discourse-ai/spec/lib/completions/xml_tool_processor_spec.rb
Sam e817b7dc11
FEATURE: improve tool support (#904)
This re-implements tool support in DiscourseAi::Completions::Llm #generate

Previously tool support was always returned via XML and it would be the responsibility of the caller to parse XML

New implementation has the endpoints return ToolCall objects.

Additionally this simplifies the Llm endpoint interface and gives it more clarity. Llms must implement

decode, decode_chunk (for streaming)

It is the implementers responsibility to figure out how to decode chunks, base no longer implements. To make this easy we ship a flexible json decoder which is easy to wire up.

Also (new)

    Better debugging for PMs, we now have a next / previous button to see all the Llm messages associated with a PM
    Token accounting is fixed for vllm (we were not correctly counting tokens)
2024-11-12 08:14:30 +11:00

189 lines
4.5 KiB
Ruby

# frozen_string_literal: true
RSpec.describe DiscourseAi::Completions::XmlToolProcessor do
let(:processor) { DiscourseAi::Completions::XmlToolProcessor.new }
it "can process simple text" do
result = []
result << (processor << "hello")
result << (processor << " world ")
expect(result).to eq([["hello"], [" world "]])
expect(processor.finish).to eq([])
expect(processor.should_cancel?).to eq(false)
end
it "is usable for simple single message mode" do
xml = (<<~XML).strip
hello
<function_calls>
<invoke>
<tool_name>hello</tool_name>
<parameters>
<hello>world</hello>
<test>value</test>
</parameters>
</invoke>
XML
result = []
result << (processor << xml)
result << (processor.finish)
tool_call =
DiscourseAi::Completions::ToolCall.new(
id: "tool_0",
name: "hello",
parameters: {
hello: "world",
test: "value",
},
)
expect(result).to eq([["hello"], [tool_call]])
expect(processor.should_cancel?).to eq(false)
end
it "handles multiple tool calls in sequence" do
xml = (<<~XML).strip
start
<function_calls>
<invoke>
<tool_name>first_tool</tool_name>
<parameters>
<param1>value1</param1>
</parameters>
</invoke>
<invoke>
<tool_name>second_tool</tool_name>
<parameters>
<param2>value2</param2>
</parameters>
</invoke>
</function_calls>
end
XML
result = []
result << (processor << xml)
result << (processor.finish)
first_tool =
DiscourseAi::Completions::ToolCall.new(
id: "tool_0",
name: "first_tool",
parameters: {
param1: "value1",
},
)
second_tool =
DiscourseAi::Completions::ToolCall.new(
id: "tool_1",
name: "second_tool",
parameters: {
param2: "value2",
},
)
expect(result).to eq([["start"], [first_tool, second_tool]])
expect(processor.should_cancel?).to eq(true)
end
it "handles non-English parameters correctly" do
xml = (<<~XML).strip
<function_calls>
<invoke>
<tool_name>translator</tool_name>
<parameters>
<text>世界</text>
</parameters>
</invoke>
XML
result = []
result << (processor << xml)
result << (processor.finish)
tool_call =
DiscourseAi::Completions::ToolCall.new(
id: "tool_0",
name: "translator",
parameters: {
text: "世界",
},
)
expect(result).to eq([["こんにちは"], [tool_call]])
end
it "processes input character by character" do
xml =
"hi<function_calls><invoke><tool_name>test</tool_name><parameters><p>v</p></parameters></invoke>"
result = []
xml.each_char { |char| result << (processor << char) }
result << processor.finish
tool_call =
DiscourseAi::Completions::ToolCall.new(id: "tool_0", name: "test", parameters: { p: "v" })
filtered_result = result.reject(&:empty?)
expect(filtered_result).to eq([["h"], ["i"], [tool_call]])
end
it "handles malformed XML gracefully" do
xml = (<<~XML).strip
text
<function_calls>
<invoke>
<tool_name>test</tool_name>
<parameters>
<param>value
</parameters>
</invoke>
malformed
XML
result = []
result << (processor << xml)
result << (processor.finish)
# Should just do its best to parse the XML
tool_call =
DiscourseAi::Completions::ToolCall.new(id: "tool_0", name: "test", parameters: { param: "" })
expect(result).to eq([["text"], [tool_call]])
end
it "correctly processes empty parameter sets" do
xml = (<<~XML).strip
hello
<function_calls>
<invoke>
<tool_name>no_params</tool_name>
<parameters>
</parameters>
</invoke>
XML
result = []
result << (processor << xml)
result << (processor.finish)
tool_call =
DiscourseAi::Completions::ToolCall.new(id: "tool_0", name: "no_params", parameters: {})
expect(result).to eq([["hello"], [tool_call]])
end
it "properly handles cancelled processing" do
xml = "start<function_calls></function_calls>"
result = []
result << (processor << xml)
result << (processor << "more text")
result << processor.finish
expect(result).to eq([["start"], [], []])
expect(processor.should_cancel?).to eq(true)
end
end