New embed script

.envrc

@@ -1,3 +1,2 @@
 source_up_if_exists
 source_env_if_exists ./.envrc.local
-layout anaconda bible-embeddings

.python-version

@@ -0,0 +1 @@
+3.13

README.md

@@ -83,7 +83,7 @@ For high-level information about these models, see the [MTEB Leaderboard](https:
 - [intfloat/e5-small-v2](https://huggingface.co/intfloat/e5-small-v2)
   - 384 dimensions
   - BSB: 336.522 MB
-- [jinaai/jina-embeddings-v3]
+- [jinaai/jina-embeddings-v3](https://huggingface.co/jinaai/jina-embeddings-v3)
   - 1024 dimensions
   - BSB: 738.597 MB
 - [nvidia/NV-Embed-V2](https://huggingface.co/nvidia/NV-Embed-V2)

embed-gemini.py

@@ -1,54 +0,0 @@
-import argparse
-import json
-import os
-from google import genai
-import logging
-from itertools import groupby
-from more_itertools import chunked
-from tenacity import retry, wait_exponential, before_sleep_log
-
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-arg_parser = argparse.ArgumentParser(prog="embed-gemini", description="Creates embeddings of Bible verses using Gemini API")
-arg_parser.add_argument("translation", help="Located at text/<translation>.json")
-arg_parser.add_argument("model", help="Gemini model")
-
-args = arg_parser.parse_args()
-
-with open(f"text/{args.translation}.json") as f:
-    bible_data = [verse for verse in json.load(f) if len(verse["text"]) > 0]
-
-grouped_bible_data = groupby(bible_data, lambda item: (item["book"], item["chapter"]))
-
-model_name = args.model
-
-client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))
-
-@retry(wait=wait_exponential(multiplier=2, min=60, max=300), before_sleep=before_sleep_log(logger, logging.WARNING))
-def call_embeddings_api(sentences):
-    response = client.models.embed_content(model=model_name, contents=sentences)
-    return response
-
-def get_embeddings(sentences):
-    all_embeddings = []
-    batch_size = 100  # API limit
-
-    for batch in chunked(sentences, batch_size):
-        response = call_embeddings_api(batch)
-        all_embeddings.extend(embedding.values for embedding in response.embeddings)
-
-    return all_embeddings
-
-for (book, chapter), grouped in grouped_bible_data:
-    print(f"Embedding {book} {chapter}")
-    verses = list(grouped)
-    data = []
-    base_path = os.path.join("embeddings", book, str(chapter).zfill(3))
-    os.makedirs(base_path, exist_ok=True)
-    sentences = [verse["text"] for verse in verses]
-    embeddings = get_embeddings(sentences)
-    for i, verse in enumerate(verses):
-        data.append({ "book": verse["book"], "chapter": verse["chapter"], "verse": verse["verse"], "embedding": embeddings[i] })
-    with open(os.path.join(base_path, f"{args.translation}-{model_name}.json"), "w", encoding='utf-8') as f:
-        json.dump(data, f, indent=2)

embed-ollama.py

@@ -1,41 +0,0 @@
-import argparse
-import json
-import ollama
-import os
-from itertools import groupby
-import numpy as np
-
-arg_parser = argparse.ArgumentParser(prog="embed-ollama", description="Creates embeddings of Bible verses using Ollama")
-arg_parser.add_argument("translation", help="Located at text/<translation>.json")
-arg_parser.add_argument("model", help="Ollama model")
-
-args = arg_parser.parse_args()
-
-with open(f"text/{args.translation}.json") as f:
-    bible_data = [verse for verse in json.load(f) if len(verse["text"]) > 0]
-
-grouped_bible_data = groupby(bible_data, lambda item: (item["book"], item["chapter"]))
-
-model_name = args.model
-
-def get_embeddings(sentences):
-    embeddings = []
-    for _, s in enumerate(sentences):
-        res = ollama.embeddings(model=model_name, prompt=s)
-        embedding = np.array(res["embedding"])
-        norm = np.linalg.norm(embedding)
-        embeddings.append((embedding / norm).tolist())
-    return embeddings
-
-for (book, chapter), grouped in grouped_bible_data:
-    print(f"Embedding {book} {chapter}")
-    verses = list(grouped)
-    data = []
-    base_path = os.path.join("embeddings", book, str(chapter).zfill(3))
-    os.makedirs(base_path, exist_ok=True)
-    sentences = [verse["text"] for verse in verses]
-    embeddings = get_embeddings(sentences)
-    for i, verse in enumerate(verses):
-        data.append({ "book": verse["book"], "chapter": verse["chapter"], "verse": verse["verse"], "embedding": embeddings[i] })
-    with open(os.path.join(base_path, f"{args.translation}-{model_name.replace("/", "__")}.json"), "w", encoding='utf-8') as f:
-        json.dump(data, f, indent=2)

embed-openai.py

@@ -1,39 +0,0 @@
-import argparse
-import json
-import os
-import openai
-from itertools import groupby
-from tenacity import retry, wait_exponential, before_sleep_log
-
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-arg_parser = argparse.ArgumentParser(prog="embed-openai", description="Creates embeddings of Bible verses using OpenAI API")
-arg_parser.add_argument("translation", help="Located at text/<translation>.json")
-arg_parser.add_argument("model", help="OpenAI model")
-
-args = arg_parser.parse_args()
-
-with open(f"text/{args.translation}.json") as f:
-    bible_data = [verse for verse in json.load(f) if len(verse["text"]) > 0]
-
-grouped_bible_data = groupby(bible_data, lambda item: (item["book"], item["chapter"]))
-
-model_name = args.model
-
-@retry(wait=wait_exponential(multiplier=2, min=2, max=30), before_sleep=before_sleep_log(logger, logging.WARNING))
-def get_embeddings(sentences):
-    return openai.Embedding.create(input=sentences, model=model_name)["data"]
-
-for (book, chapter), grouped in grouped_bible_data:
-    print(f"Embedding {book} {chapter}")
-    verses = list(grouped)
-    data = []
-    base_path = os.path.join("embeddings", book, str(chapter).zfill(3))
-    os.makedirs(base_path, exist_ok=True)
-    sentences = [verse["text"] for verse in verses]
-    embeddings = get_embeddings(sentences)
-    for i, verse in enumerate(verses):
-        data.append({ "book": verse["book"], "chapter": verse["chapter"], "verse": verse["verse"], "embedding": embeddings[i]["embedding"] })
-    with open(os.path.join(base_path, f"{args.translation}-{model_name}.json"), "w", encoding='utf-8') as f:
-        json.dump(data, f, indent=2)

embed-voyageai.py

@@ -1,51 +0,0 @@
-import argparse
-import json
-import os
-import voyageai
-from itertools import groupby
-import logging
-from tenacity import retry, wait_exponential, before_sleep_log
-
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-arg_parser = argparse.ArgumentParser(prog="embed-voyageai", description="Creates embeddings of Bible verses using VoyageAI API")
-arg_parser.add_argument("translation", help="Located at text/<translation>.json")
-arg_parser.add_argument("model", help="VoyageAI model")
-
-args = arg_parser.parse_args()
-
-vo = voyageai.Client()
-
-with open(f"text/{args.translation}.json") as f:
-    bible_data = [verse for verse in json.load(f) if len(verse["text"]) > 0]
-
-grouped_bible_data = groupby(bible_data, lambda item: (item["book"], item["chapter"]))
-
-model_name = args.model
-
-def chunk_sentences(sentences, chunk_size=128):
-    """Split sentences into chunks of specified size."""
-    for i in range(0, len(sentences), chunk_size):
-        yield sentences[i:i + chunk_size]
-
-@retry(wait=wait_exponential(multiplier=2, min=2, max=30), before_sleep=before_sleep_log(logger, logging.WARNING))
-def get_embeddings(sentences):
-    all_embeddings = []
-    for chunk in chunk_sentences(sentences):
-        embeddings = vo.embed(texts=chunk, model=model_name).embeddings
-        all_embeddings.extend(embeddings)
-    return all_embeddings
-
-for (book, chapter), grouped in grouped_bible_data:
-    print(f"Embedding {book} {chapter}")
-    verses = list(grouped)
-    data = []
-    base_path = os.path.join("embeddings", book, str(chapter).zfill(3))
-    os.makedirs(base_path, exist_ok=True)
-    sentences = [verse["text"] for verse in verses]
-    embeddings = get_embeddings(sentences)
-    for i, verse in enumerate(verses):
-        data.append({ "book": verse["book"], "chapter": verse["chapter"], "verse": verse["verse"], "embedding": embeddings[i] })
-    with open(os.path.join(base_path, f"{args.translation}-{model_name}.json"), "w", encoding='utf-8') as f:
-        json.dump(data, f, indent=2)

embed.py

@@ -1,32 +0,0 @@
-import argparse
-import json
-import os
-from itertools import groupby
-from sentence_transformers import SentenceTransformer
-
-arg_parser = argparse.ArgumentParser(prog="embed", description="Creates embeddings of Bible verses")
-arg_parser.add_argument("translation", help="Located at text/<translation>.json")
-arg_parser.add_argument("model", help="SentenceTransformer model")
-
-args = arg_parser.parse_args()
-
-with open(f"text/{args.translation}.json") as f:
-    bible_data = json.load(f)
-
-grouped_bible_data = groupby(bible_data, lambda item: (item["book"], item["chapter"]))
-
-model = SentenceTransformer(args.model, trust_remote_code=True)
-model_name = args.model.split("/")[-1]
-
-for (book, chapter), grouped in grouped_bible_data:
-    print(f"Embedding {book} {chapter}")
-    verses = list(grouped)
-    data = []
-    base_path = os.path.join("embeddings", book, str(chapter).zfill(3))
-    os.makedirs(base_path, exist_ok=True)
-    sentences = [verse["text"] for verse in verses]
-    embeddings = model.encode(sentences, normalize_embeddings=True)
-    for i, verse in enumerate(verses):
-        data.append({ "book": verse["book"], "chapter": verse["chapter"], "verse": verse["verse"], "embedding": embeddings[i].tolist() })
-    with open(os.path.join(base_path, f"{args.translation}-{model_name}.json"), "w", encoding='utf-8') as f:
-        json.dump(data, f, indent=2)