lex-llm-ollama 0.2.22 → 0.2.24

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9105e1f109fd5c83078224391ceb669ea321943a2075d85d5a77cf48b73d16e7
-  data.tar.gz: 1bd18adb284f8b8fa5c12e0049f310973989efb5a20070dcbb172754f64f941b
+  metadata.gz: fe9668a684e60453a26c187e8e5dddfa15835ce18b4b6d627cb32bd17be1fef8
+  data.tar.gz: a858a4b969945856c39ba64397067ef1f7c8c776de47b669a1e7b7c8a6704abf
 SHA512:
-  metadata.gz: 175570e4fdf0574998741b731718675d643f03c1163183bd41a28a7c90151c71919150060c9d87ff27840f0bcf5203b90dba85e81680f1ec74d3f494b22c85f0
-  data.tar.gz: 9ebb03d6cdf2078303f8ac674cfe939a73770a0ce7b721e86c6daaeb63169c0039dfca9e7063138e83896322820939b203f68265da9a2125066ee4e05ecefcd4
+  metadata.gz: 52b01a8dfa63402561d7e5e9b09ed09d92ac7d01ef9dfa0fa04e15463399da4168f6948f0f23e5d1807d6b2f09f04d103079ff35959077d717bbb2f2889fa7aa
+  data.tar.gz: 3fccbb4c70de27e38f87d4809f628819f33ff9ce4a3640b7cf29c67deda5b01ed394797b7c1ddb438f980d680d2bd0c8768b4388a2e57ff8e1d62f1cc4ccc5b5

data/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## [0.2.24] - 2026-07-24
+
+### Fixed
+- **Translator propagates `done_reason` and usage through content chunks.** When ollama sends `done: true` on a chunk that also has content, the stop_reason and usage were previously lost. Now extracts them before branching and passes through to whatever chunk type is emitted.
+
+## [0.2.23] - 2026-07-05
+
+### Changed
+- Stop_reason mapping now uses the shared `Legion::Extensions::Llm::StopReasonMapping` mixin from lex-llm (>= 0.6.9) instead of a local `OLLAMA_STOP_REASON_MAP` (a copy of the same drifting 3-entry map vLLM carried). The shared vocabulary maps `tool_calls`/`tool_use`/`function_call` to `:tool_use` (plus `stop`/`end_turn`/`eos`, `length`/`max_tokens`, `stop_sequence`, `content_filter`) and is inherited by every provider so it no longer drifts per gem. Ollama's `done_reason`-to-stop_reason resolution is unchanged in behavior.
+
 ## [0.2.22] - 2026-06-20
 
 ### Fixed

data/lex-llm-ollama.gemspec

@@ -27,5 +27,5 @@ Gem::Specification.new do |spec|
   spec.add_dependency 'legion-logging', '>= 1.3.2'
   spec.add_dependency 'legion-settings', '>= 1.3.14'
   spec.add_dependency 'legion-transport', '>= 1.4.14'
-  spec.add_dependency 'lex-llm', '>= 0.6.0'
+  spec.add_dependency 'lex-llm', '>= 0.6.9'
 end

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

@@ -21,13 +21,10 @@ module Legion
         # - assistant_prefill: false — Ollama does not support assistant prefill.
         class Translator
           include Legion::Logging::Helper
+          # Shared stop_reason vocabulary. Ollama returns structured tool_calls and
+          # OpenAI-style done_reason strings; inherits the common map.
+          include Legion::Extensions::Llm::StopReasonMapping
 
-          # Ollama-specific stop_reason mapping (done_reason field).
-          OLLAMA_STOP_REASON_MAP = {
-            'stop' => :end_turn,
-            'tool_use' => :tool_use,
-            'length' => :max_tokens
-          }.freeze
           FALLBACK_STOP_REASON = :end_turn
 
           # G18 parameter mapping: canonical params -> Ollama options keys.
@@ -392,7 +389,7 @@ module Legion
 
           def map_stop_reason(done_reason, done = nil)
             if done_reason
-              OLLAMA_STOP_REASON_MAP.fetch(done_reason.to_s, FALLBACK_STOP_REASON)
+              stop_reason_lookup(done_reason) || FALLBACK_STOP_REASON
             elsif done
               FALLBACK_STOP_REASON
             end
@@ -420,16 +417,24 @@ module Legion
             done_reason = data[:done_reason] || data['done_reason']
             request_id = data[:request_id] || data['request_id'] || data[:id] || data['id']
 
+            chunk_stop_reason = done_reason ? map_stop_reason(done_reason, done) : nil
+            chunk_usage = done ? build_chunk_usage(data) : nil
+
             # Tool call delta
             tool_calls = message[:tool_calls] || message['tool_calls']
-            return build_tool_call_chunk(tool_calls, request_id) unless Array(tool_calls).empty?
+            unless Array(tool_calls).empty?
+              return build_tool_call_chunk(tool_calls, request_id,
+                                           stop_reason: chunk_stop_reason, usage: chunk_usage)
+            end
 
             # Thinking delta
             thinking_content = message[:thinking] || message['thinking']
             unless thinking_content.to_s.empty?
               return Canonical::Chunk.thinking_delta(
                 delta: thinking_content.to_s,
-                request_id: request_id
+                request_id: request_id,
+                stop_reason: chunk_stop_reason,
+                usage: chunk_usage
               )
             end
 
@@ -438,7 +443,9 @@ module Legion
             unless content.to_s.empty?
               return Canonical::Chunk.text_delta(
                 delta: content.to_s,
-                request_id: request_id
+                request_id: request_id,
+                stop_reason: chunk_stop_reason,
+                usage: chunk_usage
               )
             end
 
@@ -448,20 +455,22 @@ module Legion
             nil
           end
 
-          def build_done_chunk(data, done_reason, request_id)
-            usage = Canonical::Usage.from_hash({
-                                                 input_tokens: data[:prompt_eval_count] || data['prompt_eval_count'],
-                                                 output_tokens: data[:eval_count] || data['eval_count']
-                                               })
+          def build_chunk_usage(data)
+            Canonical::Usage.from_hash({
+                                         input_tokens: data[:prompt_eval_count] || data['prompt_eval_count'],
+                                         output_tokens: data[:eval_count] || data['eval_count']
+                                       })
+          end
 
+          def build_done_chunk(data, done_reason, request_id)
             Canonical::Chunk.done(
               request_id: request_id,
-              usage: usage,
+              usage: build_chunk_usage(data),
               stop_reason: map_stop_reason(done_reason, true)
             )
           end
 
-          def build_tool_call_chunk(tool_calls, request_id)
+          def build_tool_call_chunk(tool_calls, request_id, stop_reason: nil, usage: nil)
             first_call = tool_calls.first
             first_call = first_call.transform_keys(&:to_sym) if first_call.is_a?(Hash)
             function = first_call[:function] || first_call['function'] || {}
@@ -476,7 +485,9 @@ module Legion
 
             Canonical::Chunk.tool_call_delta(
               tool_call: tc,
-              request_id: request_id
+              request_id: request_id,
+              stop_reason: stop_reason,
+              usage: usage
             )
           end
 

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

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module Llm
       module Ollama
-        VERSION = '0.2.22'
+        VERSION = '0.2.24'
       end
     end
   end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-ollama
 version: !ruby/object:Gem::Version
-  version: 0.2.22
+  version: 0.2.24
 platform: ruby
 authors:
 - LegionIO
@@ -71,14 +71,14 @@ dependencies:
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.6.0
+        version: 0.6.9
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.6.0
+        version: 0.6.9
 description: Ollama provider integration for the LegionIO LLM routing framework.
 email:
 - matthewdiverson@gmail.com