aireview 2.2.0 → 2.2.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c37a072f5cb03c8a4e95944a3c6fc5386d1abe6c340c63ebee9715db5d44ec6a
-  data.tar.gz: b2e2ef104f8e4310dafc0dd5041a899c7ff06af5d2098cff25e0e38f010f9847
+  metadata.gz: 7919de6ffa17aab9f5250ebf07dfec9ce83db9313ef01e6ba93a3b131da927ce
+  data.tar.gz: 348cda52953ff2d22ba56fb16b192876249efa30c5ba77f6f0736c1b96e5f569
 SHA512:
-  metadata.gz: '09524e49ed6896e32e6e73d3fcc5f5e27d19eaa255273017acf699e8e6992f44a8a950e34d96f446066e58b31a6ee625188891e913ac3a4c0db27ac2b39e95a8'
-  data.tar.gz: 4a193a705c06609c07ceb2833211d51d124e1299378368feb50fc08e0fac48a6ff09f553b47406b0bd367e83ebfe7b55fd66d018f4781551725aa27234e95113
+  metadata.gz: 372867e5a1547b4d595312498107dd8c33c6fe6721efb8f4c215850660bcf57706f5a864c9d9b7d46a9d2a1e29fdf18ab31e9ddb77c1859237ddbe5d979b2946
+  data.tar.gz: b3b079dbb2b5b92ffe39fc36c6de224d9bf2268e11b353129236e15b5c14689a9067012da053b8415bb53f64b2f52247ef252062831e9e73d789b943a611ed0a

data/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 2.2.1
+
+- OpenAI reasoning models (`o1`, `o3`, `gpt-5`…) and search models
+  (`gpt-4o-search-preview`…) get no temperature: RubyLLM 2 sends it as given
+  (1.x replaced it with 1.0 or dropped it), and these models reject any
+  other value with a bad request, which failed the run without trying a
+  fallback model. The RubyLLM registry says which models take a
+  temperature; a model it does not know is judged by its name, as 1.x did.
+  `models check` is fixed the same way.
+
 ## 2.2.0
 
 - Jev shadow mode (`llm.jev.shadow`, `LLM_JEV_SHADOW`, key `JEV_API_KEY`):

data/README.md

@@ -226,6 +226,12 @@ or without auth when that is not set, and needs no key to start. Without an
 `api_base` a model goes to its provider's API, or to `LLM_API_BASE` for
 Gemini, OpenAI and OpenRouter as before.
 
+A stage's temperature goes only to the models that take one: OpenAI
+reasoning models (`o1`, `o3`, `gpt-5`…) and search models
+(`gpt-4o-search-preview`…) accept no other, so they get none and run with
+their own default. The RubyLLM registry decides, and the name
+for a model it does not know.
+
 Instead of an LLM the critic can be Jev (see "Critique engine").
 
 ### Local Ollama

data/lib/aireview/llm_client.rb

@@ -17,6 +17,16 @@ module Aireview
       end
     end
 
+    # RubyLLM 2 sends the temperature as given, while 1.x replaced it with
+    # 1.0 for OpenAI reasoning models (o1, o3, gpt-5…) and dropped it for the
+    # search ones (gpt-4o-search-preview…), which take no other: such a
+    # request is a BadRequest, fatal for the router, and the stage would not
+    # even reach a fallback model. The RubyLLM registry knows which models
+    # take a temperature; where it does not (a server of your own,
+    # assume_model_exists, search models with an empty flag) the name
+    # decides, as in 1.x.
+    NO_TEMPERATURE_MODELS = %r{\A(?:openai/)?(?:o\d|gpt-5)|-search}
+
     def initialize(config:, logger: Logger.new($stderr))
       @config = config
       @logger = logger
@@ -26,16 +36,10 @@ module Aireview
     # Returns the RubyLLM answer; read it with LlmClient.content. A request
     # error is re-raised as is — LlmFailure classifies it.
     def request(prompt, candidate:, key:, timeout:, key_index: 0)
-      load_ruby_llm
       stage = prompt.stage.to_s
       model = candidate.model
-      @logger.info("LLM #{stage} request started (model=#{model}, temperature=#{prompt.temperature})")
-      chat = build_chat(context: context(stage, candidate, key, key_index), stage: stage,
-                        model: model, provider: candidate.provider)
-      chat = configure_reasoning(chat: chat, model: model, provider: candidate.provider)
-        .with_temperature(prompt.temperature.to_f)
-        .with_schema(prompt.schema)
-      chat.with_instructions(prompt.system)
+      chat = prepare(prompt, candidate: candidate, key: key, key_index: key_index)
+      @logger.info("LLM #{stage} request started (model=#{model}, temperature=#{chat.temperature || 'model default'})")
       response = Timeout.timeout(timeout) { chat.ask(prompt.user) }
       @logger.info("LLM #{stage} request completed (model=#{model}#{token_counts(response)})")
       response
@@ -44,6 +48,21 @@ module Aireview
       raise
     end
 
+    # The request as it will go, without sending it: the chat with the model,
+    # key, schema, instructions and temperature set. chat.render builds it —
+    # that is how the specs check the real request without the network.
+    def prepare(prompt, candidate:, key:, key_index: 0)
+      load_ruby_llm
+      stage = prompt.stage.to_s
+      chat = build_chat(context: context(stage, candidate, key, key_index), stage: stage,
+                        model: candidate.model, provider: candidate.provider)
+      chat = configure_reasoning(chat: chat, model: candidate.model, provider: candidate.provider)
+        .with_temperature(temperature_for(chat, prompt, candidate))
+        .with_schema(prompt.schema)
+      chat.with_instructions(prompt.system)
+      chat
+    end
+
     # The answer as RubyLLM 1.x gave it under a schema: the parsed JSON when
     # the text is JSON, the text itself otherwise (the pipeline repairs it).
     # RubyLLM 2 always returns the text, and a Hash that breaks the schema
@@ -82,6 +101,14 @@ module Aireview
       raise ConfigError, "Missing dependency: #{e.message}"
     end
 
+    # The temperature for a model that takes one (see NO_TEMPERATURE_MODELS);
+    # nil — its own default, left out of the request.
+    def temperature_for(chat, prompt, candidate)
+      accepts = chat.model.metadata[:temperature] if chat.model.respond_to?(:metadata)
+      accepts = !candidate.model.to_s.match?(NO_TEMPERATURE_MODELS) if accepts.nil?
+      accepts ? prompt.temperature.to_f : nil
+    end
+
     def configure_reasoning(chat:, model:, provider:)
       return chat unless provider == 'ollama' && model.start_with?('gpt-oss:')
 

data/lib/aireview/version.rb

@@ -1,4 +1,4 @@
 # frozen_string_literal: true
 module Aireview
-  VERSION = '2.2.0'
+  VERSION = '2.2.1'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: aireview
 version: !ruby/object:Gem::Version
-  version: 2.2.0
+  version: 2.2.1
 platform: ruby
 authors:
 - Denis Levenko