mirror of
https://gh.wpcy.net/https://github.com/discourse/discourse.git
synced 2026-06-19 04:25:50 +08:00
1dbe92f222
- Update Anthropic presets from Claude 4.5 to Claude 4.6 (Opus and Sonnet), increasing Opus max output tokens to 128k - Add AWS Bedrock model ID mappings for Claude 4.5 and 4.6 - Remove anthropic-beta headers for prompt caching and effort features, which are now GA and no longer require beta flags - Update OpenRouter model presets: MiniMax M2.1 → M2.5, Z-AI GLM-4.7 → GLM-5, refresh Kimi K2.5 token limits and pricing - Broaden effort parameter description (no longer Opus-only) - Update eval LLM config, locales, and specs accordingly Most important thing here is the beta header removal, having it causes features to break since they are no longer supported
55 lines
1.6 KiB
Ruby
Vendored
55 lines
1.6 KiB
Ruby
Vendored
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Completions
|
|
module Endpoints
|
|
module AnthropicPromptCache
|
|
def should_apply_prompt_caching?(prompt)
|
|
caching_mode = llm_model.lookup_custom_param("prompt_caching") || "never"
|
|
return false if caching_mode == "never"
|
|
|
|
case caching_mode
|
|
when "always"
|
|
true
|
|
when "tool_results"
|
|
prompt
|
|
.messages
|
|
.last(5)
|
|
.any? do |msg|
|
|
content = msg[:content]
|
|
|
|
if content.is_a?(Array)
|
|
content.any? { |c| c.is_a?(Hash) && c[:type] == "tool_result" }
|
|
elsif content.is_a?(Hash)
|
|
content[:type] == "tool_result"
|
|
else
|
|
false
|
|
end
|
|
end
|
|
else
|
|
false
|
|
end
|
|
end
|
|
|
|
def apply_anthropic_cache_control!(payload, prompt)
|
|
if payload[:messages].present?
|
|
last_message = payload[:messages].last
|
|
|
|
if last_message[:content].is_a?(String)
|
|
last_message[:content] = [
|
|
type: "text",
|
|
text: last_message[:content],
|
|
cache_control: {
|
|
type: "ephemeral",
|
|
},
|
|
]
|
|
elsif last_message[:content].is_a?(Array)
|
|
last_content = last_message[:content].last
|
|
last_content[:cache_control] = { type: "ephemeral" } if last_content.is_a?(Hash)
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|
