From df96159e882d2e5717da460891320847ac2e3b8a Mon Sep 17 00:00:00 2001
From: Huu Le <39040748+leehuwuj@users.noreply.github.com>
Date: Fri, 12 Jul 2024 15:01:41 +0700
Subject: [PATCH] feat: Use Qdrant FastEmbed as local embedding provider (#162)
---
.changeset/happy-emus-attend.md | 5 +++
.gitignore | 3 ++
helpers/python.ts | 14 ++++++-
.../components/settings/python/settings.py | 39 +++++++++++--------
4 files changed, 42 insertions(+), 19 deletions(-)
create mode 100644 .changeset/happy-emus-attend.md
diff --git a/.changeset/happy-emus-attend.md b/.changeset/happy-emus-attend.md
new file mode 100644
index 00000000..c7e04279
--- /dev/null
+++ b/.changeset/happy-emus-attend.md
@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Use Qdrant FastEmbed as local embedding provider
diff --git a/.gitignore b/.gitignore
index 6c50932b..534068cf 100644
--- a/.gitignore
+++ b/.gitignore
@@ -46,5 +46,8 @@ e2e/cache
# intellij
**/.idea
+# Python
+.mypy_cache/
+
# build artifacts
create-llama-*.tgz
diff --git a/helpers/python.ts b/helpers/python.ts
index d73f6c62..fd755440 100644
--- a/helpers/python.ts
+++ b/helpers/python.ts
@@ -153,14 +153,24 @@ const getAdditionalDependencies = (
version: "0.2.6",
});
break;
+ case "groq":
+ dependencies.push({
+ name: "llama-index-llms-groq",
+ version: "0.1.4",
+ });
+ dependencies.push({
+ name: "llama-index-embeddings-fastembed",
+ version: "^0.1.4",
+ });
+ break;
case "anthropic":
dependencies.push({
name: "llama-index-llms-anthropic",
version: "0.1.10",
});
dependencies.push({
- name: "llama-index-embeddings-huggingface",
- version: "0.2.0",
+ name: "llama-index-embeddings-fastembed",
+ version: "^0.1.4",
});
break;
case "gemini":
diff --git a/templates/components/settings/python/settings.py b/templates/components/settings/python/settings.py
index e0c974cc..ce427645 100644
--- a/templates/components/settings/python/settings.py
+++ b/templates/components/settings/python/settings.py
@@ -98,8 +98,25 @@ def init_azure_openai():
Settings.embed_model = AzureOpenAIEmbedding(**embedding_config)
+def init_fastembed():
+ """
+ Use Qdrant Fastembed as the local embedding provider.
+ """
+ from llama_index.embeddings.fastembed import FastEmbedEmbedding
+
+ embed_model_map: Dict[str, str] = {
+ # Small and multilingual
+ "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
+ # Large and multilingual
+ "paraphrase-multilingual-mpnet-base-v2": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2", # noqa: E501
+ }
+
+ # This will download the model automatically if it is not already downloaded
+ Settings.embed_model = FastEmbedEmbedding(
+ model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
+ )
+
def init_groq():
- from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.groq import Groq
model_map: Dict[str, str] = {
@@ -108,19 +125,13 @@ def init_groq():
"mixtral-8x7b": "mixtral-8x7b-32768",
}
- embed_model_map: Dict[str, str] = {
- "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
- "all-mpnet-base-v2": "sentence-transformers/all-mpnet-base-v2",
- }
Settings.llm = Groq(model=model_map[os.getenv("MODEL")])
- Settings.embed_model = HuggingFaceEmbedding(
- model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
- )
+ # Groq does not provide embeddings, so we use FastEmbed instead
+ init_fastembed()
def init_anthropic():
- from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.anthropic import Anthropic
model_map: Dict[str, str] = {
@@ -131,15 +142,9 @@ def init_anthropic():
"claude-instant-1.2": "claude-instant-1.2",
}
- embed_model_map: Dict[str, str] = {
- "all-MiniLM-L6-v2": "sentence-transformers/all-MiniLM-L6-v2",
- "all-mpnet-base-v2": "sentence-transformers/all-mpnet-base-v2",
- }
-
Settings.llm = Anthropic(model=model_map[os.getenv("MODEL")])
- Settings.embed_model = HuggingFaceEmbedding(
- model_name=embed_model_map[os.getenv("EMBEDDING_MODEL")]
- )
+ # Anthropic does not provide embeddings, so we use FastEmbed instead
+ init_fastembed()
def init_gemini():
--
GitLab