Skip to content
Language models

Fable

LoRA для разбора инцидентов 5G SA

2 min readupdated

Fable — это LoRA-адаптер (небольшая надстройка над базовой моделью) для Mistral-Small-4-119B-2603: MoE-модели на 119B параметров, из которых активны 6.5B. Адаптер дообучен под задачи 5G Standalone: архитектуру ядра 3GPP Rel-17/18, поиск корневой причины сбоев в RAN, Transport и Core, а также операции в суверенном телеком-облаке. На выходе — разбор инцидента или конфигурации с локализацией причины, отсылками к cause codes и таймерам, а при нехватке телеметрии модель должна воздержаться от ответа. Пригодится инженерам NOC, архитекторам 5G Core и тем, кто автоматизирует runbooks; для запуска нужна базовая 119B-модель, так что это не вариант для слабого железа.

Ключевые особенности

  • LoRA ранга r=32, alpha=64 в BF16; адаптер весит 219.8 MB и ставится на все 288 проекций MLA и shared-expert MLP в 36 слоях.
  • Обучение на 240 сценариях, сгенерированных детерминированным Python-солвером по физике 3GPP Rel-17/18, link budget, Erlang-C и state machines; без дистилляции из внешних frontier-LLM.
  • Loss на обучении упал с 1.2810 до 0.0365 (-97.15%), token accuracy — с 67.9% до 98.98%.
  • На 30 unseen-сценариях общий rubric score 85.22% против 34.94% у базы и 59.94% у базы с 6.5k-токенным playbook.
  • Блок обязательного 5-блочного NOC-ответа соблюдён в 100% случаев; epistemic abstention — тоже 100%.

Что можно сделать

  • Разбирать инцидент 5G SA по телеметрии и PCAP: локализовать проблему на уровне RAN, Transport или Core и предложить root cause.
  • Сверять signaling, cause codes и timers 3GPP Rel-17/18 в NOC-runbook'ах и проверять ответы на соответствие обязательной структуре.
  • Считать link budget, PRACH, Erlang-C и dimensioning UPF/eBPF/SR-IOV с проверкой числовых допусков.
  • Оценивать blast radius и chaos-сценарии для суверенного телеком-облака; если данных не хватает — модель должна сказать об этом, а не выдумывать.

Пример

import torch
from transformers import AutoModelForCausalLM
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained(
    'mistralai/Mistral-Small-4-119B-2603',
    torch_dtype=torch.bfloat16, device_map='auto')
model = PeftModel.from_pretrained(base, 'AIHeikki/fable-mistral-small-4-119b-telco-lora')
ПодробнееБенчмарк: overall rubric 85.22%, exact numerical math ±1% — 72.50%, 3GPP recall — 82.22%, blind judge — 9.52/10.
Pairwise против базы: 29W / 0T / 1L; против RAG+Prompt: 28W / 1T / 1L.
Обучение: 8x NVIDIA H100 80GB HBM3 (640 GB VRAM), 26.2 минуты; адаптер 219.8 MB.
Ограничения: узкая телеком-доменность и необходимость держать базовую 119B MoE; лицензия Apache 2.0.
Discuss Telegram 0 19

Will it run?

Won't run

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

  • F8_E4M3recommendedweights 113 GB · ready files125 GB
  • 4-bitweights 63 GB70 GB

Tip: go 8-bit.F8_E4M3 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: Won't fit any home GPU — renting is cheaper.

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Mistral-Small-4-119B-2603.

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.

    # AIHeikki/fable-mistral-small-4-119b-telco-lora — 4× RTX A6000, F8_E4M3
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    # 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${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve mistralai/Mistral-Small-4-119B-2603 --max-model-len 8192 --tensor-parallel-size 4 --enable-lora --max-lora-rank 64 --lora-modules fable-mistral-small-4-119b-telco-lora=AIHeikki/fable-mistral-small-4-119b-telco-lora
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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

Requires the base model

mistralai/Mistral-Small-4-119B-2603

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

New posts about versions of Mistral-Small-4-119B-2603 (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

About the model

AIHeikki/fable-mistral-small-4-119b-telco-lora
Author
AIHeikki
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

190

Ask AI about this model

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

Comments