Skip to content
Language models

Liminar-Qwen3.5-122B-A10B

a language model for Argentine academia

2 min read

Built on the Qwen 3.5 122B-A10B architecture and fine-tuned by the Argentine research lab LICDIA, this model handles local university regulations, official documents, and formal Argentine Spanish. It analyzes academic texts, answers questions on education law, and assists with specialized university tasks. The project targets universities, researchers, and edtech developers who need reliable processing of local legal and institutional texts in Spanish.

Key features

  • Adapted for Argentine regulations and university workflows
  • Built on the Qwen 3.5 122B-A10B large language model
  • Fine-tuned on curated datasets from official databases and university records
  • Delivers high accuracy when generating text in formal Spanish

What you can do

  • Deploy it in university support systems to answer student and staff questions on regulations
  • Pair it with RAG (retrieval-augmented generation) to accurately cite current laws and policies
  • Integrate it into academic platform chatbots to advise on internal university rules

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "LICDIA-UNLu/Liminar-Qwen3.5-122B-A10B"
tok = AutoTokenizer.from_pretrained(repo, revision="v2026.11")
modelo = AutoModelForCausalLM.from_pretrained(
    repo, revision="v2026.11", dtype="auto", device_map="auto")

mensajes = [{"role": "user",
             "content": "¿Qué establece la Ley de Educación Superior sobre la autonomía universitaria?"}]
entrada = tok.apply_chat_template(
    mensajes, add_generation_prompt=True, return_dict=True, return_tensors="pt").to(modelo.device)
salida = modelo.generate(**entrada, max_new_tokens=400)
print(tok.decode(salida[0, entrada["input_ids"].shape[1]:], skip_special_tokens=True))
Read moreDetails
• License: Apache 2.0
• Training method: LoRA (rank 256) on the Clementina XXI supercomputer using FSDP
• Benchmarks: closed-book test accuracy jumped from 8.3% (base model) to 31.7%
• Limitations: does not track legal changes after the dataset cutoff date and can make mistakes with regulation numbers and dates
Discuss Telegram 0

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • BF16weights 233 GB · ready files234 GB
  • GGUF Q8_0recommendedweights 123 GB · ready files125 GB
  • FP8 / INT8weights 116 GB118 GB
  • GGUF Q4_K_Mweights 71 GB · ready files72 GB
  • 4-bitweights 66 GB67 GB

Tip: go 8-bit.GGUF Q8_0 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.

Language models: with a 8192-token context.

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

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent 4× RTX A6000

    $1.71 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # LICDIA-UNLu/Liminar-Qwen3.5-122B-A10B — 4× RTX A6000, GGUF Q8_0
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -fsSL https://ollama.com/install.sh | sh
    ollama run hf.co/LICDIA-UNLu/Liminar-Qwen3.5-122B-A10B:Q8_0
  3. 3. Use it

    The chat opens right in the terminal. Leave with /bye, start again with the same ollama run command.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of Qwen3.5-122B-A10B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

LICDIA-UNLu/Liminar-Qwen3.5-122B-A10B
Author
LICDIA-UNLu
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Computer vision

Script2Play-4B

Script2Play-4B