Experiencing GraphRag with OpenAI Models - An Example from The Borderlands
Before using SiliconCloud, see how GraphRag works using OpenAI's model.
GraphRAG is an AI-based content understanding and search capability that utilizes LLMs, parses data to create knowledge graphs, and answers user questions on user-supplied private datasets.
GitHub Address:/microsoft/graphrag
Official website:/graphrag
Now it's time to officially start experiencing GraphRag.
hint
The consumption of GraphRag Token is relatively large, and at the beginning of the experience you can not follow the official configuration, switching to a text with fewer words as well as switching to the gpt-4o-mini model.
Take Shen Congwen's Border Town as an example.
Create a Python virtual environment and install GraphRag:
pip install graphrag
After the installation:
mkdir biancheng
mkdir input
It is to create two folders, or you can do it manually, and then put the Borderlands txt file into the input folder as shown below:
Start initialization:
python -m --init --root ./biancheng
Upon completion, a number of files will appear, as shown below:
Enter the OpenAI Api Key in the .env file as shown below:
Do some configurations in the file, here mine is as follows:
encoding_model: cl100k_base
skip_workflows: []
llm:
api_key: ${GRAPHRAG_API_KEY}
type: openai_chat # or azure_openai_chat
model: gpt-4o-mini
model_supports_json: true # recommended if this is available for your model.
# max_tokens: 4000
# request_timeout: 180.0
# api_base: https://<instance>.
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# temperature: 0 # temperature for sampling
# top_p: 1 # top-p sampling
# n: 1 # Number of completions to generate
parallelization:
stagger: 0.3
# num_threads: 50 # the number of threads to use for parallel processing
async_mode: threaded # or asyncio
embeddings:
## parallelization: override the global parallelization settings for embeddings
async_mode: threaded # or asyncio
llm:
api_key: ${GRAPHRAG_API_KEY}
type: openai_embedding # or azure_openai_embedding
model: text-embedding-3-small
# api_base: https://<instance>.
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# batch_size: 16 # the number of documents to send in a single request
# batch_max_tokens: 8191 # the maximum number of tokens to send in a single request
# target: required # or optional
chunks:
size: 1200
overlap: 100
group_by_columns: [id] # by default, we don't allow chunks to cross documents
input:
type: file # or blob
file_type: text # or csv
base_dir: "input"
file_encoding: utf-8
file_pattern: ".*\\.txt$"
cache:
type: file # or blob
base_dir: "cache"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
storage:
type: file # or blob
base_dir: "output/${timestamp}/artifacts"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
reporting:
type: file # or console, blob
base_dir: "output/${timestamp}/reports"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
entity_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/entity_extraction.txt"
entity_types: [organization,person,geo,event]
max_gleanings: 1
summarize_descriptions:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/summarize_descriptions.txt"
max_length: 500
claim_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
# enabled: true
prompt: "prompts/claim_extraction.txt"
description: "Any claims or facts that could be relevant to information discovery."
max_gleanings: 1
community_reports:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/community_report.txt"
max_length: 2000
max_input_length: 8000
cluster_graph:
max_cluster_size: 10
embed_graph:
enabled: false # if true, will generate node2vec embeddings for nodes
# num_walks: 10
# walk_length: 40
# window_size: 2
# iterations: 3
# random_seed: 597832
umap:
enabled: false # if true, will generate UMAP embeddings for nodes
snapshots:
graphml: true
raw_entities: false
top_level_nodes: false
local_search:
# text_unit_prop: 0.5
# community_prop: 0.1
# conversation_history_max_turns: 5
# top_k_mapped_entities: 10
# top_k_relationships: 10
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
global_search:
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
# data_max_tokens: 12000
# map_max_tokens: 1000
# reduce_max_tokens: 2000
# concurrency: 32
Swapped the model for a gpt-4o-mini to save money:
Changed to true here in order to view the graphml file later in Gephi and other software:
This is configured and indexing now begins:
python -m --root ./biancheng
Screenshot of indexing completion:
The generated nodes and edges can now be viewed:
You can start your query now.
Let's start with the global query:
python -m --root . /biancheng --method global "What theme is this story about?"
Again, a partial query:
python -m --root . /biancheng --method local "What happened to Cuicui in Baiji Pass?"
The word count for Sideways is somewhere between 50,000 and 60,000 words, check the cost:
For only $0.18, the gpt-4o-mini is still a great value for the money.
Trying out GraphRag with SiliconCloud - Using The Three Kingdoms as an Example
Although the use of OpenAI models is very good, but in the domestic use of OpenAI will have some restrictions, many people may not have OpenAI Api Key, and may not be able to get, so you can choose SiliconCloud as an alternative, SiliconCloud also provides a compatible OpenAI format of the dialog model and embedded models, and a number of advanced open source models available. SiliconCloud also provides OpenAI-compatible dialog models and embedding models, and has a variety of advanced open-source models available, just sign up for SiliconCloud will send some credits, interested in trying it right away.
When trying out GraphRag with SiliconCloud, try switching to a smaller text in order to quickly run the process through, starting with the story of Chang'e Runs to the Moon as an example.
The steps are the same as the previous ones, except that you have to change a few things when configuring it.
First change the Api Key to SiliconCloud's Api Key:
Settings where changes are needed.
First is the dialog modeling section:
Here I choose the model meta-Llama/Meta-Llama-3.1-70B-Instruct, about how to write the name of the model, refer to SiliconCloud's documentation, documentation address:/docs/getting-started
Next is the embedded model section:
The embedding model used here is BAAI/bge-large-en-v1.5, using BAAI/bge-large-zh-v1.5 I get an error here, you can also try it, I don't know what's the reason at the moment.
See also the documentation for what to write for the embedded model name:
Begin indexing:
View node:
View side:
Global question:
python -m --root . /change1 --method global "What theme is this story about?"
Partial questioning:
python -m --root . /change1 --method local "What gift did Chang'e send to the emperor?"
Now that the process is run through, it's time to try The Triple Play!
Use the same settings, the three kingdoms have more words and are slower, be patient:
The process is complete:
View node:
View side:
Global question:
python -m --root . /sanguo --method global "What stories do the three kingdoms tell?"
Partial questioning:
python -m --root . /sanguo --method local "How was Cao Cao defeated at the Battle of Red Cliff?"
Trying GraphRag with a local model
Local attempts at GraphRag can use Ollama's dialog model, and since Ollama's embedded model is not compatible with OpenAI's format, the embedded model can use LM Studio.
Configuration:
encoding_model: cl100k_base
skip_workflows: []
llm:
api_key: ${GRAPHRAG_API_KEY}
type: openai_chat # or azure_openai_chat
model: llama3.1:70b
model_supports_json: true # recommended if this is available for your model.
# max_tokens: 4000
# request_timeout: 180.0
api_base: http://localhost:11434/v1
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# temperature: 0 # temperature for sampling
# top_p: 1 # top-p sampling
# n: 1 # Number of completions to generate
parallelization:
stagger: 0.3
# num_threads: 50 # the number of threads to use for parallel processing
async_mode: threaded # or asyncio
embeddings:
## parallelization: override the global parallelization settings for embeddings
async_mode: threaded # or asyncio
llm:
api_key: ${GRAPHRAG_API_KEY}
type: openai_embedding # or azure_openai_embedding
model: nomic-ai/nomic-embed-text-v1.5-GGUF/nomic-embed-text-v1.5.Q2_K.gguf
api_base: http://localhost:1234/v1
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# batch_size: 16 # the number of documents to send in a single request
# batch_max_tokens: 8191 # the maximum number of tokens to send in a single request
# target: required # or optional
chunks:
size: 300
overlap: 100
group_by_columns: [id] # by default, we don't allow chunks to cross documents
input:
type: file # or blob
file_type: text # or csv
base_dir: "input"
file_encoding: utf-8
file_pattern: ".*\\.txt$"
cache:
type: file # or blob
base_dir: "cache"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
storage:
type: file # or blob
base_dir: "output/${timestamp}/artifacts"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
reporting:
type: file # or console, blob
base_dir: "output/${timestamp}/reports"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
entity_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/entity_extraction.txt"
entity_types: [organization,person,geo,event]
max_gleanings: 1
summarize_descriptions:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/summarize_descriptions.txt"
max_length: 500
claim_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
# enabled: true
prompt: "prompts/claim_extraction.txt"
description: "Any claims or facts that could be relevant to information discovery."
max_gleanings: 1
community_reports:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/community_report.txt"
max_length: 2000
max_input_length: 8000
cluster_graph:
max_cluster_size: 10
embed_graph:
enabled: false # if true, will generate node2vec embeddings for nodes
# num_walks: 10
# walk_length: 40
# window_size: 2
# iterations: 3
# random_seed: 597832
umap:
enabled: false # if true, will generate UMAP embeddings for nodes
snapshots:
graphml: false
raw_entities: false
top_level_nodes: false
local_search:
# text_unit_prop: 0.5
# community_prop: 0.1
# conversation_history_max_turns: 5
# top_k_mapped_entities: 10
# top_k_relationships: 10
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
global_search:
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
# data_max_tokens: 12000
# map_max_tokens: 1000
# reduce_max_tokens: 2000
# concurrency: 32
Theoretically it runs, but my computer configuration is not good enough to run a slightly larger model, so I can't test it in real life.
mix
You can access the online dialog model Api and use the local one for the embedded model, but SiliconCloud is currently free to use for the embedded model, or you can directly use SiliconCloud's embedded model.
What should I do in order to test which models can run the GraphRag process through, but some vendors only provide dialog models without embedding models or the embedding models provided are not compatible with the OpenAI format?
Two Keys can be used, one Key for SiliconCloud to use the embedding model and one Key from another vendor to use the dialog model.
For example, it can be set up like this:
The configuration file can be written like this:
encoding_model: cl100k_base
skip_workflows: []
llm:
api_key: ${Other_API_KEY}
type: openai_chat # or azure_openai_chat
model: glm-4-air
model_supports_json: true # recommended if this is available for your model.
# max_tokens: 4000
# request_timeout: 180.0
api_base: /api/paas/v4
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# temperature: 0 # temperature for sampling
# top_p: 1 # top-p sampling
# n: 1 # Number of completions to generate
parallelization:
stagger: 0.3
# num_threads: 50 # the number of threads to use for parallel processing
async_mode: threaded # or asyncio
embeddings:
## parallelization: override the global parallelization settings for embeddings
async_mode: threaded # or asyncio
llm:
api_key: ${GRAPHRAG_API_KEY}
type: openai_embedding # or azure_openai_embedding
model: BAAI/bge-large-en-v1.5
api_base: /v1
# api_version: 2024-02-15-preview
# organization: <organization_id>
# deployment_name: <azure_model_deployment_name>
# tokens_per_minute: 150_000 # set a leaky bucket throttle
# requests_per_minute: 10_000 # set a leaky bucket throttle
# max_retries: 10
# max_retry_wait: 10.0
# sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
# concurrent_requests: 25 # the number of parallel inflight requests that may be made
# batch_size: 16 # the number of documents to send in a single request
# batch_max_tokens: 8191 # the maximum number of tokens to send in a single request
# target: required # or optional
chunks:
size: 300
overlap: 100
group_by_columns: [id] # by default, we don't allow chunks to cross documents
input:
type: file # or blob
file_type: text # or csv
base_dir: "input"
file_encoding: utf-8
file_pattern: ".*\\.txt$"
cache:
type: file # or blob
base_dir: "cache"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
storage:
type: file # or blob
base_dir: "output/${timestamp}/artifacts"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
reporting:
type: file # or console, blob
base_dir: "output/${timestamp}/reports"
# connection_string: <azure_blob_storage_connection_string>
# container_name: <azure_blob_storage_container_name>
entity_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/entity_extraction.txt"
entity_types: [organization,person,geo,event]
max_gleanings: 1
summarize_descriptions:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/summarize_descriptions.txt"
max_length: 500
claim_extraction:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
# enabled: true
prompt: "prompts/claim_extraction.txt"
description: "Any claims or facts that could be relevant to information discovery."
max_gleanings: 1
community_reports:
## llm: override the global llm settings for this task
## parallelization: override the global parallelization settings for this task
## async_mode: override the global async_mode settings for this task
prompt: "prompts/community_report.txt"
max_length: 2000
max_input_length: 8000
cluster_graph:
max_cluster_size: 10
embed_graph:
enabled: false # if true, will generate node2vec embeddings for nodes
# num_walks: 10
# walk_length: 40
# window_size: 2
# iterations: 3
# random_seed: 597832
umap:
enabled: false # if true, will generate UMAP embeddings for nodes
snapshots:
graphml: true
raw_entities: false
top_level_nodes: false
local_search:
# text_unit_prop: 0.5
# community_prop: 0.1
# conversation_history_max_turns: 5
# top_k_mapped_entities: 10
# top_k_relationships: 10
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
global_search:
# llm_temperature: 0 # temperature for sampling
# llm_top_p: 1 # top-p sampling
# llm_n: 1 # Number of completions to generate
# max_tokens: 12000
# data_max_tokens: 12000
# map_max_tokens: 1000
# reduce_max_tokens: 2000
# concurrency: 32
I've tried multiple big models, and after this simple test of mine, the ones that run the GraphRag process through (just through, the answers are not always good) are the following:
hint
GraphRag Token consumption is high, so watch your quota!
For a text of more than 2,000 words, a single GraphRag basically costs more than 100,000 Token:
consultation
1、/zh-cn/siliconcloud
2、/microsoft/graphrag/discussions/321
3、/microsoft/graphrag/issues/374
4、/watch?v=BLyGDTNdad0