convertible-cli 0.2.0__tar.gz → 0.2.1__tar.gz

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/CHANGELOG.md

@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 Format follows [Keep a Changelog](https://keepachangelog.com/). This project
 adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.1] - 2026-05-27
+
+### Changed
+
+- Generalized the vLLM tool-call-parser docs from hermes-specific to model-appropriate (e.g. hermes or qwen3_coder); the engine is parser-agnostic and only needs OpenAI-format tool calls. Touches vllm_openai docstring, the live-proof test docstring, README, and CLAUDE.md.
+
 ## [0.2.0] - 2026-05-27
 
 ### Added

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/CLAUDE.md

@@ -75,7 +75,8 @@ uv run teken cli doctor . --strict        # agent-first rubric gate
 ```
 
 The live vLLM proof is opt-in (the reference rig must expose tool calling:
-`--enable-auto-tool-choice --tool-call-parser hermes`):
+`--enable-auto-tool-choice` plus a model-appropriate `--tool-call-parser`, e.g.
+`hermes` or `qwen3_coder`):
 
 ```bash
 CONVERTIBLE_VLLM_E2E=1 uv run pytest tests/test_vllm_live.py -v

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: convertible-cli
-Version: 0.2.0
+Version: 0.2.1
 Summary: Convertible CLI is a swappable coder-agent harness that turns different models into repo workers behind one shared task contract.
 Project-URL: Homepage, https://github.com/agentculture/convertible
 Project-URL: Issues, https://github.com/agentculture/convertible/issues
@@ -86,6 +86,12 @@ vllm serve Qwen/Qwen3-32B \
   --tool-call-parser hermes
 ```
 
+The right `--tool-call-parser` depends on the model **and** the vLLM build:
+`hermes` works for many models (including `Qwen/Qwen3-32B` above), while other
+builds need a different one — e.g. an NVFP4 Qwen3 checkpoint served via vLLM may
+want `qwen3_coder`. The engine itself is parser-agnostic — any parser that makes
+the server emit OpenAI-format tool calls works.
+
 Then point Convertible at it (defaults already target `localhost:8001`):
 
 ```bash

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/README.md

@@ -69,6 +69,12 @@ vllm serve Qwen/Qwen3-32B \
   --tool-call-parser hermes
 ```
 
+The right `--tool-call-parser` depends on the model **and** the vLLM build:
+`hermes` works for many models (including `Qwen/Qwen3-32B` above), while other
+builds need a different one — e.g. an NVFP4 Qwen3 checkpoint served via vLLM may
+want `qwen3_coder`. The engine itself is parser-agnostic — any parser that makes
+the server emit OpenAI-format tool calls works.
+
 Then point Convertible at it (defaults already target `localhost:8001`):
 
 ```bash

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/convertible/engines/vllm_openai.py

@@ -7,8 +7,10 @@ stdlib ``urllib`` rather than any vendor SDK — pointing :class:`EngineConfig`
 any OpenAI-compatible server (vLLM, llama.cpp, an OpenAI proxy) is a config
 change, never a code change (honesty condition h2).
 
-vLLM note: Qwen-family tool calling needs the server started with
-``--enable-auto-tool-choice --tool-call-parser hermes``.
+vLLM note: tool calling needs the server started with ``--enable-auto-tool-choice``
+and a ``--tool-call-parser`` matching the model (e.g. ``hermes`` for many models,
+``qwen3_coder`` for some Qwen3 builds). The engine is parser-agnostic — it only
+needs the server to emit OpenAI-format tool calls.
 """
 
 from __future__ import annotations

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "convertible-cli"
-version = "0.2.0"
+version = "0.2.1"
 description = "Convertible CLI is a swappable coder-agent harness that turns different models into repo workers behind one shared task contract."
 readme = "README.md"
 license = "MIT"

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/tests/test_vllm_live.py

@@ -12,8 +12,8 @@ Run it (with the reference rig up) like::
     CONVERTIBLE_MODEL=Qwen/Qwen3-32B \\
     uv run pytest tests/test_vllm_live.py -v
 
-The server must expose tool calling (vLLM:
-``--enable-auto-tool-choice --tool-call-parser hermes``).
+The server must expose tool calling (vLLM: ``--enable-auto-tool-choice`` plus a
+``--tool-call-parser`` for the model, e.g. ``hermes`` or ``qwen3_coder``).
 """
 
 from __future__ import annotations

{convertible_cli-0.2.0 → convertible_cli-0.2.1}/uv.lock

@@ -72,7 +72,7 @@ wheels = [
 
 [[package]]
 name = "convertible-cli"
-version = "0.2.0"
+version = "0.2.1"
 source = { editable = "." }
 
 [package.dev-dependencies]