lex-llm-vllm 0.3.14 → 0.3.17

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: '0459c715b28893afc32019e6f2f831ee08824cd3a0224c17fa25164f5f14a9fc'
-  data.tar.gz: 72629182ec192e73316f29d1b491b471a1c8cfe6d87e348f2ce00a16f9df8a1b
+  metadata.gz: 142f55fe9f62c87ef3fe5fe545ba652ff73426af265aa9a0aaafa2bbb703ec35
+  data.tar.gz: ce24e483f5dea06abb6be25d26bedeaf531782b344500f73ed4fe1161e69c372
 SHA512:
-  metadata.gz: 60b334c314c6f29257ae7c9a073dfc034536f50f4c58e620137d49d88fb682d12f1abb5882ae792e7a99faffabd3cb01ba990f053f341d10b5478952f527e6bd
-  data.tar.gz: 3b602d7548c15a76068645d27da5a94619d7ca1954a2620c8291b2eb4af93384b2a4fe5f241b15487b3b8f7bb99a8294cd474ddac7f94ee1439d9e2d5f76ee08
+  metadata.gz: be6ebfd3ae8e78a9b0a64078cb7e947d08fb2f58cce8c61b5ea37a0181bcda83775d928b1f638e1518e9ca6640da196305203a169beeb746e6b13335b29f70a8
+  data.tar.gz: e74e87adb11b6d67071dfb053f72e3ef882fcf5f50b39e889b3d72071af3f41fe97fc524e89a6c20fe7606aea3e1db016d33459b965ea934aeeeb441c0b0dec1

data/.rubocop.yml

@@ -21,11 +21,8 @@ RSpec/ExampleLength:
   Max: 8
 RSpec/MultipleExpectations:
   Enabled: false
-
 RSpec/ExampleLength:
-  Max: 10
-
-
+  Enabled: false
 Layout/LineLength:
   Exclude:
     - spec/**/*

data/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## [0.3.17] - 2026-08-04
+
+### Fixed
+- Preserve each streamed tool call's wire index when bridging canonical vLLM chunks to legacy chunks, allowing `lex-llm` 0.6.16 to correlate interleaved argument fragments for parallel tool calls instead of attaching them by recency.
+
+## [0.3.16] - 2026-07-31
+
+### Fixed
+- **`parse_chunk` now handles multiple tool_calls batched in a single SSE delta.** vLLM can batch several parallel tool_calls into one `choices[0].delta.tool_calls` array. Previously only `tool_calls.first` was processed — the 2nd+ tool calls were silently dropped, causing parallel tool invocations to lose calls. Now returns an array of `tool_call_delta` chunks (one per tool_call), which the streaming handler already iterates (per lex-llm 0.6.13). Single tool_call deltas still return a single chunk (not wrapped in an array) for backward compatibility.
+
+## [0.3.15] - 2026-07-31
+
+### Fixed
+- **`to_legacy_chunk` now propagates `stop_reason` from canonical chunks.** The canonical translator correctly parsed vLLM's `finish_reason` into `Canonical::Chunk.stop_reason`, but `to_legacy_chunk` never passed it into the legacy `Legion::Extensions::Llm::Chunk` constructor. The `StreamAccumulator` already reads `chunk.stop_reason` (line 40), so the field was silently nil on every streamed chunk — downstream always defaulted to `:end_turn` regardless of what the provider actually said. Truncated responses (`:max_tokens`) and content-filtered responses (`:content_filter`) were indistinguishable from clean completions. Now the real finish_reason flows through: canonical chunk → legacy chunk → accumulator → assembled Message.
+
 ## [0.3.14] - 2026-07-24
 
 ### Fixed

data/lex-llm-vllm.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', '>= 0.6.9'
+  spec.add_dependency 'lex-llm', '>= 0.6.16'
 end

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

@@ -363,6 +363,7 @@ module Legion
               thinking: thinking,
               input_tokens: usage.respond_to?(:input_tokens) ? usage.input_tokens : nil,
               output_tokens: usage.respond_to?(:output_tokens) ? usage.output_tokens : nil,
+              stop_reason: canonical.stop_reason,
               raw: raw_data
             )
           end
@@ -370,7 +371,9 @@ module Legion
           # Map a canonical tool_call_delta onto the legacy chunk tool_calls hash.
           # Fragment semantics matter: an entry with a non-nil id starts a new tool
           # call in the StreamAccumulator; a nil id appends the raw arguments
-          # fragment to the most recently started call.
+          # fragment to the call at its wire index (block_index). Carrying the
+          # index is what lets the accumulator correlate interleaved parallel
+          # fragments to the right call instead of falling back to recency.
           def legacy_chunk_tool_calls(canonical)
             return nil unless canonical.type == :tool_call_delta && canonical.tool_call
 
@@ -380,7 +383,8 @@ module Legion
               key => Legion::Extensions::Llm::ToolCall.new(
                 id: tc.id,
                 name: tc.name,
-                arguments: tc.arguments
+                arguments: tc.arguments,
+                index: canonical.block_index
               )
             }
           end

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

@@ -172,8 +172,11 @@ module Legion
                 log.debug '[vllm][translator] action=content_dropped_with_tool_call ' \
                           "content=#{delta['content'][0, 100].inspect} request_id=#{request_id}"
               end
-              return build_tool_call_delta_chunk(tool_calls.first, request_id,
-                                                 stop_reason: chunk_stop_reason, usage: chunk_usage)
+              chunks = tool_calls.map do |tc|
+                build_tool_call_delta_chunk(tc, request_id,
+                                            stop_reason: chunk_stop_reason, usage: chunk_usage)
+              end
+              return chunks.size == 1 ? chunks.first : chunks
             end
 
             # Thinking delta from reasoning_content

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

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module Llm
       module Vllm
-        VERSION = '0.3.14'
+        VERSION = '0.3.17'
       end
     end
   end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-vllm
 version: !ruby/object:Gem::Version
-  version: 0.3.14
+  version: 0.3.17
 platform: ruby
 authors:
 - LegionIO
@@ -71,14 +71,14 @@ dependencies:
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.6.9
+        version: 0.6.16
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.6.9
+        version: 0.6.16
 description: vLLM provider integration for the LegionIO LLM routing framework.
 email:
 - matthewdiverson@gmail.com