Post Syndicated from Pat Patterson original https://www.backblaze.com/blog/building-a-conversational-ai-chatbot-website-with-backblaze-b2-langchain/

In an earlier blog post, I explained how to build your own LLM with Backblaze B2 + Jupyter Notebook, implementing a simple conversational AI chatbot using the LangChain AI framework to implement retrieval-augmented generation (RAG). The notebook walks you through the process of loading PDF files from a Backblaze B2 Bucket into a vector store, running a local instance of a large language model (LLM) and combining those to form a chatbot that can answer questions on its specialist subject.

That article generated a lot of interest, and a few questions:

• “Could you make this into a web app, like ChatGPT?”
• “Could you use this with OpenAI? DeepSeek?”
• “Could I load multiple collections of documents into this?”
• “Could I run multiple LLMs and compare them?”
• “Can I add new documents to the vector store as they are uploaded to the bucket?”

The answer to all of these questions is “Yes!”

Today, I’ll present a simple conversational AI chatbot web app with a ChatGPT-style UI that you can easily configure to work with OpenAI, DeepSeek, or any of a range of other LLMs. In future blog posts, I’ll extend this to allow you to configure multiple LLMs and document collections, and integrate with Backblaze B2’s Event Notifications feature to load documents into the vector store within seconds of them being uploaded.

And, here’s a very short video of the chatbot in action:

RAG basics

Retrieval-augmented generation, or RAG for short, is a technique that applies the generative features of an LLM to a collection of documents, resulting in a chatbot that can effectively answer questions based on the content of those documents.

A typical RAG implementation splits each document in the collection into a number of roughly equal-sized, overlapping chunks, and generates an embedding for each chunk. Embeddings are vectors (lists) of floating point numbers with hundreds or thousands of dimensions. The distance between two vectors indicates their similarity. Small distances indicate high similarity and large distances indicate low similarity.

The RAG app then loads each chunk, along with its embedding, into a vector store. The vector store is a special-purpose database that can perform a similarity search–given a piece of text, the vector store can retrieve chunks ranked by their similarity to the query text by comparing the embeddings.

Let’s put the pieces together:

Given a question from the user (1), the RAG app can query the vector store for chunks of text that are similar to the question (2). This will be the context that helps the LLM answer the user’s question. Here’s a concrete example using the Backblaze documentation collection: Given the question, “Tell me about object lock,” the vector store returns four document chunks, each of about 170 words, to the app (3). Here is a link to the text of, and a short extract from, each chunk:

• Object Lock: With Object Lock Legal Hold, files are prevented from being changed or deleted, but the immutability does not have a defined retention period.
• Object Lock: You can enable Object Lock on a bucket when you create a new bucket or on an existing bucket.
• Object Lock: Object Lock in Backblaze B2 was designed with the Immutability feature in Veeam Backup & Replication in mind.
• Enable Object Lock with the S3 Compatible API: To enable Object Lock on an existing S3 compatible bucket, add the x-amz-bucket-object-lock-token header with a value of 1 to the S3 Put Object Lock Configuration API call.

Unsurprisingly, three of the four results are extracts from the main documentation article on Object Lock.

In order to provide a conversational experience, the RAG app maintains a chat history in a message store. The RAG app retrieves the message history (4) to form part of the query it will send to the LLM.

A system prompt sets the guidelines for the LLM to generate responses. In a RAG, the system prompt typically specifies that the response should be based on retrieved document data rather than model memorization, and attempts to prevent hallucinations by ensuring that the model does not fabricate information if no relevant documents are found. Here is the system prompt from the RAG app:

Use the following pieces of context and the message history to answer the question at the end. If you don't know the answer,  just say that you don't know, don't try to make up an answer.

The RAG submits the system prompt, the context, the last few messages in the conversation, and the user’s question to the LLM (5), which generates a, hopefully useful, response (6). Here’s the answer it provided to the above question:

The RAG app adds the user’s question and the LLM’s response to the message store (7), returns the answer to the user (8), and awaits the next question.

A quick tour of the sample app

The sample app is on GitHub at https://github.com/backblaze-b2-samples/ai-rag-app. The app is open source, under the MIT license, so you can use it as a basis for your own experimentation without any restrictions. The app was originally written to demonstrate RAG with Backblaze B2 Cloud Storage, but it works with any S3 compatible object store.  

The README file covers configuration and deployment in some detail; in this blog post, I’ll just give you a high-level overview. The sample app is written in Python using the Django web framework. API credentials and related settings are configured via environment variables, while the LLM and vector store are configured via Django’s settings.py file:

CHAT_MODEL: ModelSpec = {
    'name': 'OpenAI',
    'llm': {
        'cls': ChatOpenAI,
        'init_args': {
            'model': "gpt-4o-mini",
        }
    },
}

# Change source_data_location and vector_store_location to match your environment
# search_k is the number of results to return when searching the vector store
DOCUMENT_COLLECTION: CollectionSpec = {
    'name': 'Docs',
    'source_data_location': 's3://blze-ev-ai-rag-app/pdfs',
    'vector_store_location': 's3://blze-ev-ai-rag-app/vectordb/docs/openai',
    'search_k': 4,
    'embeddings': {
        'cls': OpenAIEmbeddings,
        'init_args': {
            'model': "text-embedding-3-large",
        },
    },
}

The sample app is configured to use OpenAI GPT-4o mini, but the README explains how to use different online LLMs such as DeepSeek V3 or Google Gemini 2.0 Flash, or even a local LLM such as Meta Llama 3.1 via the Ollama framework. If you do run a local LLM, be sure to pick a model that fits your hardware. I tried running Meta’s Llama 3.3, which has 70 billion parameters (70B), on my MacBook Pro with the M1 Pro CPU. It took nearly three hours to answer a single question! Llama 3.1 8B was a much better fit, answering questions in less than 30 seconds.

Notice that the document collection is configured with the location of a vector store containing the Backblaze documentation as a sample dataset. The README file contains an application key with read-only access to the PDFs and vector store so you can try the application without having to load your own set of documents.

If you want to use your own document collection, a pair of custom commands allow you to load them from a Backblaze B2 Bucket into the vector store and then query the vector store to test that it all worked.

First, you need to load your data:

% python manage.py load_vector_store
Deleting existing LanceDB vector store at s3://blze-ev-ai-rag-app/vectordb/docs
Creating LanceDB vector store at s3://blze-ev-ai-rag-app/vectordb/docs
Loading data from s3://blze-ev-ai-rag-app/pdfs in pages of 1000 results
Successfully retrieved page 1 containing 618 result(s) from s3://blze-ev-ai-rag-app/pdfs
Skipping pdfs/.bzEmpty
Skipping pdfs/cloud_storage/.bzEmpty
Loading pdfs/cloud_storage/cloud-storage-about-backblaze-b2-cloud-storage.pdf
Loading pdfs/cloud_storage/cloud-storage-add-file-information-with-the-native-api.pdf
Loading pdfs/cloud_storage/cloud-storage-additional-resources.pdf
...
Loading pdfs/v1_api/s3-put-object.pdf
Loading pdfs/v1_api/s3-upload-part-copy.pdf
Loading pdfs/v1_api/s3-upload-part.pdf
Loaded batch of 614 document(s) from page
Split batch into 2758 chunks
[2025-02-28T01:26:11Z WARN  lance_table::io::commit] Using unsafe commit handler. Concurrent writes may result in data loss. Consider providing a commit handler that prevents conflicting writes.
Added chunks to vector store
Added 614 document(s) containing 2758 chunks to vector store; skipped 4 result(s).
Created LanceDB vector store at s3://blze-ev-ai-rag-app/vectordb/docs. "vectorstore" table contains 2758 rows

Now you can verify that the data is stored by querying the vector store. Notice how the raw results from the vector store include an S3 URI identifying the source document:

% python manage.py search_vector_store 'Which B2 native APIs would I use to upload large files?' 
2025-03-01 02:38:07,740 ai_rag_app.management.commands.search INFO     Opening vector store at s3://blze-ev-ai-rag-app/vectordb/docs/openai
2025-03-01 02:38:07,740 ai_rag_app.utils.vectorstore DEBUG    Populating AWS environment variables from the b2 profile
Found 4 docs in 2.30 seconds
2025-03-01 02:38:11,074 ai_rag_app.management.commands.search INFO     
page_content='Parts of a large file can be uploaded and copied in parallel, which can significantly reduce the time it takes to upload terabytes of data. Each part can be  anywhere from 5 MB to 5 GB, and you can pick the size that is most convenient for your application. For best upload performance, Backblaze recommends that you use the recommendedPartSize parameter that is returned by the b2_authorize_account operation.  To upload larger files and data sets, you can use the command-line interface (CLI), the Native API, or an integration, such as Cyberduck.  Usage for Large Files  Generally, large files are treated the same as small files. The costs for the API calls are the same.  You are charged for storage for the parts that you uploaded or copied. Usage is counted from the time the part is stored. When you call the b2_finish_large_file' metadata={'source': 's3://blze-ev-ai-rag-app/pdfs/cloud_storage/cloud-storage-large-files.pdf'}
...

The core of the sample application is the RAG class. There are several methods that create the basic components of the RAG, but here we’ll look at how the _create_chain() method brings together the system prompt, vector store, message history, and LLM.

First, we define the system prompt, which includes a placeholder for the context—those chunks of text that the RAG will retrieve from the vector store:

# These are the basic instructions for the LLM
system_prompt = (
    "Use the following pieces of context and the message history to "
    "answer the question at the end. If you don't know the answer, "
    "just say that you don't know, don't try to make up an answer. "
    "\n\n"
    "Context: {context}"
)

Then we create a prompt template that brings together the system prompt, message history, and the user’s question:

# The prompt template brings together the system prompt, context, message history and the user's question
prompt_template = ChatPromptTemplate(
    [
        ("system", system_prompt),
        MessagesPlaceholder(variable_name="history", optional=True, n_messages=10),
        ("human", "{question}"),
    ]
)

Now we use LangChain Expression Language (LCEL) to bring the various components together to form a chain. LCEL allows us to define a chain of components declaratively; that is, we provide a high-level representation of the chain we want, rather than specifying how the components should fit together. 

Notice the log_data() helper method—it simply logs its input and passes it on to the next component in the chain.

# Create the basic chain
# When loglevel is set to DEBUG, log_input will log the results from the vector store
chain = (
    {
        "context": (
            itemgetter("question")
            | retriever
            | log_data('Documents from vector store', pretty=True)
        ),
        "question": itemgetter("question"),
        "history": itemgetter("history"),
    }
    | prompt_template
    | model
    | log_data('Output from model', pretty=True)
)

Assigning a name to the chain allows us to add instrumentation when we invoke it:

# Give the chain a name so the handler can see it
named_chain: Runnable[Input, Output] = chain.with_config(run_name="my_chain")

Now, we use LangChain’s RunnableWithMessageHistory class to manage adding and retrieving messages from the message store:

# Add message history management
return RunnableWithMessageHistory(
    named_chain,
    lambda session_id: RAG._get_session_history(store, session_id),
    input_messages_key="question",
    history_messages_key="history",
)

Finally, the log_chain() function prints an ASCII representation of the chain to the debug log:

log_chain(history_chain, logging.DEBUG, {"configurable": {'session_id': 'dummy'}})

This is the output:

The RAG class’ invoke() function, in contrast, is very simple. Here is the key section of code:

response = self._chain.invoke(
    {"question": question},
    config={
        "configurable": {
            "session_id": session_key
        },
        "callbacks": [
            ChainElapsedTime("my_chain")
        ]
    },
)

The input to the chain is a Python dictionary containing the question, while the config argument configures the chain with the Django session key and a callback that annotates the chain output with its execution time. Since the chain output contains Markdown formatting, the API endpoint that handles requests from the front end uses the open source markdown-it library to render the output to HTML for display.

The remainder of the code is mostly concerned with rendering the web UI. One interesting facet is that the Django view, responsible for rendering the UI as the page loads, uses the RAG’s message store to render the conversation, so if you reload the page, you don’t lose your context.

Take this code and run it!

The sample AI RAG application is open source under the MIT license, and I encourage you to use it as the basis for your own RAG exploration. The README file suggests a few ways you could extend it, and I also draw your attention to conclusion of the README if you are thinking of running the app in production:

Above all, have fun! AI is a rapidly evolving technology, with vendors and open source projects releasing new capabilities every day. I hope you find this app a useful way of jumping in.

The post Building a Conversational AI Chatbot Website with Backblaze B2 + LangChain appeared first on Backblaze Blog | Cloud Storage & Cloud Backup

Post Syndicated from Pat Patterson original https://www.backblaze.com/blog/experimenting-with-deepseek-backblaze-b2-and-drive-stats/

As we explained in our recent blog post, AI Reasoning Models: OpenAI o3-mini, o1-mini, and DeepSeek R1, Chinese startup DeepSeek caused a stir when it released its R1 reasoning model in January of this year. Interestingly, DeepSeek R1 has an OpenAI-compatible API, so applications written for OpenAI should work with DeepSeek R1 with just a configuration change. Since I had a suitable sample app all ready to go, I decided to put their claim to the test.

Why, and why not, use DeepSeek?

A major difference between DeepSeek and OpenAI is cost. At the time of writing, DeepSeek charges $0.55 per million input tokens and $2.19 per million output tokens for its R1 model. That’s about 3.6% of OpenAI’s $15.00 per million input tokens and $60.00 per million output tokens for its flagship o1 reasoning model, and about half of o3-mini’s $1.10 per million input tokens and $4.40 per million output tokens.

Set against this is the fact that, in using the DeepSeek platform’s API, you are sending your data to a startup located in China that has been accused by OpenAI of “inappropriately” basing its work on the output of OpenAI’s models. It’s up to you, and your organizations’ data governance policy, whether the trade-off is worthwhile.

Another consideration is the ability to run DeepSeek’s models locally, on your own infrastructure, or, more likely, your chosen provider’s infrastructure, rather than sending requests to the DeepSeek platform. Spinning up my own DeepSeek instance was out of scope for this blog post, but I’ll likely return to it in a future blog post.

Swapping OpenAI for DeepSeek

Last month, I explained how you can build an AI agent with Backblaze B2, LangChain, and Drive Stats, walking you through a simple chatbot that can answer questions based on our Drive Stats data set—11 years of metrics gathered from the Backblaze B2 Cloud Storage platform’s fleet of hard drives. In that example, the chatbot accepted a natural language question, used OpenAI’s GPT‑4o mini large language model (LLM) to generate a SQL query that might help provide an answer, executed the query against the Drive Stats data set via the Trino SQL engine, and then used OpenAI again to interpret the result set and either repeat the query-interpret cycle, or generate a natural language answer.

I copied the Jupyter notebook from that example and used it as the basis for investigating the feasibility of swapping out OpenAI for DeepSeek. The DeepSeek version of the notebook contains the full source code of my experiments; I’ll include relevant extracts here, edited for clarity.

Since I used the LangChain AI framework, which provides a layer above a range of AI models, the only place that OpenAI surfaced in my code was in creating an instance of LangChain’s ChatOpenAI wrapper:

# OPENAI_API_KEY must be defined in the .env file
load_dotenv()
llm = ChatOpenAI(model="gpt-4o-mini")

The ChatOpenAI class contains all the code required to communicate with OpenAI via its API.

According to the DeepSeek documentation, all you should need to do is:

• Provide your DeepSeek API key in the same OPENAI_API_KEY environment variable.
• Set the API base URL to https://api.deepseek.com.
• Provide a DeepSeek model name in place of the OpenAI one.

If this reminds you of the steps for using Backblaze B2’s S3-compatible API, you’re not alone. The OpenAI API has become a de facto standard for integrating with LLMs in much the same way as Amazon’s S3 API allows an ecosystem of apps and tools to interoperate with object storage systems from a variety of vendors.

Looking at the DeepSeek documentation, you can use one of two models, deepseek-reasoner (aka DeepSeek R1) or deepseek-chat. Let’s see what the much-talked-about DeepSeek R1 came up with.

Using DeepSeek R1 in the AI agent

To make it easy to use both the OpenAI and DeepSeek notebooks, I created a second entry in the .env file for the DeepSeek API key, and copied it to the OpenAI environment variable in the notebook code:

# The .env file needs at least DEEPSEEK_API_KEY, and may also contain
# OPENAI_API_KEY. Move the DeepSeek API key to the OpenAI environment
# variable
load_dotenv()

os.environ["OPENAI_API_KEY"] = os.environ.pop("DEEPSEEK_API_KEY")

llm = ChatOpenAI(model="deepseek-reasoner", base_url='https://api.deepseek.com')

As I set about repeating the steps from the Jupyter notebook that supported my previous blog post, I was disappointed to see DeepSeek fall at the very first hurdle: generating a SQL query for a simple natural language question. Here is the code:

question = {"question": "How many drives are there?"}

write_query(question)

Looking back at the original notebook, OpenAI’s response was valid SQL, although it didn’t have enough information to construct the correct query:

{'query': 'SELECT COUNT(*) AS drive_count FROM drivestats'}

DeepSeek, on the other hand, responded with a Python stack trace and this error:

openai.UnprocessableEntityError: Failed to deserialize the JSON body into the target type: response_format: response_format.type `json_schema` is unavailable now at line 1 column 13827

What went wrong? Searching for the error turns up a comment from a LangChain engineer explaining that we should use BaseChatOpenAI rather than ChatOpenAI since it “[…] accommodates many APIs that are similar to OpenAI. It uses tool calling for structured output by default.”

So, we can redefine llm accordingly, and try generating a query again:

llm = BaseChatOpenAI(model="deepseek-reasoner", base_url='https://api.deepseek.com')

write_query(question)

Unfortunately, DeepSeek returns another error:

BadRequestError: Error code: 400 - {'error': {'message': 'The last message of deepseek-reasoner must be a user message, or an assistant message with prefix mode on (refer to https://api-docs.deepseek.com/guides/chat_prefix_completion).', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}

Looking back at the AI agent code, we can see that we used an off-the-shelf prompt from the LangChain Prompt Hub that provides the model with a single, system, message:

================================ System Message ================================

Given an input question, create a syntactically correct {dialect} query to run to help find the answer. Unless the user specifies in his question a specific number of examples they wish to obtain, always limit your query to at most {top_k} results. You can order the results by a relevant column to return the most interesting examples in the database.

Never query for all the columns from a specific table, only ask for a few relevant columns given the question.

Pay attention to use only the column names that you can see in the schema description. Be careful to not query for columns that do not exist. Also, pay attention to which column is in which table.

Only use the following tables:
{table_info}

Question: {input}

Does this mean that DeepSeek is not, in fact, API-compatible with OpenAI? I would argue that it does not. DeepSeek implements the same API request/response syntax as OpenAI, but it is a different platform. Some variation in semantics is to be expected. We see similar variations between Backblaze B2 and Amazon S3; for example, the S3 PutObjectAcl operation sets the access control list (ACL) for an object in a bucket. Amazon S3’s access management model allows you to manipulate an object’s ACL independently of its bucket—for example, you can put a private object in a public bucket, and vice versa.

This flexibility comes with a cost: It becomes difficult to reason about the visibility of data. In fact, AWS now recommends “that you keep ACLs disabled, except in unusual circumstances where you need to control access for each object individually.”

Backblaze B2’s model is much simpler: You control access at the bucket level, and all objects have the same ACL as their bucket. Backblaze B2 implements the PutObjectAcl operation, but, if you try to set an object’s ACL to any other value than its bucket’s ACL, the service responds with an error.

Returning to the AI agent code, we can replace the single-system-message prompt with one that combines a system message with a user message:

import textwrap
from langchain_core.prompts import ChatPromptTemplate

query_prompt_template = ChatPromptTemplate([
    ("system", textwrap.dedent("""Given an input question, create a
    syntactically correct {dialect} query to run to help find the answer.
    Unless the user specifies in his question a specific number of examples
    they wish to obtain, always limit your query to at most {top_k} results.
    You can order the results by a relevant column to return the most
    interesting examples in the database.

    Never query for all the columns from a specific table, only ask for a the
    few relevant columns given the question.

    Pay attention to use only the column names that you can see in the schema
    description. Be careful to not query for columns that do not exist. Also,
    pay attention to which column is in which table.

    Only use the following tables:
    {table_info}""")),
    ("human", "Question: {input}"),
])

Trying the write_query() call for a third time, this is the response:

BadRequestError: Error code: 400 - {'error': {'message': 'deepseek-reasoner does not support Function Calling', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}

A third error! What is this “function calling” that deepseek-reasoner does not support? A helpful article on the topic at the Hugging Face AI community explains:

Function calling is a powerful capability that enables Large Language Models (LLMs) to interact with your code and external systems in a structured way. Instead of just generating text responses, LLMs can understand when to call specific functions and provide the necessary parameters to execute real-world actions.

Unfortunately, that is exactly our use case. It’s becoming clear that DeepSeek R1 is not the correct tool for implementing an AI agent—we’ve been trying to use a chisel as a screwdriver!

DeepSeek-V3: A better fit

As its name suggests, the deepseek-chat model is more appropriate for this application. The DeepSeek documentation tells us that it is based on DeepSeek-V3, released in December 2024. DeepSeek-V3 is priced at $0.27 per million input tokens and $1.10 per million output tokens; this is actually more expensive than the GPT-4o mini model I used for the OpenAI agent example ($0.15 per million input tokens, $0.600 per million output tokens), but how does it compare? Let’s take a look.

First, we need to edit the LLM creation code again to set the model name:

llm = BaseChatOpenAI(model="deepseek-chat", base_url='https://api.deepseek.com')

Now we can run write_query() again. It’s immediately clear that it’s a better fit than its “big brother:”

{'query': 'SELECT COUNT(*) AS total_drives FROM drivestats LIMIT 10'}

As with the OpenAI agent, this query is well-formed SQL, but it’s not answering the question we set—it’s giving us the total number of rows in the dataset, rather than the number of drives. Also, it’s a little odd to have a LIMIT clause in a SELECT COUNT(*) query, but it’s legal SQL, and the agent is following its instructions very literally: always limit your query to at most {top_k} results, where we set top_k to 10.

question = {"question": "Each drive has its own serial number. How many drives are there?"}

query = write_query(question)

{'query': 'SELECT COUNT(DISTINCT serial_number) AS total_drives FROM drivestats'}

So far, so good!

I’ll skip some intermediate steps here—they are all in the Jupyter notebook if you want to review them, or run them for yourself—and look at how a simple LangChain graph, built on the DeepSeek LLM, answered the question: “Each drive has its own serial number. How many drives did each data center have on 9/1/2024?”

The OpenAI version generated an invalid query, comparing the date column with the string ’2024-09-01’ without using the required DATE type identifier, but DeepSeek generates a correct SQL query and provides a useful natural language response:

DeepSeek scores a point!

Moving on to the ReAct AI Agent, which allows the LLM to perform multiple SQL queries in generating an answer to a question, DeepSeek performs similarly to OpenAI. Given the question, “Each drive has its own serial number. What is the annualized failure rate of the ST4000DM000 drive model?”, the DeepSeek agent provides the overall failure rate rather than the annualized failure rate (AFR).

When we provide explicit instructions for calculating AFR in its prompt, the DeepSeek agent provides the correct result, identical, in fact, to the OpenAI agent’s response:

The annual failure rate (AFR) for the ST4000DM000 drive model is approximately 2.63%.

However, when given the question, “What was the annual failure rate of the ST8000NM000A drive model in Q3 2024?”, the DeepSeek agent gives us:

[(1.6100573445081607,)]

While OpenAI responds:

The annual failure rate (AFR) of the ST8000NM000A drive model in Q3 2024 is approximately 1.61%.

Wrapping up the investigation, the final question from the OpenAI notebook is more complex:

Considering only drive models which had at least 100 drives in service at the end of the quarter and which accumulated 10,000 or more drive days during the quarter, which drive had the most failures in Q3 2024, and what was its failure rate?

Impressively, the OpenAI agent constructed a well-formed SQL query and provided the correct response:

The drive model with the most failures in Q3 2024 is the TOSHIBA MG08ACA16TA, which had 181 failures. Its failure rate during this period was approximately 1.84%.

BadRequestError: Error code: 400 - {'error': {'message': "An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. (insufficient tool messages following tool_calls message)", 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
During task with name 'agent' and id '0aa26ba6-a3ee-ced1-de4d-b60ed7fbca99'

The phrase “insufficient tool messages” suggested that the DeepSeek LLM might need to be reconfigured to allow more tokens. According to the documentation on models and pricing, the deepseek-chat model supports a maximum of 8K output tokens, but defaults to 4K if max_tokens is not specified.

Recreating the DeepSeek wrapper object and agent accordingly, I gave it the last question again:

llm = BaseChatOpenAI(model="deepseek-chat", base_url='https://api.deepseek.com', max_tokens=8192, **extra_kwargs)

agent_executor = create_react_agent(llm, tools, state_modifier=system_message)

response = agent_executor.invoke(
    {"messages": [{"role": "user", "content": "Considering only drive models which had at least 100 drives in service at the end of the quarter and which accumulated 10,000 or more drive days during the quarter, which drive had the most failures in Q3 2024, and what was its failure rate?"}]}
)

# Show the SQL query sent to the database
print(response['messages'][-3].tool_calls[0]['args']['query'])

# Show the final response message
display_markdown(response['messages'][-1].content, raw=True)

This time, DeepSeek was able to generate a similar SQL query to OpenAI:

WITH drive_counts AS (
    SELECT model, COUNT(DISTINCT serial_number) AS drive_count
    FROM drivestats
    WHERE date >= DATE '2024-07-01' AND date <= DATE '2024-09-30'
    GROUP BY model
    HAVING COUNT(DISTINCT serial_number) >= 100
), drive_days AS (
    SELECT model, COUNT(*) AS total_drive_days
    FROM drivestats
    WHERE date >= DATE '2024-07-01' AND date <= DATE '2024-09-30'
    GROUP BY model
    HAVING COUNT(*) >= 10000
), failures AS (
    SELECT model, COUNT(*) AS failure_count
    FROM drivestats
    WHERE date >= DATE '2024-07-01' AND date <= DATE '2024-09-30' AND failure = 1
    GROUP BY model
)
SELECT d.model,
       f.failure_count,
       100 * (CAST(f.failure_count AS DOUBLE) / (CAST(d.total_drive_days AS DOUBLE) / 365)) AS annual_failure_rate
FROM drive_days d
JOIN failures f ON d.model = f.model
JOIN drive_counts dc ON d.model = dc.model
ORDER BY f.failure_count DESC
LIMIT 1

With a correct response:

To answer the question:

The drive model with the most failures in Q3 2024 is TOSHIBA MG08ACA16TA, which had 181 failures. The annualized failure rate (AFR) for this model during that quarter was 1.84%.

Success! But, unfortunately, this isn’t the whole story.

DeepSeek Reliability

I originally set out to write this blog post at the end of January, but the DeepSeek platform website had gone offline by January 30, so I couldn’t even start until I was able to sign up for an API key on February 5.

Given my shiny new API key, and DeepSeek’s claims of OpenAI API compatibility, I naïvely expected to be able to work through my earlier OpenAI notebook and write up the results in a couple of days. The reality was more like two weeks.

In this blog post I’ve detailed some of the error messages I encountered along the way, but I saw many more that pointed to the DeepSeek API simply being overwhelmed with traffic. For example, for over a day, when the status page reported no issues, most API requests to DeepSeek terminated after a minute with the error message:

json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)

A time-consuming investigation revealed that this was caused by the DeepSeek API returning the 200 status code and headers as if the request was successful, then hanging for a minute before terminating the connection without returning any actual data. The calling code saw the 200 as success and tried to decode the non-existent API response body, resulting in the error.

I saw several more instances of intermittent errors that all seemed to point in the same direction: DeepSeek needs to add capacity to its API platform. Notably, the platform seemed faster and more stable on a Saturday morning, U.S. Pacific time, the early hours of Sunday morning in China.

Final thoughts

At present, I would have to classify the DeepSeek-V3 API as “promising, but somewhat flaky.” An agent invocation that succeeds one minute could fail the next with any of a range of error messages. That’s a shame, since when it does work, for instance, in creating the SQL query for the final question above, it tends to work very well.

One final caveat: This is a dynamic field; frameworks and services are literally being updated on a daily basis. For example, since yesterday, as I write this, four of the notebook’s module dependencies have been updated. I encourage you to experiment for yourself as your mileage will almost certainly vary, hopefully in a positive direction.

The post Experimenting with DeepSeek, Backblaze B2, and Drive Stats appeared first on Backblaze Blog | Cloud Storage & Cloud Backup