Skip to content
Other

CrispEmbed English spaCy

language bundle for native C++ runtime

2 min readupdated

This repository contains an English spaCy 3.8.0 pipeline bundle converted to the GGUF format for CrispEmbed's native C++ and ggml runtime. The model handles tokenization, part-of-speech (POS) tagging, morphological analysis, lemmatization, dependency parsing, and named entity recognition (NER) without needing Python in production. Built on the CrispEmbed library with v2 runtime support, the package requires explicit initialization via a C ABI and targets developers integrating NLP tasks into high-performance C/C++ applications.

Key features

  • Native runtime: Pure C/C++ implementation built on ggml with zero Python interpreter dependencies at runtime.
  • Full pipeline: Includes F32 weights for tok2vec, tagger, parser, attribute_ruler, lemmatizer, and ner components.
  • Integrity checks: Strict runtime verification of SHA-256 hashes and file sizes to prevent loading corrupted weights.
  • Exact parity: Matches original spaCy 3.8.18 outputs 100% on UD English-EWT test datasets.

What you can do

  • Integrate into C/C++ services: Embed fast linguistic text analysis directly into C++ backends without Python overhead.
  • Deploy on edge devices: Run a lightweight language model (around 12 MB) in resource-constrained environments.
  • Extract entities and syntax: Perform sentence parsing, named entity recognition, and lemmatization through a clean C ABI.

Example

Initializing the context and analyzing text in C:

crispembed_spacy_set_enabled(CRISPEMBED_SPACY_ENABLE_ANALYSIS);
ctx = crispembed_spacy_create_with_resources("bundle/model.gguf", MODEL_SHA256,
                                             "bundle/resources.json", RESOURCES_SHA256, err, sizeof err);
crispembed_spacy_analyze_utf8(ctx, text, len, &doc, NULL, NULL, NULL, NULL, err, sizeof err);
Read moreDetails
Size of model.gguf: ~12.7 MB, resources.json: ~3.4 MB.
Tests on UD English-EWT (2,077 texts, 25,531 tokens) showed 100% output parity with original spaCy.
Requires a CrispEmbed build with v2 runtime support (PR #244) and the matching language capsule.
License: MIT.
Discuss Telegram 0 58

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # cstr/crispembed-spacy-en-core-web-sm-3.8.0
    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 cstr/crispembed-spacy-en-core-web-sm-3.8.0 --local-dir ~/models/crispembed-spacy-en-core-web-sm-3.8.0
    # The model is in ~/models. How to run it is in the model card (link above).
  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

New posts about versions of crispembed-spacy-en-core-web-sm-3.8.0 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

cstr/crispembed-spacy-en-core-web-sm-3.8.0
Author
cstr
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

580

Ask AI about this model

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

Comments

Other

laya-cua

laya-cua