Verus-4B

This repository contains model weights and configuration files for Verus-4B in the Hugging Face Transformers format.

Compatible with Hugging Face Transformers, vLLM, SGLang, llama.cpp (GGUF export), and other major inference frameworks.

Primary intended use cases are code generation, code review, debugging, and general coding assistance.

Verus-4B Highlights

Coding-First: Fine-tuned specifically on high-quality coding datasets — handles everything from simple scripts to complex multi-file implementations cleanly.
Image + Text Input: Accepts both images and text, allowing you to describe UIs, diagrams, or screenshots alongside code questions.
262K Token Context Window: Process entire codebases, long specifications, or lengthy conversations in a single pass.
Strong Instruction Following: Stays focused, responds clearly, and redirects to the task at hand.
Efficient: At 4B parameters in bfloat16, runs comfortably on a single consumer GPU with 8GB+ VRAM.

Model Overview

Property	Value
Parameters	~4B
Context Length	262,144 tokens
Architecture	Qwen3.5
Chat Format	ChatML (`<\|im_start\|>` / `<\|im_end\|>`)
Dtype	bfloat16
License	Apache 2.0

Quickstart

Installation

pip install "transformers>=4.52.0" accelerate torch

Code Generation

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

MODEL_ID = "8F-ai/Verus-4B"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model.eval()

messages = [
    {
        "role": "system",
        "content": "You are Verus, a coding assistant made by 8F-ai. You help with coding tasks and keep responses focused and clean."
    },
    {
        "role": "user",
        "content": "Write a Python async context manager that manages a PostgreSQL connection pool using asyncpg."
    }
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.inference_mode():
    generated_ids = model.generate(**inputs, max_new_tokens=2048, temperature=0.1, top_p=0.95)

output = tokenizer.decode(generated_ids[0][len(inputs.input_ids[0]):], skip_special_tokens=True)
print(output)

Image + Text Input

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

MODEL_ID = "8F-ai/Verus-4B"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model.eval()

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "path/to/screenshot.png"},
            {"type": "text", "text": "Convert this UI screenshot into a React component using Tailwind CSS."}
        ]
    }
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.inference_mode():
    generated_ids = model.generate(**inputs, max_new_tokens=2048, temperature=0.1, top_p=0.95)

output = tokenizer.decode(generated_ids[0][len(inputs.input_ids[0]):], skip_special_tokens=True)
print(output)

Quantized Inference (4-bit NF4, ~4 GB VRAM)

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
)

tokenizer = AutoTokenizer.from_pretrained("8F-ai/Verus-4B")
model = AutoModelForCausalLM.from_pretrained(
    "8F-ai/Verus-4B",
    quantization_config=quantization_config,
    device_map="auto",
)

Intended Use Cases

Use Case	Example
Code Generation	Write functions, classes, scripts in any language
Debugging	Identify and fix bugs from error messages or code
Code Review	Suggest improvements, catch issues, explain code
UI to Code	Convert screenshots or diagrams into working code
Long Context Codebase	Reason over entire repos up to ~200K tokens
General Q&A	Answer programming questions clearly and concisely

Limitations

English-Primary: Fine-tuning was conducted predominantly on English-language code and documentation.
Not for Math/Science: Not optimized for mathematical proofs or scientific computation.

Citation

@misc{verus4b2026,
  title        = {Verus-4B: A Coding-Focused Multimodal Language Model with 262K Context},
  author       = {8F-ai},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/8F-ai/Verus-4B}},
  note         = {Apache 2.0 License}
}