Skip to content
Other

sriq-clover-35B-A3B-v1.6

LoRA для Clover с китайским рассуждением

1 min read

Это PEFT LoRA-адаптер (не merged) для Diffuspace/Clover-35B-A3B-Preview — блочно-диффузионной MoE-модели. Адаптер от SRIQ учит рассуждать на сжатом упрощённом китайском, проверять себя через 验: и отвечать на языке вопроса.

Ключевые особенности

  • Рассуждение на китайском: 55 из 57 некитайских held-out вопросов.
  • Самопроверка 验: появляется в 54 из 64 трасс.
  • Ответ на языке вопроса — 64 из 64 примеров.
  • Обязателен system prompt: без него эффект файнтюна не проявляется.
  • Обучение: QLoRA r=32, α=64, эксперты в NF4, затем отфильтрованный SFT.
  • Точность ответов не бенчмаркалась — проверяйте на своей нагрузке.

Пример

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = 'Diffuspace/Clover-35B-A3B-Preview'
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, trust_remote_code=True, dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, 'kaivoss/sriq-clover-35B-A3B-v1.6').merge_and_unload()
messages = [{'role':'system','content':'You are a helpful assistant'},{'role':'user','content':'4 + 7 = ?'}]
ПодробнееНа 64 held-out примерах: 55/57 некитайских вопросов рассуждают только на китайском, 验: — 54/64, ответ на языке вопроса — 64/64.
Без system prompt адаптер рассуждает как базовая модель. В SGLang трассы смешивают языки; 1/64 зацикливается.
В bf16 нужно ~70 ГБ VRAM; для SGLang адаптер надо слить с базой, поддержка только BF16.
Discuss Telegram 0 20

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 65 GB · ready files72 GB
  • FP8 / INT8recommendedweights 32 GB36 GB
  • 4-bitweights 18 GB21 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

This is a LoRA: counted together with its base model Clover-35B-A3B-Preview.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # kaivoss/sriq-clover-35B-A3B-v1.6
    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 Diffuspace/Clover-35B-A3B-Preview --local-dir ~/models/Clover-35B-A3B-Preview
    hf download kaivoss/sriq-clover-35B-A3B-v1.6 --local-dir ~/models/sriq-clover-35B-A3B-v1.6
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Clover-35B-A3B-Preview
  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

Requires the base model

Diffuspace/Clover-35B-A3B-Preview

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. Qwen3.6-35B-A3Bfamily root
  2. Clover-35B-A3B-Preview
  3. sriq-clover-35B-A3B-v1.6

New posts about versions of Qwen3.6-35B-A3B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 4All on Hugging Face

About the model

kaivoss/sriq-clover-35B-A3B-v1.6
Author
kaivoss
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments