Hugging Face's logo

cstr
/
pixie-rune-v1-GGUF

Feature Extraction
GGUF
multilingual
embeddings
ggml
text-embeddings
xlm-r
crispembed
ollama
Model card Files
xet
 Community
cstr commited on
Commit
2fb50bc
·
verified ·
1 Parent(s): c7a82b6

Update README with Ollama instructions and quality metrics

Browse files
Files changed (1) hide show
  1. README.md +32 -39
README.md CHANGED
@@ -1,37 +1,45 @@
1
  ---
2
- license: apache-2.0
3
  language: [multilingual]
4
- tags: [embeddings, gguf, ggml, text-embeddings, xlm-r, crispembed]
5
  pipeline_tag: feature-extraction
6
  base_model: telepix/PIXIE-Rune-v1.0
7
  ---
8
 
9
  # pixie-rune-v1 GGUF
10
 
11
- GGUF format of [telepix/PIXIE-Rune-v1.0](https://huggingface.co/telepix/PIXIE-Rune-v1.0) for use with [CrispEmbed](https://github.com/CrispStrobe/CrispEmbed).
12
-
13
- PIXIE-Rune v1.0. 74-language embedding model, 1024-dimensional CLS-pooled.
14
 
15
  ## Files
16
 
17
  | File | Quantization | Size |
18
  |------|-------------|------|
19
- | [pixie-rune-v1-q4_k.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1-q4_k.gguf) | Q4_K | 437 MB |
20
- | [pixie-rune-v1-q8_0.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1-q8_0.gguf) | Q8_0 | 582 MB |
21
- | [pixie-rune-v1.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1.gguf) | F32 | 2171 MB |
22
 
 
23
 
24
  ## Quick Start
25
 
 
26
  ```bash
27
- # Download
28
- huggingface-cli download cstr/pixie-rune-v1-GGUF pixie-rune-v1-q4_k.gguf --local-dir .
 
29
 
30
- # Run with CrispEmbed
31
- ./crispembed -m pixie-rune-v1-q4_k.gguf "Hello world"
 
 
 
 
32
 
33
- # Or with auto-download
34
- ./crispembed -m pixie-rune-v1 "Hello world"
 
 
 
35
  ```
36
 
37
  ## Model Details
@@ -44,34 +52,19 @@ huggingface-cli download cstr/pixie-rune-v1-GGUF pixie-rune-v1-q4_k.gguf --local
44
  | Layers | 24 |
45
  | Pooling | CLS |
46
  | Tokenizer | SentencePiece |
47
- | Base Model | [telepix/PIXIE-Rune-v1.0](https://huggingface.co/telepix/PIXIE-Rune-v1.0) |
48
-
49
- ## Verification
50
 
51
- Verified bit-identical to HuggingFace sentence-transformers (cosine similarity >= 0.999 on test texts).
52
 
53
- ## Usage with CrispEmbed
54
-
55
- CrispEmbed is a lightweight C/C++ text embedding inference engine using ggml.
56
- No Python runtime, no ONNX. Supports BERT, XLM-R, Qwen3, and Gemma3 architectures.
57
-
58
- ```bash
59
- # Build CrispEmbed
60
- git clone https://github.com/CrispStrobe/CrispEmbed
61
- cd CrispEmbed
62
- cmake -S . -B build && cmake --build build -j
63
-
64
- # Encode
65
- ./build/crispembed -m pixie-rune-v1-q4_k.gguf "query text"
66
-
67
- # Server mode
68
- ./build/crispembed-server -m pixie-rune-v1-q4_k.gguf --port 8080
69
- curl -X POST http://localhost:8080/v1/embeddings \
70
- -d '{"input": ["Hello world"], "model": "pixie-rune-v1"}'
71
- ```
72
 
73
  ## Credits
74
 
75
  - Original model: [telepix/PIXIE-Rune-v1.0](https://huggingface.co/telepix/PIXIE-Rune-v1.0)
76
- - Inference engine: [CrispEmbed](https://github.com/CrispStrobe/CrispEmbed) (ggml-based)
77
- - Conversion: `convert-bert-embed-to-gguf.py`
 
1
  ---
2
+ license: mit
3
  language: [multilingual]
4
+ tags: [embeddings, gguf, ggml, text-embeddings, xlm-r, crispembed, ollama]
5
  pipeline_tag: feature-extraction
6
  base_model: telepix/PIXIE-Rune-v1.0
7
  ---
8
 
9
  # pixie-rune-v1 GGUF
10
 
11
+ GGUF format of [telepix/PIXIE-Rune-v1.0](https://huggingface.co/telepix/PIXIE-Rune-v1.0) for use with [CrispEmbed](https://github.com/CrispStrobe/CrispEmbed) and [Ollama](https://ollama.com).
 
 
12
 
13
  ## Files
14
 
15
  | File | Quantization | Size |
16
  |------|-------------|------|
17
+ | [pixie-rune-v1-q4_k.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1-q4_k.gguf) | Q4_K | 0 MB |
18
+ | [pixie-rune-v1-q8_0.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1-q8_0.gguf) | Q8_0 | 0 MB |
19
+ | [pixie-rune-v1.gguf](https://huggingface.co/cstr/pixie-rune-v1-GGUF/resolve/main/pixie-rune-v1.gguf) | F32 | 0 MB |
20
 
21
+ **Recommended:** Q8_0 for quality (cos vs HF: cross-lingual OK), Q4_K for size (cross-lingual OK).
22
 
23
  ## Quick Start
24
 
25
+ ### CrispEmbed
26
  ```bash
27
+ ./crispembed -m pixie-rune-v1 "Hello world"
28
+ ./crispembed-server -m pixie-rune-v1 --port 8080
29
+ ```
30
 
31
+ ### Ollama (with [CrispStrobe fork](https://github.com/CrispStrobe/ollama/tree/feat/xlmr-embedding))
32
+ ```bash
33
+ echo "FROM pixie-rune-v1-q8_0.gguf" > Modelfile
34
+ ollama create pixie-rune-v1 -f Modelfile
35
+ curl http://localhost:11434/api/embed -d '{"model":"pixie-rune-v1","input":["Hello world"]}'
36
+ ```
37
 
38
+ ### Python (CrispEmbed)
39
+ ```python
40
+ from crispembed import CrispEmbed
41
+ model = CrispEmbed("pixie-rune-v1-q8_0.gguf")
42
+ vectors = model.encode(["Hello world", "Goodbye world"])
43
  ```
44
 
45
  ## Model Details
 
52
  | Layers | 24 |
53
  | Pooling | CLS |
54
  | Tokenizer | SentencePiece |
55
+ | Language | multilingual |
56
+ | Q8_0 vs HuggingFace | cross-lingual OK |
57
+ | Q4_K vs HuggingFace | cross-lingual OK |
58
 
59
+ ## Server API
60
 
61
+ CrispEmbed server supports four API dialects:
62
+ - `POST /embed` -- native
63
+ - `POST /v1/embeddings` -- OpenAI-compatible
64
+ - `POST /api/embed` -- Ollama-compatible
65
+ - `POST /api/embeddings` -- Ollama legacy
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66
 
67
  ## Credits
68
 
69
  - Original model: [telepix/PIXIE-Rune-v1.0](https://huggingface.co/telepix/PIXIE-Rune-v1.0)
70
+ - Inference: [CrispEmbed](https://github.com/CrispStrobe/CrispEmbed) (MIT, ggml-based)