streaming

Streaming

The Streaming API allows you to receive responses in real-time chunks rather than waiting for the complete response.

Basic Usage

client = ChatGPT::Client.new
client.chat_stream([
  { role: "user", content: "Tell me a story" }
]) do |chunk|
  print chunk.dig("choices", 0, "delta", "content")
end

Parameters

client.chat_stream(
  messages,
  {
    model: 'gpt-3.5-turbo',
    temperature: 0.7,
    max_tokens: 100
  }
) do |chunk|
  # Process chunk
end

Chunk Format

{
  "choices" => [{
    "delta" => {
      "content" => "chunk of text"
    },
    "index" => 0,
    "finish_reason" => nil
  }]
}

Error Handling

begin
  client.chat_stream(messages) do |chunk|
    print chunk.dig("choices", 0, "delta", "content")
  end
rescue ChatGPT::APIError => e
  puts "Error: #{e.message}"
end

Best Practices

1. Buffer Management

buffer = StringIO.new
client.chat_stream(messages) do |chunk|
  content = chunk.dig("choices", 0, "delta", "content")
  buffer << content if content
end
complete_response = buffer.string

2. Progress Indicators

dots = 0
client.chat_stream(messages) do |chunk|
  print chunk.dig("choices", 0, "delta", "content")
  dots = (dots + 1) % 4
  print "\rProcessing#{'.' * dots}#{' ' * (3 - dots)}"
end

3. Error Recovery

def stream_with_retry(messages, max_attempts = 3)
  attempts = 0
  begin
    attempts += 1
    client.chat_stream(messages) do |chunk|
      yield chunk if block_given?
    end
  rescue ChatGPT::RateLimitError => e
    retry if attempts < max_attempts
    raise
  end
end

Examples

Basic Streaming

client.chat_stream([
  { role: "user", content: "Tell me a story" }
]) do |chunk|
  print chunk.dig("choices", 0, "delta", "content")
end

With Response Building

full_response = ""
client.chat_stream(messages) do |chunk|
  content = chunk.dig("choices", 0, "delta", "content")
  full_response += content if content
  print content if content
end

Real-time Processing

client.chat_stream(messages) do |chunk|
  content = chunk.dig("choices", 0, "delta", "content")
  if content
    processed = process_content(content)  # Your processing logic
    print processed
  end
end