ErebusTN/EGen-V1

EGen V1

EGen V1 is a compact, fast conversational language model from EGen Labs, an independent AI research organization. It is designed for helpful, direct, general-purpose assistance question answering, writing, analysis, and coding help on modest hardware, including laptops and edge devices.

Model summary

Field	Value
Developer	EGen Labs
Model	EGen V1
Model ID	ErebusTN/EGen-V1
Type	Instruction-following chat model (causal language model)
Architecture	EGen hybrid: gated short-convolution blocks + grouped-query attention (GQA)
Parameters	~2.6B
Context window	Up to 128K tokens (architecture limit)
Primary language	English
Chat format	ChatML-style (`<
License	Apache-2.0

Highlights

• Small and efficient : runs comfortably on a single consumer GPU; GGUF builds run on CPU.
• Consistent identity : knows who it is and who built it, and answers identity questions reliably.
• System-prompt friendly : follows system instructions and multi-turn conversation context.
• Flexible deployment : supports Transformers, vLLM, llama.cpp, Ollama, and LM Studio.

Quickstart

Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ErebusTN/EGen-V1"

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

messages = [
    {"role": "system", "content": "You are a helpful and friendly customer support agent for [Business Name]. Your role is to answer questions about our products, handle billing inquiries, and guide users through troubleshooting steps. Always maintain a professional yet empathetic tone. If you cannot resolve an issue, escalate it to a human agent using [Escalation Steps]. Do not invent information; if unsure, ask clarifying questions.."},
    {"role": "user", "content": "What are you?"},
]

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

output = model.generate(
    **inputs,
    max_new_tokens=256,
    do_sample=True,
    temperature=0.1,
    top_k=50,
    top_p=0.1,
    repetition_penalty=1.05,
)

reply = tokenizer.decode(
    output[0][inputs["input_ids"].shape[-1]:],
    skip_special_tokens=True,
)

# The chat template opens an (empty) reasoning block; strip it before printing.
print(reply.split("</think>")[-1].strip())

llama.cpp / Ollama (GGUF)

Quantized GGUF builds are available at ErebusTN/EGen-V1-GGUF.

llama-cli -hf ErebusTN/EGen-V1-GGUF:Q4_K_M

or:

ollama run hf.co/ErebusTN/EGen-V1-GGUF:Q4_K_M

Note on the reasoning block

EGen V1's chat template opens a reasoning block (<think>) at the start of each reply. The model normally closes it immediately, so responses are direct. However, raw output may contain an empty <think></think> span. Strip everything up to and including </think> as shown above, or use a runtime that parses reasoning blocks natively—for example, llama.cpp with its reasoning-format option.

Recommended sampling

Parameter	Value
temperature	0.1
top_k	50
top_p	0.1
repetition_penalty	1.05

For more creative writing, raise temperature (for example, 0.3–0.7) and top_p.

Intended use

EGen V1 is intended for:

• Assistants and chat interfaces
• Drafting and summarization
• Lightweight coding help
• Research and prototyping as a small local model

Limitations

• Like all language models of this size, EGen V1 can produce incorrect, outdated, or fabricated information. Verify important facts.
• It is not a substitute for professional medical, legal, or financial advice and should not be used as the sole basis for high-stakes decisions.
• Performance is strongest in English; other languages may be weaker.
• Very long-context use should be validated for your own workload.

Safety and responsible use

EGen V1 has more relaxed refusal behavior than typical consumer assistant models. Anyone deploying it in a user-facing product is responsible for adding moderation, filtering, and usage policies suited to their audience and applicable law. Do not use the model to harm others or to violate the law.

License

Released under the Apache-2.0 license. See the full text in the LICENSE file in this repository.

Citation

@misc{egen_v1_2026,
  title        = {EGen V1},
  author       = {{EGen Labs}},
  year         = {2026},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/ErebusTN/EGen-V1}
}

Contact

Questions, feedback, and collaboration: reach out to EGen Labs through the profile on Hugging Face.