Rewnozom/Rewnozom-GGUF

Rewnozom

Rewnozom is a locally configured variant of Qwen/Qwen2.5-7B-Instruct-1M.

The model keeps the Qwen2 architecture and tokenizer compatibility, while the local configuration identifies the model as Rewnozom/Rewnozom and uses a custom default system prompt focused on senior software engineering behavior, correctness, stability, maintainability, performance, and direct high-signal answers.

Model Details

• Repository: Rewnozom/Rewnozom
• Base model: Qwen/Qwen2.5-7B-Instruct-1M
• Architecture: Qwen2ForCausalLM
• Model type: qwen2
• Context length config: 1,010,000 tokens
• Format: safetensors
• Library: transformers
• License: Apache 2.0, following the base model license

Intended Use

This model is intended for assistant-style text generation, especially:

• Software engineering assistance
• Logical reasoning
• Code review and implementation planning
• Technical writing
• Long-context analysis

It is designed to be used through the standard Hugging Face transformers chat-template flow.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Rewnozom/Rewnozom"

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

messages = [
    {"role": "user", "content": "Review this function for correctness and edge cases."},
]

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=512,
    temperature=0.63,
    top_p=0.8,
    top_k=15,
    min_p=0.03,
    repetition_penalty=1.05,
)

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

Default Generation Config

The included generation_config.json sets:

• do_sample: true
• temperature: 0.63
• top_p: 0.8
• top_k: 15
• min_p: 0.03
• repetition_penalty: 1.05

These settings are intended to keep responses fairly controlled while still allowing enough variation for useful assistant behavior. min_p is a decoding/runtime parameter and belongs in generation_config.json or the inference call; it is not stored in the .safetensors weight files.

For stricter code or reasoning output, top_p: 0.7 can be better because it narrows the token pool. For more open-ended writing or brainstorming, top_p: 0.9 can be better because it allows more alternatives. top_p: 0.8 is a reasonable default for this model card because it is balanced, especially together with temperature: 0.63, top_k: 15, and min_p: 0.03.

For deterministic code or evaluation workflows, lower temperature further or disable sampling.

System Prompt

The tokenizer chat template includes a default system prompt when the caller does not provide one. The prompt emphasizes:

• Senior software engineering judgment
• Correctness before speed
• Stability and maintainability
• Explicit tradeoffs and risks
• Concise, direct answers
• Avoiding invented facts, dependencies, APIs, or requirements

If your application passes its own system message, that message takes precedence over the default prompt.

Limitations

• This model can still produce incorrect or unsupported claims.
• Generated code must be reviewed and tested before production use.
• Long-context inference requires substantial memory and runtime resources.
• The default system prompt changes behavior, not the underlying model weights.
• Safety, licensing, and data-handling requirements remain the responsibility of the deployer.

Attribution

This model is based on Qwen/Qwen2.5-7B-Instruct-1M by Qwen. The base model is distributed under the Apache 2.0 license.

Base model page: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct-1M