Class: Yorishiro::Provider::Ollama

Inherits:
Base
  • Object
show all
Defined in:
lib/yorishiro/provider/ollama.rb

Constant Summary collapse

DEFAULT_BASE_URL = 
"http://localhost:11434"
DEFAULT_NUM_CTX = 
8192
OUTPUT_TOKEN_RESERVE = 
2048
MIN_CONTEXT_BUDGET = 
1024

Instance Attribute Summary

Attributes inherited from Base

#api_key, #model_name

Class Method Summary collapse

Instance Method Summary collapse

Methods inherited from Base

#debug?, #debug_log

Constructor Details

#initialize(api_key: nil, model: nil, base_url: nil, num_ctx: nil) ⇒ Ollama

Returns a new instance of Ollama.



11
12
13
14
15
 
# File 'lib/yorishiro/provider/ollama.rb', line 11

def initialize(api_key: nil, model: nil, base_url: nil, num_ctx: nil)
  @base_url = base_url || ENV.fetch("OLLAMA_HOST", DEFAULT_BASE_URL)
  @num_ctx = num_ctx
  super(api_key: api_key || "unused", model: model)
end

Class Method Details

.supported_models ⇒ Object 



17
18
19
20
21
22
23
24
25
26
27
 
# File 'lib/yorishiro/provider/ollama.rb', line 17

def self.supported_models
  base_url = ENV.fetch("OLLAMA_HOST", DEFAULT_BASE_URL)
  uri = URI("#{base_url}/api/tags")
  response = Net::HTTP.get_response(uri)
  return [] unless response.is_a?(Net::HTTPSuccess)

  data = JSON.parse(response.body)
  data.fetch("models", []).map { |m| m["name"] }
rescue StandardError
  []
end

Instance Method Details

#chat(conversation, tools: []) ⇒ Object 



29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
 
# File 'lib/yorishiro/provider/ollama.rb', line 29

def chat(conversation, tools: [], &)
  uri = URI("#{@base_url}/api/chat")

  body = {
    model: @model_name,
    messages: format_messages(conversation.to_api_messages),
    keep_alive: ENV.fetch("OLLAMA_KEEP_ALIVE", "10m"),
    options: { num_ctx: num_ctx },
    stream: true
  }

  headers = { "Content-Type" => "application/json" }

  body[:tools] = format_tools(tools) unless tools.empty?

  debug_log("Ollama request", body)

  @last_meta = {}
  result = post_stream(uri, headers: headers, body: body, &)
  result[:meta] = @last_meta
  result[:usage] = { input: @last_meta[:prompt_eval_count], output: @last_meta[:eval_count] }
  debug_log("Ollama response", result)
  result
end

#context_budget_tokens ⇒ Object

Token budget the conversation should be trimmed to before each request. Reserves headroom for the model's output within the configured context window.



56
57
58
 
# File 'lib/yorishiro/provider/ollama.rb', line 56

def context_budget_tokens
  [num_ctx - OUTPUT_TOKEN_RESERVE, MIN_CONTEXT_BUDGET].max
end