Skip to content
Computer vision

Novalyte-v2-27B

Qwen3.8-based conversational model

2 min readupdated

The new Novalyte-v2-27B model is a full-size merged checkpoint built on the Qwen3.8-27B architecture. It is fine-tuned using SFT (supervised fine-tuning) for a specific identity and conversational style while maintaining the strong baseline capabilities in Chinese and English. Weights are provided in Safetensors format matching the original structure, so no extra LoRA adapters are required to run it. This option suits researchers and developers exploring how fine-tuning affects the behavior of large language models.

Key features

  • Base model: Built on Qwen3.8-27B with around 27.78 billion parameters in BF16 format.
  • Training method: Fine-tuned via QLoRA using the LLaMA-Factory framework on a specialized identity dataset.
  • Open license: Distributed under the permissive Apache-2.0 license.
  • Transparent benchmarks: The authors honestly note that version two trails version one in metrics and publish the comparative results.

What you can do

  • Local deployment: Run the model locally on suitable hardware to test conversational abilities and text generation.
  • Pipeline integration: Use the weights as a ready starting point for your own language model fine-tuning experiments.
  • Quality analysis: Review C-Eval, MMLU, and CMMLU benchmark results to evaluate score drops or gains after custom SFT.

Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_id = "voblcn/Novalyte-v2-27B"
tok = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                         bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16)
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True,
                                            quantization_config=bnb, device_map="auto")
prompt = "<|im_start|>user\n你是谁?<|im_end|>\n<|im_start|>assistant\n"
ids = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**ids, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0][ids["input_ids"].shape[1]:], skip_special_tokens=True))
Read moreDetails
• Weight format: Safetensors (18 shards, around 51.75 GiB of data).
• Hardware requirements: Full BF16 precision requires approximately 52 GiB VRAM/RAM, while 4-bit quantization is available for running on 24–32 GiB GPUs.
• Metrics: On a subset of 433 tests, the base model scores 74.8%, v1 scores 80.4%, and the current v2 scores 77.6%.
• Important note: The authors recommend using the previous v1 version for practical use due to better stability.
Discuss Telegram 0 11

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # voblcn/Novalyte-v2-27B
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download voblcn/Novalyte-v2-27B --local-dir ~/models/Novalyte-v2-27B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Novalyte-v2-27B
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Novalyte-v2-27B (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

voblcn/Novalyte-v2-27B
Author
voblcn
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

110

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments