Class: ActiveHarness::Providers::GPUStack

Inherits:
Base
  • Object
show all
Defined in:
lib/active_harness/providers/gpustack.rb

Overview

GPUStack — self-hosted GPU inference server, OpenAI-compatible API. docs.gpustack.ai/latest/user-guide/inference-openai-compatible-apis/

GPUSTACK_API_BASE is required (e.g. “my-gpustack-server:80”). GPUSTACK_API_KEY is optional (needed only if the server has auth enabled).

Example:

model do
  use provider: :gpustack, model: "Qwen/Qwen2.5-7B-Instruct-GGUF"
end

Constant Summary

Constants inherited from Base

Base::HTTP, Base::STREAMING_HTTP

Instance Method Summary collapse

Instance Method Details

#call(model:, messages:, temperature: 0.7) ⇒ Object 



16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
 
# File 'lib/active_harness/providers/gpustack.rb', line 16

def call(model:, messages:, temperature: 0.7)
  url = URI("#{api_base}/v1/chat/completions")

  headers = { "Content-Type" => "application/json" }
  key     = api_key
  headers["Authorization"] = "Bearer #{key}" if key

  raw  = post_json(url,
    headers: headers,
    body:    { model: model, messages: messages, temperature: temperature }
  )
  data = parse!(raw)
  handle_error!(data)

  {
    content:  data.dig("choices", 0, "message", "content").to_s.strip,
    provider: :gpustack,
    model:    data["model"] || model,
    usage:    extract_usage_openai(data)
  }
end