Skip to content
Language models

cbx-rag-context-utilization-68M

fast RAG context utilization scorer

2 min readupdated

A compact model from CobuildX-AI that checks which exact sentences from the retrieved context an AI model actually used to generate its answer. You pass in a question, text passages, and the generated answer, and it outputs a score and flag for each sentence along with the overall utilization fraction (following the RAGBench standard). Built on the jhu-clsp/ettin-encoder-68m encoder (68M parameters), it belongs to a four-model lineup of specialized RAG (retrieval-augmented generation) evaluators. Unlike heavy LLM judges such as GPT-3.5, it runs locally in fractions of a second on a GPU and achieves a lower RMSE error on benchmarks. It is built for RAG developers looking to automate retrieval evaluation, debug chunking strategies, and filter out noisy context.

Key features

  • Ultra-lightweight architecture: only 68M parameters, running a single pass over several short passages in 0.12s on an NVIDIA TITAN RTX (fp16) or 3.04s on a CPU.
  • Sentence-level granularity: scores every context sentence individually, applies isotonic calibration to the probabilities, and computes the overall utilization fraction.
  • Adaptive profiles: supports a contract mode for long legal agreements and a financial mode for shorter financial reports packed with numbers and tables.
  • Smart prefilter: in the contract profile, selects the top 64 BM25-matched sentences and their neighbors, retaining 98.8% of actually used sentences while scanning only 39% of the text.

What you can do

  • Replace expensive LLM-as-a-Judge calls: plug it into your RAG evaluation CI/CD pipeline instead of commercial model APIs to compute Context Utilization on English documents.
  • Tune your retriever and chunk size: spot dead-weight chunks that consistently make it into the prompt but get ignored by the answer generator.
  • Audit legal and financial knowledge bases: verify assistant answers over contracts and financial filings using the built-in domain profiles.

Example

from cbx_rag_eval import ContextUtilization

judge = ContextUtilization(".")
out = judge(question, passages, answer)
print(out["utilization"])
used = [s for s in out["sentences"] if s["utilized"]]
Read moreDetails
Benchmarks (answer-level RMSE, lower is better): on CUAD contracts — 0.039 vs. 0.059 for GPT-3.5; on FinQA reports — 0.033 vs. 0.134 for GPT-3.5.
Sentence-level performance: AUROC 0.981 on CUAD (precision 0.78 / recall 0.60) and 0.995 on FinQA (precision 0.90 / recall 0.88).
Training and limitations: fine-tuned on RAGBench data (CUAD and FinQA) with LLM-generated labels; English only, Apache-2.0 license.
Discuss Telegram 0 12

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.3 GB · ready files0.9 GB
  • BF16weights 0.1 GB0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.7 GB
  • 4-bitweights 0.0 GB0.6 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

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

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 A30

    $0.26 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.

    # CobuildX-AI/cbx-rag-context-utilization-68M — A30, FP8 / INT8
    # 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 CobuildX-AI/cbx-rag-context-utilization-68M --max-model-len 8192 --quantization fp8
  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

New posts about versions of ettin-encoder-68m (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

CobuildX-AI/cbx-rag-context-utilization-68M
Author
CobuildX-AI
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

120

Ask AI about this model

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

Comments

Language models

SEL Qwen 9B

SEL Qwen 9B