- Meivin Embed 0.1
- The Meivin family | ืกืืจืช ืืืืืื
- English overview
- ืืืื ืืื
- ืืชืืื ืืืืจื
- Regular weights: PyTorch / Safetensors
- ืืืื ืฉืืืื
- ืืงืืจ ืืืจืืืืงืืืจื
- ืืขืจืื ืืืืืืืช
- ืืืืืืืช ืืชืืื ืืช
- ืงืืืฆื ืืืืืจ
- ืื ืืืื?
- ืจืืฉืื ืืืืืืก
- ืชืืขืื ืืืืืื
- Languages and textual coverage | ืฉืคืืช ืื ืืื ืืฉืื
- Downloads and parameter display | ืืืจืืืช ืืคืจืืืจืื
- Citation
- The Meivin family | ืกืืจืช ืืืืืื
Meivin Embed 0.1
ืืืืขืืช ืกืื ืืืืช ืืืืคืืฉ ืืกืคืจืืช ืชืืจื ืืช ืืขืืจืืช ืืืืจืืืช โ ืืืฉืงืืื ืจืืืืื ืืึพONNX ืืืจืฆื ืืงืืืืช.
Meivin Embed ืืืืจ ืฉืืืืช ืืืงืืขื ืืงืกื ืืืืงืืืจืื ืฉื 256 ืืืืื, ืืื ืืืคืฉืจ ืืืคืืฉ ืืคื ืืฉืืขืืช ืื ืืืฉืจ ื ืืกืื ืืฉืืื ืฉืื ื ืืืฉืื ืืืงืืจ. ืืื ืืืืขื ืืืคืชืื ืื ืืขื ืืืคืืฉ ืชืืจื ืืื, ืกืคืจืืืช ืืืืืืืืืช ืืืขืจืืืช ืืืืืจ ืืงืืจืืช ืขืืืจ RAG.
ืืืืืื ืืืืืช ืืฉืงืืื ืจืืืืื ืึพSafetensors, ืืจืกืช ONNX FP32 ืืืจืกืช ONNX INT8 ืงืืืคืงืืืช. ืืืืจ ืืืจืืช ืืงืืฆืื ืืคืฉืจ ืืงืืื ืืงืกื ืืงืืืืช ืืืืฆืขืืช ONNX Runtime. ืืืืื ืืคืืง ืืงืืืจืื; ืืฆืืจืช ืืื ืืงืก, ืืืคืืฉ ืื ืืืคืงืช ืชืฉืืื ืืงืกืืืืืืช ื ืขืฉืื ืืจืืืืื ืืืจืื ืฉื ืืืืฉืื.
ืื ืฉืืืืฉ ืืกืืจื ืืกืืจ ืืื ืืืฉืืจ ืืคืืจืฉ ืืจืืฉ ืืืืชื ืืืจืืื ืื ืืืื โ arieldaniely@gmail.com. ืืกืืจ ืืฉืื ืื ืืืฉืชืืฉ ืืืืื ืืื ืฆืืจื ืืืขืจืืช, ืชืืื ื ืื ืฉืืจืืช ืืืืจืืช ืจืืื, ืื ืื ืืจืืื ืฉืื ืืื ืืืืืข ืืื ืื. ืจืื ืืจืืฉืืื ืืืื.
The Meivin family | ืกืืจืช ืืืืืื
| Model | Purpose | Output |
|---|---|---|
| Meivin Embed 0.1 | Semantic retrieval for Judaic libraries and RAG source retrieval | L2-normalized 256-dimensional embeddings |
| Meivin MLM 0.1 | Contextual masked-token prediction and language research | Scores over a 32,000-token vocabulary |
Built for the language of Jewish texts, with compact models that run locally. The two products share a documented training foundation and serve complementary tasks. This release is available under Meivin Non-Commercial License 1.0; commercial licensing and collaboration: Ariel Daniely.
English overview
A compact Hebrew and Aramaic encoder for semantic search in Judaic literature. It maps queries and passages to normalized 256-dimensional vectors, runs locally, and ships as complete PyTorch/Safetensors weights, ONNX FP32 and ONNX INT8. The regular checkpoint contains 42,260,993 parameters; the ONNX files are approximately 168.2 MB and 42.5 MB respectively. Use [QUERY] and [PASSAGE] prefixes with the included tokenizer. Full training report documents the training lineage and retrieval evaluations.
ืืืื ืืื
| ืืืคืืื | ืขืจื |
|---|---|
| ืชืืื | ืกืคืจืืช ืชืืจื ืืช ืืขืืจืืช ืืืืจืืืช |
| ืคืื | ืืงืืืจ ืฉื 256 ืืืืื, ืื ืืจืื L2 |
| ืคืืจืื | Safetensors FP32; ONNX opset 17 |
| ืืจืกืืืช | PyTorch Safetensors FP32, โONNX FP32, โONNX INT8 |
| ืคืจืืืจืื | 42,260,993 ืืืืืืช ืืืฉืงืืื ืืืืื |
| ืืืื ืงืืืฅ ืืืืื | FP32: ืึพ168.2 MB; INT8: ืึพ42.5 MB, ืืืืืืืช ืขืฉืจืื ืืืช |
| ืงืื | ืจืฆืฃ ืืื ืืื ืงืจืืื, ืืืืจื ืืฉืชื ื |
| ืชืงืจืช ืืืจืืืืงืืืจื | 512 ืืืงื ืื, ืืืื ืงืืืืืช ืืืืงื ืื ืืืืืืื |
| ืืืจื ืืืืืืช ืืฉืืืืฉ | 256 ืืืงื ืื |
| ืืืฉื | ืืงืฉืช ืืืฉื ืืืืฉืืจ ืืื ื ืึพHugging Face |
| ืจืืฉืืื ืืจืืืคืงืื ืืืืื | Meivin Non-Commercial License 1.0; ืฉืืืืฉ ืืกืืจื ืจืง ืืืืฉืืจ ืืืชื |
0.1 ืืื ืืจืกืช ืืืืฆืจ ืืืชืืขืืช ืืืจืืืก ืื. ืืืฉืงืืืืช ืืืืกืกืืช ืขื checkpoint ืืืืืืจ Round 2; ืืืืชืื ืืื ื ืืฆืืื ืืืืื ืืืฉ. ืืืืืืืืช ืืฉืชืืฉืืช ืืฉื ืืืืืจ ืืจืฉืื Meivin-Embed-0.1.
ืืชืืื ืืืืจื
ืืงืฉื ืืืฉื ืึพืขืืื ืืืืื. ืืืืจ ืืืฉืืจื, ืืชืืืจื ืืืฉืืื ืืขื ืืืจืฉืื:
pip install numpy onnxruntime tokenizers huggingface_hub
hf auth login
ืืืืืื ืืงืืืืช ืฉืืื ืืฉื ื ืืงืืขื ืืืืื ืืืืจืืช ืืืชื ืืคื ืืืืื ืงืืกืื ืืก:
import numpy as np
import onnxruntime as ort
from huggingface_hub import hf_hub_download
from tokenizers import Tokenizer
REPO = "ArieLLL123/Meivin-Embed-0.1"
MODEL_FILE = "seforim-embed-round2-int8.onnx"
import json
config_path = hf_hub_download(REPO, "config.json", token=True)
config = json.load(open(config_path, encoding="utf-8"))
model_path = hf_hub_download(REPO, MODEL_FILE, token=True)
tokenizer_path = hf_hub_download(REPO, "tokenizer.json", token=True)
tokenizer = Tokenizer.from_file(tokenizer_path)
tokenizer.enable_truncation(max_length=config["query_max_length"])
session = ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])
def encode(text: str, role: str) -> np.ndarray:
prefixes = {"query": "[QUERY]", "passage": "[PASSAGE]"}
tokens = tokenizer.encode(f"{prefixes[role]} {text}", add_special_tokens=True)
inputs = {
"input_ids": np.asarray([tokens.ids], dtype=np.int64),
"attention_mask": np.asarray([tokens.attention_mask], dtype=np.int64),
}
return session.run(["embedding"], inputs)[0][0]
query = encode("ืื ืืฉืืขืืช ืืืืืื ืืืจ ืฉืืื ื ืืชืืืื?", "query")
passages = [
"ืืืจ ืฉืืื ื ืืชืืืื ืืื ืืขืฉื ืฉืชืืฆืืชื ืืืกืืจื ืืื ื ืืืื ืช ืืขืืฉื.",
"ืืฆืืืช ืกืคืืจืช ืืขืืืจ ื ืืืืช ืืื ืืื ืืืื ืืกืคืืจื.",
]
vectors = np.stack([encode(text, "passage") for text in passages])
scores = vectors @ query # ืืืืงืืืจืื ืื ืืจืืืื, ืืืคืื ืกืงืืจืืช ืฉืืื ืืงืืกืื ืืก
for index in np.argsort(-scores):
print(f"{scores[index]:.4f}\t{passages[index]}")
ืืฉืืืืฉ ืึพFP32, ืฉื ื ืืช MODEL_FILE ืึพseforim-embed-round2-fp32.onnx. ืงืืืื ืฉืืืืช ืืืงืืขื ืืื ืืงืก ืืืืชื ืืจืกืช ืงืืืฅ ืืืื ืืืืืชื ืืืงื ืืืืจ.
Regular weights: PyTorch / Safetensors
model.safetensors contains the complete trained sentence encoder: BERT backbone, projection, LayerNorm and the stored training logit scale. A plain AutoModel loader is not sufficient for this custom sentence encoder: it would omit the retrieval head. Use the small explicit loader below after reviewing meivin_embed.py; no trust_remote_code=True is needed.
pip install torch transformers safetensors huggingface_hub numpy
Download the weights and loader at one pinned revision (see RELEASE_NOTES.md), then run:
from huggingface_hub import snapshot_download
import importlib.util
from transformers import AutoTokenizer
folder = snapshot_download("ArieLLL123/Meivin-Embed-0.1", token=True,
allow_patterns=["config.json", "model.safetensors", "tokenizer*", "special_tokens_map.json", "meivin_embed.py"])
spec = importlib.util.spec_from_file_location("meivin_embed", folder + "/meivin_embed.py")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module) # reviewed repository helper
model = module.MeivinEmbed.from_directory(folder)
tokenizer = AutoTokenizer.from_pretrained(folder)
query = module.encode(model, tokenizer, ["ืืื ืืืจ ืฉืืื ื ืืชืืืื?"], role="query")
passages = module.encode(model, tokenizer, ["ืืื ื ืฉืืช ืืืืืื", "ืกืคืืจืช ืืขืืืจ"], role="passage")
print(passages @ query[0])
The explicit loader returns normalized vectors and supports padded batches. Regular weights retain gradients for further training; encode is an inference helper. Existing ONNX files retain their single-sequence input contract. PyTorch and ONNX FP32 were compared using short inputs and a 512-token input; see validation.
ืืืื ืฉืืืื
- ืืืกืืคื
[QUERY]ืืคื ื ืฉืืืืช ืึพ[PASSAGE]ืืคื ื ืืงืืขืื, ืืคื ื ืืืืงื ืืืฆืื. ืืื ืืืงื ืื ืืืืืืื ืฉื ืืืื ืืืืืื. - ืืฉืชืืฉื ืืืืงื ืืืืจ ืืืฆืืจืฃ ืืฉืืจื ืขื ืืื ืช ืืงืกื ืขืงืืืช ืืืื ืืืงืก ืืืืืคืืฉ.
- ืืงืืืื
input_idsืึพattention_maskืื ืืกืืint64ืืืฆืืจื[1, sequence_length]. ืืืจืฃ ืชืืื ืืจืฆืฃ ืืื ืืื ืงืจืืื. - ืืคืื
embeddingืืืfloat32[1, 256], ืื ืึพINT8. ืึพpooling, ืืืืื ืืื ืจืืื ืืืืืื ืืืจืฃ. - ืืืืืื ืืืคืขืืช ืืืืื ืฉื 256 ืืืงื ืื, ืืืืื ืืืจืืจืช ืืืืื ืฉื ืืืงืืื ืืืงืืื. ืืืจืืืืงืืืจื ืืืคืฉืจืช ืขื 512; ืืืงื ืืืืืช ืืืชืืื ืืคื ื ืฉืื ืื ืืืืืื.
- ืืืงื ืืกืืืื ืืจืืืื ืืืงืืขืื ืืฉืืจื ืืคื ืืืช ืืืงืืจ ืืฆื ืืืืงืืืจืื. ืืงืกื ืืขืืจ ืืืืืื ื ืืชื.
- ืฉืื ืื ืืืืื, ืืืืืืช, ืืืืงื ืืืืจ ืื ืืื ืช ืืืงืกื ืืืืื ืืืืงืช ืืชืืื ืืืืจื ืืื ืงืืืื ืืืืฉ ืฉื ืืืื ืืงืก.
ืฆืืื ืืืืืื ืืฉืืฉ ืืืืจืื ืชืืฆืืืช. ืืื ืืื ื ืืกืชืืจืืช ืื ืืื ืืช ืืชืืฆืื ืื ืืืื ืืืืืช.
ืืงืืจ ืืืจืืืืงืืืจื
ืืืืืื ืืืืกืกืช ืขื ืืืื ืืืืืืจ Round 2. ืื ืืืฆืข ืืืืื ื ืืกืฃ ืืฆืืจื ืืืืฆืื.
ืืืงืืื ืืืื BERT ืขื 8 ืฉืืืืช, ืจืืื ืืืื 512, ืฉืืื ื ืจืืฉื ืงืฉื, ืจืืื ืืื ืืื 2048 ืืืืฆืจ ืืืืื ืฉื 32,000 ืืืงื ืื; ืึพ42.26 ืืืืืื ืคืจืืืจืื ืืคื ื ืืืืืช. ืืืจืฃ ืืืื ืืช ืฉืจืฉืจืช ืืงืืืื ืืืืื: backbone, ืืืืฆืข ืืคื ืืกืืช ืืงืฉื, ืืืื ื ืืืืช ืึพ512 ืึพ256, LayerNorm ืื ืจืืื L2.
ืืงืืจ ืืืืฆืื ืืื ืชืืงืืืช final ืืืจืกื 8d7016cf472fb436dbd3ac843ef4d407c28d08f4. ืคืจืื ืืืงืืจ ืืืืชืืืืช ืืืคืืขืื ืึพmanifest.json.
ืืขืจืื ืืืืืืืช
ืืืืืช ืืืืืจ ืฉื checkpoint ืืืงืืจ
ืืืึพืืืืช checkpoint ืืืงืืจ ืืืืืืช ืขื ืกื ืคืืชืื ืฉื 410 ืฉืืืืช ืึพ12,002 ืืงืืขืื, ืืื ืื ื ืืืง ืขื ืฉืืืืช ืกืื ืืืืืช ืืืื ื ืืฉืงืฃ ืืช ืืืืืืช ืืืืื ืขื ืฉืืืืช ืื ืืฉืืืช:
| ืืื | ืขืจื |
|---|---|
| nDCG@10 | 0.4040 |
| MRR | 0.3662 |
| Recall@5 | 0.4683 |
| Recall@10 | 0.5537 |
ืืื ืืืื checkpoint ืืืงืืจ ืืกื ืืคืืชืื ืฉืฉืืืฉ ืื ืืืืืจืชื. ืื ืืื ื ืืืืงื ืขืฆืืืืช ืฉื INT8 ืื ืืืืื ืืืืืืช ืขื ืืืืจ ืืืฉ. ืืืื ืืืืืจ, ืืืืงื ืืฉืืจืืช ืืืงืื ืืืงืืขืื ืืกืื ืืฉืืืืช ืขืฉืืืื ืืฉื ืืช ืืช ืืชืืฆืืืช. ืืืืืฅ ืืืืืง ืฉืืืืช ืืืืฆืืืช ืฉื ืืืืฉืื ืืืืฉืืืช ืืืคืืฉ ืกืื ืื, ืืืืืื ืืืืืจืืื.
ืืืืืช ืืืฆืื ONNX
ืฉื ื ืืงืืฆืื ืขืืจื ืืืืงืช ืชืงืื ืืช ONNX ืืืจืฆื ืขื CPU. ืืืจืืขื ืงืืื ืืืืงื ืืขืืจืืช ืืืืจืืืช, ืืืืจืืื ืฉืื ืื ืขื 256 ืืืงื ืื, ืืืฉืื ืืืืงืืืจืื ืืืงืืื PyTorch ืืืงืืจื:
| ืืืฆืื | ืืคืจืฉ ืืืืื ืืจืื ืืงืืืืจืืื ืื | ืืืืื ืงืืกืื ืืก ืืืขืจื ืืืงืืจ |
|---|---|---|
| FP32 | 2.85e-7 | 0.99999994 |
| INT8 | 0.01123 | 0.99863 |
ืื ืืืืงืช ืงืจืื ืืกืคืจืืช ืืืจืืข ืืืืืืืช, ืืื ืืื ืืืืืช ืืืืืจ. ืืืืืช INT8 ืขืฉืื ืืฉื ืืช ืืืจืืืื, ืืืืืื ืืืฉืจ ืืฆืืื ืื ืงืจืืืื. ืื ืืคืืจืกื ืืื ืืื ืืืืจืืช: ืืื ื ืงืืืื ืืืืคืืฉ ืชืืืืื ืืืืืจื, ืืืืจื ืืงืื ืืืืื ืืงืก.
ืฉืืืื ืืืืช (Zayit)
ืืืืืื ื ืืขืื ืืืคืฉืจ ืฉืืืื ืืืืช. ืืคื ืชืืืืจ ืืชืืืืืช ืฉืฉืืืฉ ืืืื ืช ืืืืฆืื, ืืืฉืง v5 ืืงืืื ืืฆืคื ืึพ384 ืืืืื, ืืืืืื ืฉื 128 ืืืงื ืื ืืืงืื ืืื ืงืืืืืืช. ืฉืืืื ืืืื ืื ืืืจืฉ ืชืืืื ืึพ256 ืืืืื, ืืืกืคืช ืืงืืืืืืช, ืงืืืื ืืืืฉ ืฉื ืืืืงืืืจืื ืืื ืืืช ืืื ืืงืก Lucene ืืืชืื. ืืฉ ืืืืืง ืื ืคืจื ื ืจืืื ืืงืกื ืืืืืืช ืืืืืจ ืืจืืช ืฉืืจื; ืฉืื ืื ืฉื ืืงืืืฅ ืืืื ืืื ื ืืฉืืื ืืช ืืฉืืืื.
ืืืืืืืช ืืชืืื ืืช
ืืชืืื ืืช ืืืืืช ืฆืืจืคื ืืืืืฉืช ืืฉืืืืฉ. ืื ืืื ื ืืื ืืขืจืื ืฉืืืชื ืื ืืชืืืืืืช ืืชืืฆืืืช. ืงืืฉืืจื ืืืงืืจ.
ืงืืืฆื ืืืืืจ
| ืงืืืฅ | ืชืืื |
|---|---|
seforim-embed-round2-fp32.onnx |
ืืืฆืื FP32 |
seforim-embed-round2-int8.onnx |
ืืืฆืื INT8 ืงืืืคืงืื ืืืืืืช ืืื ืื |
tokenizer.json |
ืืืืงื ืืืืจ ืืชืืื ืืืืื |
manifest.json |
ืืืื ืืงืื ืืืคืื, ืืงืืจ, ืชืืฆืืืช ืืืืืช, ืืืืื ืืืชืืืืช SHA-256 |
export_round2_onnx.py |
ืงืื ืืืืฆืื ืืืืืงืืช ืืงืจืื ืืืงืืจ |
LICENSE.md |
ืชื ืื ืืจืืฉืื ืืคื ืจืืื |
ืื ืืืื?
ืืืืืช ืืืชืื ืืช ืืืืื ืฉืคื ืชืืจื ื ืงื ืืฉืงื.
ืืืืื ืืืื ืืืืขื ืืขื ืืืฆืืขืื ืฉืืืืฉืืื.
ืฉืืคืืจ ืืืจืกืืื ืฉื ืืืืื ืืืืฆืขืืช ืืฉืื ืื ืืฉื ืืืฉืชืืฉืื.
ืคืืชืื ืืืืืื ืืืงืื ืืฉืืืืฉืืื ืืืชืจ:
- ืืชืืงืืืช ืืืืืื ืืฉืืืืช โ ืืชื ืืฉืงื ืืืื ืืืชืจ ืืืืืื ืื ืืืื ืื ืืฉืืขืืชืืื, ืืืื ืฆืืืืืื.
- ืืงืฉืจ ืืจืื โ ืชืืืื ืึพ4K ืืืงื ืื.
- ืืืจื ืืงืืืจ ืืืืฉ โ ืืคืฉืจืืช ืืืฉืชืืฉ ืจืง ืึพ32/64/128/256/512/1024 ืืขืจืืื ืืจืืฉืื ืื ืฉื ืืืืงืืืจ, ืชืื ืจืืืื ืืืืืข ืืืฉืื ืืชืืืืช ืืืืงืืืจ.
- ืืฉืืืื ืืฉืืจื ืืื ืฉืืืืชื ืืงืืข โ ืชืืืื ืืืื ืช ืฉืืืืชื ืืงืืข ืืื ืืงืืืช ืฆืืื ืจืืืื ืืืืช ืืืืืง ืืืชืจ.
ืืืืื ืืฉืคืืช ืืืืืื ืืืืืื ืฉืื ืื, ืืืฉืชืคืื ืืจืื ืกืื ืื ืืฉืืชืฃ ืืืืคืฉืจืื ืืืืคื ืืืื ืืื ืืื:
- Sinai โ ืืืื ืืืืจ ืืงื ืืฉืงื [ืืฉืืขืจ: ~30M ืคืจืืืจืื].
- OkerHarim โ ืืืื ืืืง ืืืืืืง [ืืฉืืขืจ: ~80M ืคืจืืืจืื]. ื ืืชื ืืืฆืืจ ืืื ืืงืก ืืืืฆืขืืช OkerHarim ืืืืฆืข ืืืคืืฉ ืืืืฆืขืืช Sinai, ืื ืืืคื, ืืื ืฆืืจื ืืืฆืืจืช ืืื ืืงืก ืืืืฉ, ืืฉืื ืฉืฉื ื ืืืืืืื ืืืคืื ืืงืกืืื ืืืืชื ืืจืื ืกืื ืื.
ืจืืฉืื ืืืืืืก
ืื ืฉืืืืฉ ืืกืืจื ืืกืืจ ืืื ืืืฉืืจ ืืคืืจืฉ ืืจืืฉ ืืืืชื ืืืจืืื ืื ืืืื โ arieldaniely@gmail.com. ืืกืืจ ืืฉืื ืื ืืืฉืชืืฉ ืืืืื ืืื ืฆืืจื ืืืขืจืืช, ืชืืื ื ืื ืฉืืจืืช ืืืืจืืช ืจืืื, ืื ืื ืืจืืื ืฉืื ืืื ืืืืืข ืืื ืื. ืจืื ืืจืืฉืืื ืืืื.
ืืืืื ืืืืืงื ืืืืจ ืืืคืฆืื ืืืจืกื ืื ืชืืช Meivin Non-Commercial License 1.0. ืืืืกืืจ ืืืื ืฉืืืื ืืฉืืจ ืื ืขืงืืฃ, API, ืชืืกืฃ, ืฉืืจืืช ืืจืืืง ืืฉืืืืฉ ืืืืงืืืจืื ืื ืืืื ืืงืกืื ืืจืืื ืืืขืจืืช ืืกืืจืืช. ืืืฉืืจ ืืืจืื ืื ืืืฉื ืืืืืจ ืืื ื ืืืฉืืจ ืืฉืืืืฉ ืืกืืจื. ืืคืฆื ืืขืืืืืื ืืืืืืื ืฉืืืจืช ืืจืืฉืืื ืืืืืืืก. ืงืื ืืงืืจื ื ืืืื ื ืฉืืจ ืืคืืฃ ืึพPersonal Use License 1.0 ืฉืืงืืืฅ ืืจืืฉืืื; ืืืืืืช ืฆื ืฉืืืฉื ื ืฉืืจืืช.
ืืืืืก: ืืจืืื ืื ืืืื โ Judaic Semantic Embedding project, Otzaria. ืืืืืืก ืืชืืืืก ืืคืืชืื ืืืจืืืคืงืืื ืืืื ื ืืขื ื ืืืืืจืึผืช ืฉื ืืงืกืื ืืืงืืจ.
ืชืืขืื ืืืืืื
ืืื ืืืืืื ืืืคืืจื ืืชืขื ืืช ืืกืืื ืืืฉืงืืืืช ืฉื Round 2, ืืืืื ื ืึพMLM, ืืืืฉืืจ ืืขืืจืโืืจืื, ืกืืื ืืืืืืจ ืืื ืืกืืืื ืฉืื ืงืืืื. ืืื ืืืื ืืืคืจึพืคืจืืืจืื, ืืจืคืื, ื ืชืื ื ืืืืื ืืืืืืืช ืืืขืจืื.
ืืืจืฃ ืืฆืื ืืืืงืช MLM ืงืืืขื ืืืืจืืื ืืขืืจืืื; ืืื ืืื ื ืืื ืืฆืืื ืฉื ืืืคืืฉ ืืงืืจืืช. ืชืืฆืืืช ืืืืืืจ ืืคืจืืืืงืืื ืืืืืงื ืืืคืืขืื ืืืื.
Languages and textual coverage | ืฉืคืืช ืื ืืื ืืฉืื
| Metadata | Coverage | Evidence and interpretation |
|---|---|---|
he โ Hebrew |
Biblical, Mishnaic/Rabbinic, medieval and modern Hebrew | Tanakh, Mishnah, halakhic and commentarial literature, and modern HeQ text |
tmr โ Jewish Babylonian Aramaic (ca. 200โ1200 CE) |
Aramaic in the Babylonian Talmud | Bavli corpus bucket; Bavli also contains Hebrew |
jpa โ Jewish Palestinian Aramaic |
Palestinian rabbinic Aramaic represented in the Jerusalem Talmud | Yerushalmi corpus bucket; Yerushalmi also contains Hebrew |
arc โ Official/Imperial Aramaic code |
Biblical Aramaic within Daniel and Ezra; broad Aramaic discovery label | Both books were verified in retained training rows; not every verse is Aramaic |
ืืกืืืื ืืืื ื he, tmr, jpa, arc ืืชืืื ืืืืงืฃ ืืืงืืจืืช. tmr ืืื ืืืจืืืช ืืืืืืืช ืืืืืืช, jpa ืืืจืืืช ืืืืืืืช ืืืจืฅึพืืฉืจืืืืช, ืึพarc ืืฉืืฉ ืืื ืื ืืกืืืื ืืืจืืืช ืืืงืจืืืช. ืื ืืื ืืขืืจื ืืืืืื ืขืืจืืช ืืืจืืืช; ืืืื ืืืจืืฉืืื ืื ืฉืืืช ืงืืจืคืืกืื, ืื ืฉืืืช ืฉืคืืช. ืืื ืืกืื ืืช ืื ืืชืืืืืื ืืฉืคื ืืืช ืื ืืช ืื ืกืคืจ ืื ืืื ืืืจืืืช.
The arc ISO label is historical and is not a precise label for every Aramaic variety in the corpus. Targumic and Zoharic/literary Aramaic are also represented in the broader Judaic source collection; they are described here rather than assigned unrelated modern Aramaic codes. Presence of texts establishes training coverage, not independently benchmarked mastery of each dialect. No per-language quality scores or exact percentages of Aramaic are available, because corpus buckets are not language annotations. English instructions in this repository do not imply English model support.
See aggregate corpus checks. Language code references: Jewish Babylonian Aramaic, Jewish Palestinian Aramaic, and Biblical Hebrew/Aramaic language codes in BHSA.
Downloads and parameter display | ืืืจืืืช ืืคืจืืืจืื
The full FP32 checkpoint is distributed as root model.safetensors, enabling Hugging Face to inspect its tensor headers.
Hugging Face currently reports 42,260,992 parameters in its native Safetensors display. The direct PyTorch count is 42,260,993, including the scalar training parameter logit_scale. The one-parameter difference does not affect embeddings or model size in practical terms. Both counts refer to the same released file.
The count includes stored training parameters; ONNX can omit unused components. INT8 is a numerical export of the same model, not a reduction in architectural parameter count.
Use the documented loader/configuration together with the weights. The Hub counts server-side requests to designated query files such as config.json; it does not count unique users or every arbitrary file download. ONNX integrations below load the real configuration as part of model initialization. Direct downloads of ONNX files alone may not be represented in this counter. Historical missed requests cannot be reconstructed from repository metadata.
The model page shows recent downloads; Hub API also exposes downloadsAllTime for cumulative downloads, when available. The release includes an example that reads both fields live, without hardcoding a launch-day number. See Hugging Face counting rules.
Citation
@misc{daniely2026meivinembed,
author = {Ariel Daniely},
title = {Meivin Embed 0.1: Semantic Retrieval for Hebrew and Aramaic Judaic Texts},
year = {2026},
url = {https://huggingface.co/ArieLLL123/Meivin-Embed-0.1}
}
- Downloads last month
- 3








