.000

⚡ Veloxy-Flash-15B: Faz 8 Tool-Call Master & Autonomous Reasoning Engine ⚡

📌 1. Overview

🌟 2. Key Capabilities in Faz 8

🛠️ 1. 38 Native Full-OS & Browser Tools (<tool_call>)

🔍 2. Dynamic Unknown Tool Discovery

🔓 3. Zero-Refusal Cybersecurity & Binary Exploitation

☕ 4. Elite Java 21 LTS & JVM Internals

⏳ 5. 2026 Temporal Anchor

🧠 6. Scalable Reasoning Depth (Reasoning Effort)

🧰 3. 38 Native Tool Catalog

📦 4. Multi-Quant GGUF Family

🚀 5. Quickstart & Deployment Guide

A. Run with Ollama (Recommended for Local PC)

B. Python / Transformers (Native bfloat16 & 4-bit)

C. llama.cpp Server (OpenAI-Compatible Local API)

🛠️ 6. Tool-Calling Format Example

🌐 7. Veloxy AI Ecosystem

📄 8. License

Model tree for zveloxy/Veloxy-Flash-15B

Evaluation results

🛠️ 1. 38 Native Full-OS & Browser Tools (`<tool_call>`)