gmakstutis/LightRAG
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #1913 from danielaskdd/fix-base64-for-embedding

Browse source 
Feat: Change embedding formats from float to base64 for efficiency
This commit is contained in:
Daniel.y 2025-08-05 11:48:53 +08:00 • committed by GitHub
commit 793e82ae89
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 52 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
env.example
View file

@ -121,6 +121,16 @@ LLM_MODEL=gpt-4o
LLM_BINDING_HOST=https://api.openai.com/v1 LLM_BINDING_HOST=https://api.openai.com/v1
LLM_BINDING_API_KEY=your_api_key LLM_BINDING_API_KEY=your_api_key
### Optional for Azure
# AZURE_OPENAI_API_VERSION=2024-08-01-preview
# AZURE_OPENAI_DEPLOYMENT=gpt-4o
### Openrouter example
# LLM_MODEL=google/gemini-2.5-flash
# LLM_BINDING_HOST=https://openrouter.ai/api/v1
# LLM_BINDING_API_KEY=your_api_key
# LLM_BINDING=openai
### Most Commont Parameters for Ollama Server ### Most Commont Parameters for Ollama Server
### Time out in seconds, None for infinite timeout ### Time out in seconds, None for infinite timeout
TIMEOUT=240 TIMEOUT=240
@ -132,14 +142,11 @@ OLLAMA_LLM_NUM_CTX=32768
# OLLAMA_LLM_TEMPERATURE=0.85 # OLLAMA_LLM_TEMPERATURE=0.85
### see also env.ollama-binding-options.example for fine tuning ollama ### see also env.ollama-binding-options.example for fine tuning ollama
### Optional for Azure
# AZURE_OPENAI_API_VERSION=2024-08-01-preview
# AZURE_OPENAI_DEPLOYMENT=gpt-4o
#################################################################################### ####################################################################################
### Embedding Configuration (Should not be changed after the first file processed) ### Embedding Configuration (Should not be changed after the first file processed)
#################################################################################### ####################################################################################
### Embedding Binding type: openai, ollama, lollms, azure_openai, jina ### Embedding Binding type: ollama, openai, azure_openai, jina, lollms
### see also env.ollama-binding-options.example for fine tuning ollama ### see also env.ollama-binding-options.example for fine tuning ollama
EMBEDDING_BINDING=ollama EMBEDDING_BINDING=ollama
@ -149,7 +156,7 @@ EMBEDDING_BINDING_API_KEY=your_api_key
# If the embedding service is deployed within the same Docker stack, use host.docker.internal instead of localhost # If the embedding service is deployed within the same Docker stack, use host.docker.internal instead of localhost
EMBEDDING_BINDING_HOST=http://localhost:11434 EMBEDDING_BINDING_HOST=http://localhost:11434
### OpenAI compatible ### OpenAI compatible (VoyageAI embedding openai compatible)
# EMBEDDING_BINDING=openai # EMBEDDING_BINDING=openai
# EMBEDDING_MODEL=text-embedding-3-large # EMBEDDING_MODEL=text-embedding-3-large
# EMBEDDING_DIM=3072 # EMBEDDING_DIM=3072

35
lightrag/llm/jina.py
View file

@ -8,6 +8,7 @@ if not pm.is_installed("tenacity"):
pm.install("tenacity") pm.install("tenacity")
import numpy as np import numpy as np
import base64
import aiohttp import aiohttp
from tenacity import ( from tenacity import (
retry, retry,
@ -23,12 +24,34 @@ async def fetch_data(url, headers, data):
async with session.post(url, headers=headers, json=data) as response: async with session.post(url, headers=headers, json=data) as response:
if response.status != 200: if response.status != 200:
error_text = await response.text() error_text = await response.text()
logger.error(f"Jina API error {response.status}: {error_text}")
# Check if the error response is HTML (common for 502, 503, etc.)
content_type = response.headers.get("content-type", "").lower()
is_html_error = (
error_text.strip().startswith("<!DOCTYPE html>")
or "text/html" in content_type
)
if is_html_error:
# Provide clean, user-friendly error messages for HTML error pages
if response.status == 502:
clean_error = "Bad Gateway (502) - Jina AI service temporarily unavailable. Please try again in a few minutes."
elif response.status == 503:
clean_error = "Service Unavailable (503) - Jina AI service is temporarily overloaded. Please try again later."
elif response.status == 504:
clean_error = "Gateway Timeout (504) - Jina AI service request timed out. Please try again."
else:
clean_error = f"HTTP {response.status} - Jina AI service error. Please try again later."
else:
# Use original error text if it's not HTML
clean_error = error_text
logger.error(f"Jina API error {response.status}: {clean_error}")
raise aiohttp.ClientResponseError( raise aiohttp.ClientResponseError(
request_info=response.request_info, request_info=response.request_info,
history=response.history, history=response.history,
status=response.status, status=response.status,
message=f"Jina API error: {error_text}", message=f"Jina API error: {clean_error}",
) )
response_json = await response.json() response_json = await response.json()
data_list = response_json.get("data", []) data_list = response_json.get("data", [])
@ -82,6 +105,7 @@ async def jina_embed(
"model": "jina-embeddings-v4", "model": "jina-embeddings-v4",
"task": "text-matching", "task": "text-matching",
"dimensions": dimensions, "dimensions": dimensions,
"embedding_type": "base64",
"input": texts, "input": texts,
} }
@ -108,7 +132,12 @@ async def jina_embed(
f"Jina API returned {len(data_list)} embeddings for {len(texts)} texts" f"Jina API returned {len(data_list)} embeddings for {len(texts)} texts"
) )
embeddings = np.array([dp["embedding"] for dp in data_list]) embeddings = np.array(
[
np.frombuffer(base64.b64decode(dp["embedding"]), dtype=np.float32)
for dp in data_list
]
)
logger.debug(f"Jina embeddings generated: shape {embeddings.shape}") logger.debug(f"Jina embeddings generated: shape {embeddings.shape}")
return embeddings return embeddings

10
lightrag/llm/openai.py
View file

@ -34,6 +34,7 @@ from lightrag.types import GPTKeywordExtractionFormat
from lightrag.api import __api_version__ from lightrag.api import __api_version__
import numpy as np import numpy as np
import base64
from typing import Any, Union from typing import Any, Union
from dotenv import load_dotenv from dotenv import load_dotenv
@ -472,6 +473,11 @@ async def openai_embed(
async with openai_async_client: async with openai_async_client:
response = await openai_async_client.embeddings.create( response = await openai_async_client.embeddings.create(
model=model, input=texts, encoding_format="float" model=model, input=texts, encoding_format="base64"
)
return np.array(
[
np.frombuffer(base64.b64decode(dp.embedding), dtype=np.float32)
for dp in response.data
]
) )
return np.array([dp.embedding for dp in response.data])