JDSS/mem0
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs(rerankers): align reranker config docs with v3 SDK source

Browse Source
This commit is contained in:
kartik-mem0
2026-06-25 11:49:24 +05:30
parent ac296f7534
commit 5345ed5e0b
9 changed files with 55 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/components/rerankers/config.mdx
+7 -5
View File
@@ -61,11 +61,13 @@ All rerankers share these common configuration parameters:
### LLM Reranker ### LLM Reranker
| Parameter | Description | Type | Default | | Parameter | Description | Type | Default |
| -------------- | --------------------------- | ------ | -------- | | ----------- | -------------------------------------------------------------------------------------------- | ------ | -------------- |
| `llm.provider` | LLM provider for reranking | `str` | Required | | `provider` | LLM provider (`openai`, `anthropic`, etc.) | `str` | `"openai"` |
| `llm.config` | LLM configuration object | `dict` | Required | | `model` | LLM model to use for scoring | `str` | `"gpt-4o-mini"` |
| `top_n` | Number of results to return | `int` | `None` | | `api_key` | API key for LLM provider | `str` | `None` |
| `top_k` | Number of results to return after reranking | `int` | `None` |
| `llm` | Optional nested LLM config (`{"provider": ..., "config": {...}}`) for provider-specific fields (e.g. `ollama_base_url`) | `dict` | `None` |
## Environment Variables ## Environment Variables
docs/components/rerankers/custom-prompts.mdx
+4
View File
@@ -27,6 +27,10 @@ Provide only a single numerical score between 0.0 and 1.0. Do not include any ex
## Custom Prompt Configuration ## Custom Prompt Configuration
<Warning>
The `scoring_prompt` field is **deprecated** and will be removed in a future version. Setting it still works but emits a `DeprecationWarning` at runtime. The field currently sets the system prompt used by the LLM; custom prompts may be exposed through a different API in a future release.
</Warning>
You can provide a custom prompt template using the `scoring_prompt` parameter: You can provide a custom prompt template using the `scoring_prompt` parameter:
```python ```python
docs/components/rerankers/models/cohere.mdx
+2 -2
View File
@@ -74,7 +74,7 @@ os.environ["COHERE_API_KEY"] = "your-api-key"
config = { config = {
"vector_store": {"provider": "chroma"}, "vector_store": {"provider": "chroma"},
"llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}}, "llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}},
"rerank": { "reranker": {
"provider": "cohere", "provider": "cohere",
"config": { "config": {
"model": "rerank-english-v3.0", "model": "rerank-english-v3.0",
@@ -110,7 +110,7 @@ For multilingual applications, use the multilingual model:
```python Python ```python Python
config = { config = {
"rerank": { "reranker": {
"provider": "cohere", "provider": "cohere",
"config": { "config": {
"model": "rerank-multilingual-v3.0", "model": "rerank-multilingual-v3.0",
docs/components/rerankers/models/huggingface.mdx
+11 -17
View File
@@ -35,7 +35,7 @@ m = Memory.from_config(config)
| `device` | str | "cpu" | Device to run model on ("cpu", "cuda", "mps") | | `device` | str | "cpu" | Device to run model on ("cpu", "cuda", "mps") |
| `batch_size` | int | 32 | Batch size for processing | | `batch_size` | int | 32 | Batch size for processing |
| `max_length` | int | 512 | Maximum input sequence length | | `max_length` | int | 512 | Maximum input sequence length |
| `trust_remote_code` | bool | False | Allow remote code execution | | `normalize` | bool | True | Whether to normalize scores to [0, 1] range |
### Advanced Configuration ### Advanced Configuration
@@ -48,10 +48,7 @@ config = {
"device": "cuda", "device": "cuda",
"batch_size": 16, "batch_size": 16,
"max_length": 512, "max_length": 512,
"trust_remote_code": False, "normalize": True
"model_kwargs": {
"torch_dtype": "float16"
}
} }
} }
} }
@@ -154,7 +151,7 @@ m.add("I enjoy reading science fiction books", user_id="alice")
# Search with reranking # Search with reranking
results = m.search( results = m.search(
"What outdoor activities do I enjoy?", "What outdoor activities do I enjoy?",
user_id="alice", filters={"user_id": "alice"},
rerank=True rerank=True
) )
@@ -208,10 +205,7 @@ config = {
"model": "BAAI/bge-reranker-base", "model": "BAAI/bge-reranker-base",
"device": "cpu", "device": "cpu",
"batch_size": 8, # Smaller batch size "batch_size": 8, # Smaller batch size
"max_length": 256, # Shorter sequences "max_length": 256 # Shorter sequences
"model_kwargs": {
"torch_dtype": "float16" # Half precision
}
} }
} }
} }
@@ -232,7 +226,7 @@ config = {
try: try:
results = m.search( results = m.search(
"test query", "test query",
user_id="alice", filters={"user_id": "alice"},
rerank=True rerank=True
) )
except Exception as e: except Exception as e:
@@ -240,7 +234,7 @@ except Exception as e:
# Fall back to vector search only # Fall back to vector search only
results = m.search( results = m.search(
"test query", "test query",
user_id="alice", filters={"user_id": "alice"},
rerank=False rerank=False
) )
``` ```
@@ -251,13 +245,13 @@ except Exception as e:
```python ```python
# Use a private model from Hugging Face # Use a private model from Hugging Face
# Set HF_TOKEN environment variable for private model access
config = { config = {
"reranker": { "reranker": {
"provider": "huggingface", "provider": "huggingface",
"config": { "config": {
"model": "your-org/custom-reranker", "model": "your-org/custom-reranker",
"device": "cuda", "device": "cuda"
"use_auth_token": "your-hf-token"
} }
} }
} }
@@ -311,13 +305,13 @@ config = {
import os import os
os.environ["TRANSFORMERS_CACHE"] = "/path/to/cache" os.environ["TRANSFORMERS_CACHE"] = "/path/to/cache"
# Or use offline mode # Or use offline mode by setting the environment variable
# TRANSFORMERS_OFFLINE=1 before running your script
config = { config = {
"reranker": { "reranker": {
"provider": "huggingface", "provider": "huggingface",
"config": { "config": {
"model": "BAAI/bge-reranker-base", "model": "BAAI/bge-reranker-base"
"local_files_only": True
} }
} }
} }
docs/components/rerankers/models/llm.mdx
+5 -5
View File
@@ -40,7 +40,7 @@ config = {
} }
}, },
"reranker": { "reranker": {
"provider": "llm", "provider": "llm_reranker",
"config": { "config": {
"model": "gpt-4o-mini", "model": "gpt-4o-mini",
"provider": "openai", "provider": "openai",
@@ -90,7 +90,7 @@ config = {
"vector_store": {"provider": "chroma"}, "vector_store": {"provider": "chroma"},
"llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}}, "llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}},
"reranker": { "reranker": {
"provider": "llm", "provider": "llm_reranker",
"config": { "config": {
"model": "gpt-4o-mini", "model": "gpt-4o-mini",
"provider": "openai", "provider": "openai",
@@ -150,7 +150,7 @@ Score from 0.0 to 1.0. Provide only the numerical score."""
config = { config = {
"reranker": { "reranker": {
"provider": "llm", "provider": "llm_reranker",
"config": { "config": {
"model": "gpt-4o-mini", "model": "gpt-4o-mini",
"provider": "openai", "provider": "openai",
@@ -169,7 +169,7 @@ Use different LLM providers for reranking:
# Using Anthropic Claude # Using Anthropic Claude
anthropic_config = { anthropic_config = {
"reranker": { "reranker": {
"provider": "llm", "provider": "llm_reranker",
"config": { "config": {
"model": "claude-3-haiku-20240307", "model": "claude-3-haiku-20240307",
"provider": "anthropic", "provider": "anthropic",
@@ -181,7 +181,7 @@ anthropic_config = {
# Using local Ollama model # Using local Ollama model
ollama_config = { ollama_config = {
"reranker": { "reranker": {
"provider": "llm", "provider": "llm_reranker",
"config": { "config": {
"model": "llama2:7b", "model": "llama2:7b",
"provider": "ollama", "provider": "ollama",
docs/components/rerankers/models/llm_reranker.mdx
+1 -1
View File
@@ -268,7 +268,7 @@ m.add("I'm vegetarian", user_id="alice")
# Search with LLM reranking # Search with LLM reranking
results = m.search( results = m.search(
"What foods should I avoid?", "What foods should I avoid?",
user_id="alice", filters={"user_id": "alice"},
rerank=True rerank=True
) )
docs/components/rerankers/models/sentence_transformer.mdx
+4 -4
View File
@@ -39,7 +39,7 @@ config = {
"model": "gpt-4o-mini" "model": "gpt-4o-mini"
} }
}, },
"rerank": { "reranker": {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
@@ -60,7 +60,7 @@ For better performance, use GPU acceleration:
```python Python ```python Python
config = { config = {
"rerank": { "reranker": {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
@@ -80,7 +80,7 @@ from mem0 import Memory
config = { config = {
"vector_store": {"provider": "chroma"}, "vector_store": {"provider": "chroma"},
"llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}}, "llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}},
"rerank": { "reranker": {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
@@ -117,7 +117,7 @@ You can use any HuggingFace cross-encoder model:
```python Python ```python Python
# Using a different model # Using a different model
config = { config = {
"rerank": { "reranker": {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"model": "cross-encoder/stsb-distilroberta-base", "model": "cross-encoder/stsb-distilroberta-base",
docs/components/rerankers/models/zero_entropy.mdx
+2 -2
View File
@@ -37,7 +37,7 @@ config = {
"model": "gpt-4o-mini" "model": "gpt-4o-mini"
} }
}, },
"rerank": { "reranker": {
"provider": "zero_entropy", "provider": "zero_entropy",
"config": { "config": {
"model": "zerank-1", # or "zerank-1-small" "model": "zerank-1", # or "zerank-1-small"
@@ -71,7 +71,7 @@ os.environ["ZERO_ENTROPY_API_KEY"] = "your-api-key"
config = { config = {
"vector_store": {"provider": "chroma"}, "vector_store": {"provider": "chroma"},
"llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}}, "llm": {"provider": "openai", "config": {"model": "gpt-4o-mini"}},
"rerank": {"provider": "zero_entropy", "config": {"model": "zerank-1"}} "reranker": {"provider": "zero_entropy", "config": {"model": "zerank-1"}}
} }
memory = Memory.from_config(config) memory = Memory.from_config(config)
docs/components/rerankers/optimization.mdx
+19 -28
View File
@@ -13,10 +13,10 @@ The number of candidates sent to the reranker significantly impacts performance:
```python ```python
# Optimal candidate sizes for different rerankers # Optimal candidate sizes for different rerankers
config_map = { config_map = {
"cohere": {"initial_candidates": 100, "top_n": 10}, "cohere": {"initial_candidates": 100, "top_k": 10},
"sentence_transformer": {"initial_candidates": 50, "top_n": 10}, "sentence_transformer": {"initial_candidates": 50, "top_k": 10},
"huggingface": {"initial_candidates": 30, "top_n": 5}, "huggingface": {"initial_candidates": 30, "top_k": 5},
"llm_reranker": {"initial_candidates": 20, "top_n": 5} "llm_reranker": {"initial_candidates": 20, "top_k": 5}
} }
``` ```
@@ -31,7 +31,7 @@ config = {
"config": { "config": {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
"batch_size": 16, # Process multiple candidates at once "batch_size": 16, # Process multiple candidates at once
"top_n": 10 "top_k": 10
} }
} }
} }
@@ -48,7 +48,7 @@ config = {
"provider": "cohere", "provider": "cohere",
"config": { "config": {
"model": "rerank-english-v3.0", "model": "rerank-english-v3.0",
"top_n": 10, "top_k": 10,
"max_chunks_per_doc": 10, # Limit chunk processing "max_chunks_per_doc": 10, # Limit chunk processing
"return_documents": False # Reduce response size "return_documents": False # Reduce response size
} }
@@ -73,7 +73,7 @@ config = {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
"device": "cuda", # Use GPU when available "device": "cuda", # Use GPU when available
"batch_size": 32, "batch_size": 32,
"top_n": 10, "top_k": 10,
"max_length": 512 # Limit input length "max_length": 512 # Limit input length
} }
} }
@@ -107,10 +107,9 @@ config = {
"provider": "huggingface", "provider": "huggingface",
"config": { "config": {
"model": "BAAI/bge-reranker-base", "model": "BAAI/bge-reranker-base",
"use_fp16": True, # Half precision for speed
"max_length": 512, "max_length": 512,
"batch_size": 8, "batch_size": 8,
"top_n": 10 "top_k": 10
} }
} }
} }
@@ -124,17 +123,11 @@ config = {
"reranker": { "reranker": {
"provider": "llm_reranker", "provider": "llm_reranker",
"config": { "config": {
"llm": { "provider": "openai",
"provider": "openai", "model": "gpt-4o-mini", # Faster than gpt-4o
"config": { "temperature": 0.0, # Deterministic results
"model": "gpt-3.5-turbo", # Faster than gpt-4 "max_tokens": 100, # Limit response length
"temperature": 0, # Deterministic results "top_k": 5 # Fewer results for faster processing
"max_tokens": 500 # Limit response length
}
},
"batch_ranking": True, # Rank multiple at once
"top_n": 5, # Fewer results for faster processing
"timeout": 10 # Request timeout
} }
} }
} }
@@ -201,12 +194,12 @@ class CachedReranker:
### Model Caching ### Model Caching
```python ```python
# Pre-load models to avoid initialization overhead # Pre-load models to avoid initialization overhead
# Set SENTENCE_TRANSFORMERS_HOME env var to control cache location
config = { config = {
"reranker": { "reranker": {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
"cache_folder": "/path/to/model/cache",
"device": "cuda" "device": "cuda"
} }
} }
@@ -218,14 +211,14 @@ config = {
### Async Configuration ### Async Configuration
```python ```python
import asyncio import asyncio
from mem0 import Memory from mem0 import AsyncMemory
async def parallel_search(config, queries, user_id): async def parallel_search(config, queries, user_id):
memory = Memory.from_config(config) memory = AsyncMemory.from_config(config)
# Process multiple queries concurrently # Process multiple queries concurrently
tasks = [ tasks = [
memory.search_async(query, user_id=user_id) memory.search(query, filters={"user_id": user_id})
for query in queries for query in queries
] ]
@@ -249,8 +242,7 @@ config = {
"config": { "config": {
"device": "cuda", "device": "cuda",
"model": "cross-encoder/ms-marco-electra-base", "model": "cross-encoder/ms-marco-electra-base",
"batch_size": 64, # Larger batch for GPU "batch_size": 64 # Larger batch for GPU
"fp16": True # Half precision
} }
} }
} }
@@ -268,8 +260,7 @@ config = {
"provider": "sentence_transformer", "provider": "sentence_transformer",
"config": { "config": {
"device": "cpu", "device": "cpu",
"model": "cross-encoder/ms-marco-MiniLM-L-6-v2", "model": "cross-encoder/ms-marco-MiniLM-L-6-v2"
"num_workers": 4 # Parallel processing
} }
} }
} }