LoNebula/jervis-v1

Jervis v1

A bilingual (Japanese / English) assistant model fine-tuned from Qwen3-8B. This is the eighth model in the LoNebula workshop series and an evolution of the private "Jarvis" personal adapter: same warm, plain-spoken, think-out-loud personality, but packaged as a public, merged, ready-to-use chat model (no PEFT adapter required). It answers in whichever language you write in.

Model overview

Property	Details
Base model	Qwen3-8B (8.2B parameters, Apache-2.0)
Method	4-bit QLoRA (NF4 + double quant, bf16 compute), then merged into the base
Languages	Japanese, English
License	Apache-2.0 (adapter merged; subject to the base model license)
Author	LoNebula (寝坊ロケット / Nebou Rocket)
Weights format	Merged bfloat16, loads directly with transformers

Training data

This model is trained only on public data and contains no private or personal information.

• Aya Dataset (CohereLabs/aya_dataset): all Japanese and English instruction rows from the training split (the full dataset has ~202,362 examples across 65 languages; this model uses the JA/EN subset). Apache-2.0.
• Japanese instruction corpus (izumi-lab/llm-japanese-dataset): a capped sample of Japanese instruction/response pairs, used to strengthen Japanese instruction-following. CC BY-SA 4.0.
• Persona seed: a small, hand-written set of public-safe bilingual dialogue examples that establish the assistant's identity and tone. No personal facts, contacts, or private details are included.

The private, identity-specific corpus that personalizes the account owner lives only in the separate private "Jarvis Personal" repository and is never part of this public model.

Use

The repository contains merged weights, a tokenizer, and a generation_config.json, so it loads directly and generates with sensible defaults. Use a GPU with enough memory for the quantized or half-precision base.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "LoNebula/jervis-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 Jervis, a bilingual (Japanese/English) assistant. Answer in the user's language."},
    {"role": "user", "content": "難しいバグにぶつかったとき、どう進めればいい？"},
]
inputs = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt",
    chat_template_kwargs={"enable_thinking": False},  # Qwen3: non-thinking chat mode
).to(model.device)
output = model.generate(inputs, max_new_tokens=512)
print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True))

Install the runtime with pip install torch transformers accelerate. The generation_config.json sets temperature=0.7, top_p=0.8, top_k=20; override as needed. This model was fine-tuned with the Qwen3 chat template in non-thinking mode (enable_thinking=False), so pass the same chat_template_kwargs at inference; if your transformers version does not recognize the kwarg, drop it and the template falls back to its default.

Limitations

• This is an experimental assistant, not a benchmarked or production-ready system. It may be inaccurate, reproduce training text, or drift from its intended tone.
• It is a general bilingual assistant and does not carry the private, account-specific knowledge of the separate private Jarvis adapter.
• Japanese and English coverage depends on the public corpora above; specialized or low-resource domains may be weak.
• Review outputs before relying on them for important decisions.

License

Apache-2.0, subject to the base model license (Qwen3-8B, Apache-2.0) and the licenses of the public training data (Aya: Apache-2.0; izumi-lab Japanese corpus: CC BY-SA 4.0). Review source terms before redistribution.