ltcai 0.4.0 → 0.5.0

package/docs/CHANGELOG.md

@@ -1,5 +1,27 @@
 # Changelog
 
+## [0.5.0] - 2026-05-31
+
+> MLX 샘플링 API 호환성 버그 수정 + 릴리스 워크플로 build-only 전환.
+
+### Fixed
+
+- **MLX `temp` kwarg 제거 대응** — `llm_router.py`의 로컬 MLX 추론 경로(텍스트/
+  비전, 동기/스트리밍, 문서 생성 4계열·총 8개 호출부)가 `mlx_lm.generate` /
+  `mlx_vlm.generate`에 `temp=temperature`를 직접 넘기다가
+  `generate_step() got an unexpected keyword argument 'temp'`로 실패하던 문제
+  수정. mlx_lm ≥ 0.20 / mlx_vlm는 `temp` 키워드를 제거하고 `sampler` 콜러블을
+  받도록 API가 바뀌었으므로, `make_sampler(temp=...)`로 만든 sampler를
+  `sampler=`로 전달하도록 `_mlx_sampler()` 헬퍼를 도입.
+
+### Changed
+
+- **릴리스 워크플로 build-only 전환** — `.github/workflows/release.yml`이 v* 태그
+  push 시 단위 테스트와 빌드 산출물 생성(`python -m build`, `twine check`,
+  `npm pack`, `vsce package`)까지만 수행. `publish-pypi`/`publish-npm`/
+  `publish-vscode`/`publish-ovsx` job과 GitHub Secrets 의존(`if: secrets.*`)을
+  제거. 배포는 로컬에서 수동 인증 후 진행.
+
 ## [0.4.0] - 2026-05-31
 
 > Knowledge Graph v2 read/write cutover — legacy/v2 동등성 보장, dual-write

package/llm_router.py

@@ -227,6 +227,18 @@ def ensure_mlx_runtime() -> None:
     except Exception as e:
         raise RuntimeError(f"MLX runtime is not available after install: {e}") from e
 
+def _mlx_sampler(temperature: float):
+    """Build an MLX sampler callable for the given temperature.
+
+    mlx_lm >= 0.20 removed the ``temp`` keyword from generate_step in favour of a
+    ``sampler`` callable, and mlx_vlm follows the same convention. Passing
+    ``temp=`` to generate/stream_generate now raises
+    ``generate_step() got an unexpected keyword argument 'temp'``. Both libraries
+    accept ``sampler=`` and share make_sampler from mlx_lm.sample_utils.
+    """
+    from mlx_lm.sample_utils import make_sampler
+    return make_sampler(temp=temperature)
+
 class LLMRouter:
     def __init__(self):
         self._cache: Dict[str, Tuple] = {}
@@ -514,10 +526,10 @@ class LLMRouter:
             is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
             if is_gemma4 and VLM_AVAILABLE:
                 from mlx_vlm import generate as vlm_gen
-                return vlm_gen(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, temp=temperature, draft_model=draft_model, draft_kind="mtp")
+                return vlm_gen(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
             else:
                 from mlx_lm import generate as lm_gen
-                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, temp=temperature, draft_model=draft_model)
+                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
         result = await loop.run_in_executor(executor, _gen)
         # mlx-vlm might return a GenerationResult object; extract the text
         if hasattr(result, "text"):
@@ -571,10 +583,10 @@ class LLMRouter:
                 is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
                 if is_gemma4 and VLM_AVAILABLE:
                     from mlx_vlm import stream_generate as vlm_stream
-                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, temp=temperature, draft_model=draft_model, draft_kind="mtp")
+                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=self._prep_image(image_data), max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
                 else:
                     from mlx_lm import stream_generate as lm_stream
-                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, temp=temperature, draft_model=draft_model)
+                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
                 
                 for chunk in gen:
                     text = chunk.text if hasattr(chunk, "text") else (chunk[0] if isinstance(chunk, tuple) else str(chunk))
@@ -666,10 +678,10 @@ class LLMRouter:
             is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
             if is_gemma4 and VLM_AVAILABLE:
                 from mlx_vlm import generate as vlm_gen
-                return vlm_gen(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, temp=temperature, draft_model=draft_model, draft_kind="mtp")
+                return vlm_gen(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
             else:
                 from mlx_lm import generate as lm_gen
-                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, temp=temperature, draft_model=draft_model)
+                return lm_gen(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
         result = await loop.run_in_executor(executor, _gen)
         if hasattr(result, "text"):
             return normalize_branding(result.text)
@@ -733,10 +745,10 @@ class LLMRouter:
                 is_gemma4 = "gemma-4" in self._current.lower() or "gemma4" in self._current.lower()
                 if is_gemma4 and VLM_AVAILABLE:
                     from mlx_vlm import stream_generate as vlm_stream
-                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, temp=temperature, draft_model=draft_model, draft_kind="mtp")
+                    gen = vlm_stream(model, tokenizer, prompt=prompt, image=None, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model, draft_kind="mtp")
                 else:
                     from mlx_lm import stream_generate as lm_stream
-                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, temp=temperature, draft_model=draft_model)
+                    gen = lm_stream(model, tokenizer, prompt=prompt, max_tokens=max_tokens, sampler=_mlx_sampler(temperature), draft_model=draft_model)
                 for chunk in gen:
                     text = chunk.text if hasattr(chunk, "text") else (chunk[0] if isinstance(chunk, tuple) else str(chunk))
                     loop.call_soon_threadsafe(queue.put_nowait, text)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ltcai",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "Lattice AI local MLX/cloud LLM workspace server",
   "homepage": "https://github.com/TaeSooPark-PTS/LatticeAI#readme",
   "repository": {

package/server.py

@@ -1121,7 +1121,7 @@ async def lifespan(app: FastAPI):
             except Exception:
                 pass
 
-app = FastAPI(title=f"Lattice AI Server ({APP_MODE})", version="0.4.0", lifespan=lifespan)
+app = FastAPI(title=f"Lattice AI Server ({APP_MODE})", version="0.5.0", lifespan=lifespan)
 
 CORS_ALLOWED_ORIGINS = [
     f"http://localhost:{DEFAULT_PORT}",
@@ -3466,7 +3466,7 @@ async def verify_cloud_models(force: bool = False, provider_filter: Optional[str
 
 @app.get("/health")
 async def health(request: Request):
-    base = {"status": "ok", "version": "0.4.0", "mode": APP_MODE}
+    base = {"status": "ok", "version": "0.5.0", "mode": APP_MODE}
     if not get_current_user(request) and REQUIRE_AUTH:
         return base
     engines = await asyncio.to_thread(engine_status)