lex-llm-ollama 0.2.7 → 0.2.9

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: bd3556fdfe59e46af218aa97edcc1f81b3ffb39c8af1c144f4b3f0e23bf17bea
-  data.tar.gz: cff453247d66070976c3cb74e1f5524e718fbded408d340dc7cb74892750b39c
+  metadata.gz: 90c9bd5b9b740637320d528012401d93d38e1d0b38396c3baa51ad45067547aa
+  data.tar.gz: 1024e115de9bea52c58b3fb20bbbf1fc561bc6299f4bdfea4a95d2c7899618c8
 SHA512:
-  metadata.gz: a3cac452d99bfb4e0c34285b38979a59e0937a98b3dac849fa401fde6022d9e04887062b062185199294c6d4dafe1ea5f65c9fdbb9be4b5e3cfb0fcbfd9a1159
-  data.tar.gz: ecbd0541f0ca31ee2979f1424034fd719533d0f86b6631be510ff696b775ee23b4a314bf86596e3062c223a90d3b65d4bc5a1b30c5f69833cf7596cea00295fe
+  metadata.gz: 7222b463dd74c7ba154d363dabaf451979562831313583ac2e6fc034ddda595c13b3c4b1ae6772906842a92aa83842bb9a37c2c99f198c5a4985c82a3f61174c
+  data.tar.gz: 682b890f4c30d28d0786c6d3f7d03ac1d024fbec025119373346f9bcb87cad22391328b02238007ab9550cf6dd8ef07f565cc69628e26084fc2815a4b2d8e095

data/CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## 0.2.9 - 2026-05-13
+
+- Add `fetch_model_detail` — calls POST `/api/show` to retrieve the real context window from Ollama.
+- Add `resolve_context_window` — tries live model detail cache first, falls back to static prefix map.
+- Add `extract_context_window` — parses `num_ctx` from `model_info` hash or `parameters` string in the `/api/show` response.
+- Add `CONTEXT_WINDOWS` static fallback map covering common Ollama model families.
+- Add `rescue Faraday::ConnectionFailed` in `discover_offerings` with a concise warn log instead of an unhandled exception.
+- Add `show_model_url` endpoint helper returning `/api/show`.
+
+## 0.2.8 - 2026-05-12
+
+- Include `Legion::Logging::Helper` directly in Ollama provider, actor, and fleet runner runtime surfaces.
+- Add sanitized debug logging for provider discovery, payload rendering, tool formatting, embeddings, offerings, and fleet handoff.
+
 ## 0.2.7 - 2026-05-07
 
 - Render Ollama embedding payloads with the canonical model id when callers pass `Model::Info` objects.

data/lib/legion/extensions/llm/ollama/actors/fleet_worker.rb

@@ -12,6 +12,7 @@ end
 
 require 'legion/extensions/llm/ollama'
 require 'legion/extensions/llm/fleet/provider_responder'
+require 'legion/logging/helper'
 
 module Legion
   module Extensions
@@ -20,6 +21,8 @@ module Legion
         module Actor
           # Subscription actor for Ollama fleet request consumption.
           class FleetWorker < Legion::Extensions::Actors::Subscription
+            include Legion::Logging::Helper
+
             def runner_class
               'Legion::Extensions::Llm::Ollama::Runners::FleetWorker'
             end
@@ -33,7 +36,9 @@ module Legion
             end
 
             def enabled?
-              Legion::Extensions::Llm::Fleet::ProviderResponder.enabled_for?(Ollama.discover_instances)
+              enabled = Legion::Extensions::Llm::Fleet::ProviderResponder.enabled_for?(Ollama.discover_instances)
+              log.debug { "ollama fleet worker actor enabled=#{enabled}" }
+              enabled
             end
           end
         end

data/lib/legion/extensions/llm/ollama/provider.rb

@@ -1,6 +1,7 @@
 # frozen_string_literal: true
 
 require 'legion/extensions/llm'
+require 'legion/logging/helper'
 
 module Legion
   module Extensions
@@ -8,6 +9,8 @@ module Legion
       module Ollama
         # Ollama provider implementation for the Legion::Extensions::Llm base provider contract.
         class Provider < Legion::Extensions::Llm::Provider # rubocop:disable Metrics/ClassLength
+          include Legion::Logging::Helper
+
           class << self
             attr_writer :registry_publisher
 
@@ -54,7 +57,7 @@ module Legion
           def version_url = '/api/version'
 
           def list_running_models
-            log.info { "listing running models from #{api_base}#{running_models_url}" }
+            log.debug { "ollama provider listing running models endpoint=#{api_base}#{running_models_url}" }
             connection.get(running_models_url).body.fetch('models', [])
           rescue StandardError => e
             handle_exception(e, level: :error, handled: true, operation: 'ollama.list_running_models')
@@ -62,29 +65,40 @@ module Legion
           end
 
           def readiness(live: false)
-            log.info { "checking readiness live=#{live} at #{api_base}" }
+            log.debug { "ollama provider checking readiness live=#{live} endpoint=#{api_base}" }
             super.tap do |metadata|
               self.class.registry_publisher.publish_readiness_async(metadata) if live
             end
           end
 
           def list_models
-            log.info { "discovering models from #{api_base}#{models_url}" }
+            log.debug { "ollama provider discovering models endpoint=#{api_base}#{models_url}" }
             super.tap do |models|
-              log.info { "discovered #{models.size} model(s) from Ollama" }
+              log.debug { "ollama provider discovered model_count=#{models.size}" }
               self.class.registry_publisher.publish_models_async(models, readiness: readiness(live: false))
             end
           end
 
           def show_model(model)
-            log.info { "fetching model details for #{model}" }
+            log.debug { "ollama provider fetching model details model=#{model}" }
             connection.post(show_model_url, { model: model }).body
           rescue StandardError => e
             handle_exception(e, level: :error, handled: true, operation: 'ollama.show_model')
             raise
           end
 
+          def fetch_model_detail(model_name)
+            raw = show_model(model_name)
+            context_window = extract_context_window(raw)
+            { context_window: context_window }.compact
+          rescue StandardError => e
+            handle_exception(e, level: :warn, handled: true, operation: 'ollama.fetch_model_detail',
+                                model: model_name)
+            nil
+          end
+
           def pull_model(model, stream: false)
+            log.debug { "ollama provider pulling model=#{model} stream=#{stream}" }
             log.info { "pulling model #{model} stream=#{stream}" }
             connection.post(pull_url, { model: model, stream: stream }).body
           rescue StandardError => e
@@ -93,19 +107,52 @@ module Legion
           end
 
           def discover_offerings(live: false, **)
-            models = if live
-                       @cached_models = list_models
-                     else
-                       Array(@cached_models)
-                     end
-            models.map { |model_info| offering_from_model(model_info) }
+            log.debug do
+              "ollama provider discovering offerings live=#{live} cached_model_count=#{Array(@cached_models).size}"
+            end
+            resolve_models(live).map { |model_info| offering_from_model(model_info) }.tap do |offerings|
+              log.debug { "ollama provider built offering_count=#{offerings.size} live=#{live}" }
+            end
+          rescue Faraday::ConnectionFailed => e
+            log.warn("[ollama] instance=#{provider_instance_id} unreachable: #{e.message}")
+            []
           rescue StandardError => e
-            handle_exception(e, level: :warn, handled: true, operation: 'ollama.discover_offerings')
+            handle_exception(e, level: :warn, handled: true, operation: 'ollama.discover_offerings',
+                                backtrace_limit: 3)
             []
           end
 
+          CONTEXT_WINDOWS = {
+            'qwen3' => 128_000,
+            'qwen2.5' => 128_000,
+            'llama3' => 128_000,
+            'llama3.1' => 128_000,
+            'llama3.2' => 128_000,
+            'llama3.3' => 128_000,
+            'gemma2' => 8_192,
+            'gemma3' => 128_000,
+            'mistral' => 128_000,
+            'deepseek' => 128_000,
+            'phi3' => 128_000,
+            'phi4' => 16_384,
+            'command-r' => 128_000,
+            'codellama' => 16_384,
+            'nomic-embed' => 8_192,
+            'mxbai-embed' => 512,
+            'snowflake' => 512,
+            'bge' => 512
+          }.freeze
+
           private
 
+          def resolve_models(live)
+            if live
+              @cached_models = list_models
+            else
+              Array(@cached_models)
+            end
+          end
+
           def offering_from_model(model_info)
             Legion::Extensions::Llm::Routing::ModelOffering.new(
               provider_family: :ollama,
@@ -137,7 +184,46 @@ module Legion
           end
 
           def offering_limits(model_info)
-            { context_window: model_info.context_length }.compact
+            ctx = model_info.context_length || resolve_context_window(model_info.id)
+            ctx ? { context_window: ctx } : {}
+          end
+
+          def resolve_context_window(model_id)
+            detail = model_detail(model_id)
+            return detail[:context_window] if detail.is_a?(Hash) && detail[:context_window]
+
+            infer_context_window(model_id)
+          end
+
+          def infer_context_window(model_id)
+            name = model_id.to_s.split(':').first
+            CONTEXT_WINDOWS.find { |prefix, _| name.start_with?(prefix) }&.last
+          end
+
+          def extract_context_window(raw)
+            return nil unless raw.is_a?(Hash)
+
+            from_model_info(raw) || from_parameters_string(raw)
+          end
+
+          def from_model_info(raw)
+            model_info = raw['model_info'] || raw[:model_info]
+            return unless model_info.is_a?(Hash)
+
+            num_ctx_from_hash(model_info)&.to_i
+          end
+
+          def num_ctx_from_hash(model_info)
+            model_info['num_ctx'] || model_info[:num_ctx] ||
+              model_info.find { |k, _| k.to_s.end_with?('.context_length') }&.last
+          end
+
+          def from_parameters_string(raw)
+            params = raw['parameters'] || raw[:parameters]
+            return unless params.is_a?(String)
+
+            match = params.match(/num_ctx\s+(\d+)/)
+            match[1].to_i if match
           end
 
           def offering_metadata(model_info)
@@ -153,8 +239,14 @@ module Legion
           end
 
           def render_payload(messages, tools:, temperature:, model:, stream:, schema:, thinking:, tool_prefs:) # rubocop:disable Metrics/ParameterLists
+            model_id = model.respond_to?(:id) ? model.id : model
+            log.debug do
+              "ollama provider rendering chat payload model=#{model_id} message_count=#{messages.size} " \
+                "stream=#{stream} tools=#{tools.size} schema=#{!schema.nil?} thinking=#{thinking ? true : false}"
+            end
+
             {
-              model: model.id,
+              model: model_id,
               messages: format_messages(messages),
               stream: stream,
               think: thinking ? true : nil,
@@ -196,6 +288,9 @@ module Legion
           def format_tools(tools)
             return nil if tools.empty?
 
+            tool_names = tools.values.filter_map { |tool| tool.respond_to?(:name) ? tool.name : nil }
+            log.debug { "ollama provider formatting tools count=#{tools.size} names=#{tool_names.join(',')}" }
+
             tools.values.map do |tool|
               {
                 type: 'function',
@@ -243,6 +338,8 @@ module Legion
           def parse_tool_calls(tool_calls)
             return nil unless tool_calls
 
+            log.debug { "ollama provider parsing tool_call_count=#{tool_calls.size}" }
+
             tool_calls.to_h do |call|
               function = call.fetch('function', {})
               [
@@ -289,7 +386,11 @@ module Legion
           end
 
           def render_embedding_payload(text, model:, dimensions:)
-            { model: model.respond_to?(:id) ? model.id : model, input: text, dimensions: dimensions }.compact
+            model_id = model.respond_to?(:id) ? model.id : model
+            input_count = text.respond_to?(:size) ? text.size : 1
+            log.debug { "ollama provider rendering embedding payload model=#{model_id} input_count=#{input_count}" }
+
+            { model: model_id, input: text, dimensions: dimensions }.compact
           end
 
           def parse_embedding_response(response, model:, text:)
@@ -302,6 +403,9 @@ module Legion
                         body['embeddings']&.first
                       end
 
+            vector_count = vectors.respond_to?(:size) ? vectors.size : 0
+            log.debug { "ollama provider parsed embedding response model=#{model} vector_count=#{vector_count}" }
+
             Legion::Extensions::Llm::Embedding.new(vectors: vectors, model: model,
                                                    input_tokens: body['prompt_eval_count'].to_i)
           end

data/lib/legion/extensions/llm/ollama/runners/fleet_worker.rb

@@ -2,6 +2,7 @@
 
 require 'legion/extensions/llm/fleet/provider_responder'
 require 'legion/extensions/llm/ollama'
+require 'legion/logging/helper'
 
 module Legion
   module Extensions
@@ -10,9 +11,17 @@ module Legion
         module Runners
           # Runner entrypoint for Ollama fleet request execution.
           module FleetWorker
+            extend Legion::Logging::Helper
+
             module_function
 
             def handle_fleet_request(payload, delivery: nil, properties: nil)
+              payload_keys = payload.respond_to?(:keys) ? payload.keys.join(',') : payload.class
+              log.debug do
+                "ollama fleet worker handing off request payload_keys=#{payload_keys} " \
+                  "delivery=#{!delivery.nil?} properties=#{!properties.nil?}"
+              end
+
               Legion::Extensions::Llm::Fleet::ProviderResponder.call(
                 payload: payload,
                 provider_family: Ollama::PROVIDER_FAMILY,

data/lib/legion/extensions/llm/ollama/version.rb

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module Llm
       module Ollama
-        VERSION = '0.2.7'
+        VERSION = '0.2.9'
       end
     end
   end

data/lib/legion/extensions/llm/ollama.rb

@@ -3,6 +3,7 @@
 require 'legion/extensions/llm'
 require 'legion/extensions/llm/ollama/provider'
 require 'legion/extensions/llm/ollama/version'
+require 'legion/logging/helper'
 
 module Legion
   module Extensions
@@ -49,24 +50,10 @@ module Legion
         def self.discover_instances
           instances = {}
 
-          if CredentialSources.socket_open?('127.0.0.1', 11_434, timeout: 0.1)
-            instances[:local] = {
-              base_url: 'http://127.0.0.1:11434',
-              tier: :local,
-              capabilities: %i[completion embedding vision]
-            }
-          end
-
-          configured = CredentialSources.setting(:extensions, :llm, :ollama, :instances)
-          if configured.is_a?(Hash)
-            configured.each do |name, config|
-              instances[name.to_sym] = normalize_instance_config(config).merge(
-                tier: :direct,
-                capabilities: %i[completion embedding vision]
-              )
-            end
-          end
+          discover_local_instance(instances)
+          discover_configured_instances(instances)
 
+          log.debug { "ollama discovery returning instance_count=#{instances.size}" }
           instances
         end
 
@@ -77,6 +64,33 @@ module Legion
           normalized[:base_url] ||= normalized.delete(:endpoint)
           normalized.compact
         end
+
+        def self.discover_local_instance(instances)
+          log.debug { 'ollama discovery probing local socket host=127.0.0.1 port=11434' }
+          return unless CredentialSources.socket_open?('127.0.0.1', 11_434, timeout: 0.1)
+
+          log.debug { 'ollama discovery found local socket instance' }
+          instances[:local] = {
+            base_url: 'http://127.0.0.1:11434',
+            tier: :local,
+            capabilities: %i[completion embedding vision]
+          }
+        end
+
+        def self.discover_configured_instances(instances)
+          configured = CredentialSources.setting(:extensions, :llm, :ollama, :instances)
+          return unless configured.is_a?(Hash)
+
+          log.debug { "ollama discovery loading configured instance_count=#{configured.size}" }
+          configured.each do |name, config|
+            instances[name.to_sym] = normalize_instance_config(config).merge(
+              tier: :direct,
+              capabilities: %i[completion embedding vision]
+            )
+          end
+        end
+
+        private_class_method :discover_local_instance, :discover_configured_instances
       end
     end
   end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-ollama
 version: !ruby/object:Gem::Version
-  version: 0.2.7
+  version: 0.2.9
 platform: ruby
 authors:
 - LegionIO