Skip to content
Language models

Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps

рассуждения о крикете в 6 раз короче

2 min readupdated

Это LoRA-адаптер (небольшая надстройка над базовой моделью) для языковой модели Qwen/Qwen3.8-27B от автора Balab2021, обученный решать математические и статистические задачи по крикету в формате T20. Модель училась методом RLVR (обучение с подкреплением на основе проверяемых наград) без готовых примеров рассуждений: она сама генерировала цепочки мыслей в режиме размышлений (thinking mode) и получала награду только за точный итоговый ответ. По сравнению с оригинальной базовой моделью адаптер поднимает общую точность с 97.00% до 99.96% и сокращает среднюю длину рассуждений с 734 до 119 токенов, полностью убирая обрывы генерации. Подойдёт разработчикам спортивной аналитики и исследователям, изучающим сжатие цепочек рассуждений через RL.

Ключевые особенности

  • Почти безошибочные расчёты: общая точность на отложенном тесте из 600 вопросов выросла с 97.00% до 99.96%, а в задачах на темп набора очков (run_rate) и вехи (milestone) достигла 100%.
  • Шестикратное ускорение вывода: среднее количество генерируемых токенов при рассуждении сократилось с 734 до 119 за счёт штрафа за длину в функции награды.
  • Нулевой процент обрывов: доля ответов, не уложившихся в лимит токенов, снизилась с 0.83% до 0.00%.
  • Обучение без дистилляции следов: модель не копировала чужие объяснения, а самостоятельно находила кратчайший путь к верному числу на датасете CricketData-T20-Reasoning-Qwen3.8-Full.

Что можно сделать

  • Встроить в чат-бота или сервис спортивной аналитики для быстрого расчёта темпа игры, шансов погони за счётом (chase_rate) и статистики матчей T20.
  • Использовать код и методику обучения (групповой policy gradient со штрафом за длину) как референс для оптимизации thinking-режима LLM в других узких предметных областях.

Пример

import torch
from peft import PeftModel
from transformers import AutoModelForImageTextToText, AutoTokenizer

tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.8-27B")
model = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen3.8-27B", dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(model, "Balab2021/Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps")
prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
ПодробнееБенчмарк (600 вопросов, 4 сэмпла на каждый): общая точность 99.96% (было 97.00%), chase_rate — 99.88% (было 93.62%), среднее число токенов — 119 (было 734).
Детали обучения: 120 шагов, 122 880 роллаутов, LoRA r=32, alpha=64 на проекциях attention, Gated-DeltaNet и MLP.
Железо при обучении: 1.8 часа на 32 GPU NVIDIA Vera Rubin.
Discuss Telegram 0 4

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 53 GB · ready files55 GB
  • FP8 / INT8recommendedweights 27 GB29 GB
  • 4-bitweights 15 GB18 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 Qwen3.8-27B.

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 RTX A6000

    $0.43 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.

    # Balab2021/Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps — RTX A6000, FP8 / INT8
    # ===== 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 Qwen/Qwen3.8-27B --max-model-len 8192 --quantization fp8 --enable-lora --max-lora-rank 64 --lora-modules Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps=Balab2021/Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps
    Add more LoRAs to Qwen3.8-27B
  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

Qwen/Qwen3.8-27B

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

New posts about versions of Qwen3.8-27B (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

Balab2021/Qwen3.8-27B-Cricket-Reasoning-RLVR-120steps
Author
Balab2021
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

40

Ask AI about this model

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

Comments