hedit 0.7.8.dev1__tar.gz → 0.7.9.dev1__tar.gz

{hedit-0.7.8.dev1/hedit.egg-info → hedit-0.7.9.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hedit
-Version: 0.7.8.dev1
+Version: 0.7.9.dev1
 Summary: Multi-agent system for HED annotation generation and validation
 Author-email: Annotation Garden Initiative <info@annotation.garden>
 License-Expression: MIT

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1/hedit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hedit
-Version: 0.7.8.dev1
+Version: 0.7.9.dev1
 Summary: Multi-agent system for HED annotation generation and validation
 Author-email: Annotation Garden Initiative <info@annotation.garden>
 License-Expression: MIT

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "hedit"
-version = "0.7.8.dev1"
+version = "0.7.9.dev1"
 description = "Multi-agent system for HED annotation generation and validation"
 readme = "PKG_README.md"
 requires-python = ">=3.12"

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/api/main.py

@@ -99,8 +99,8 @@ def create_openrouter_workflow(
         api_key: OpenRouter API key
         annotation_model: Model for annotation (default: ANNOTATION_MODEL env or Claude Haiku 4.5)
         annotation_provider: Provider for annotation model (default: ANNOTATION_PROVIDER env or "anthropic")
-        eval_model: Model for eval/assessment/feedback (default: EVALUATION_MODEL env or GPT-OSS-120B)
-        eval_provider: Provider for eval models (default: EVALUATION_PROVIDER env or "groq")
+        eval_model: Model for eval/assessment/feedback (default: EVALUATION_MODEL env or Qwen3.5-122B)
+        eval_provider: Provider for eval models (default: EVALUATION_PROVIDER env or "alibaba")
         temperature: LLM temperature (default: 0.1)
         user_id: User ID for cache optimization (derived from API key if not provided)
         schema_dir: Path to HED schemas (None = fetch from GitHub)
@@ -113,8 +113,8 @@ def create_openrouter_workflow(
     # Apply defaults from environment
     default_annotation_model = os.getenv("ANNOTATION_MODEL", "anthropic/claude-haiku-4.5")
     default_annotation_provider = os.getenv("ANNOTATION_PROVIDER", "anthropic")
-    default_eval_model = os.getenv("EVALUATION_MODEL", "qwen/qwen3.5-397b-a17b")
-    default_eval_provider = os.getenv("EVALUATION_PROVIDER")
+    default_eval_model = os.getenv("EVALUATION_MODEL", "qwen/qwen3.5-122b-a10b")
+    default_eval_provider = os.getenv("EVALUATION_PROVIDER", "alibaba")
 
     # Resolve final values: parameter > env var > default
     actual_annotation_model = get_model_name(annotation_model or default_annotation_model)
@@ -240,8 +240,8 @@ def create_byok_vision_agent(
         Configured VisionAgent using the user's key and model settings
     """
     # Use user-provided settings or fall back to server defaults
-    default_vision_model = os.getenv("VISION_MODEL", "qwen/qwen3-vl-32b-instruct")
-    default_vision_provider = os.getenv("VISION_PROVIDER", "novita")
+    default_vision_model = os.getenv("VISION_MODEL", "qwen/qwen3.5-122b-a10b")
+    default_vision_provider = os.getenv("VISION_PROVIDER", "alibaba")
 
     actual_model = vision_model if vision_model else default_vision_model
     actual_temperature = temperature if temperature is not None else 0.3
@@ -354,9 +354,9 @@ async def lifespan(app: FastAPI):
         # Log configuration (env vars are read by create_openrouter_workflow)
         print("Using OpenRouter with models:")
         print(f"  Annotation: {os.getenv('ANNOTATION_MODEL', 'anthropic/claude-haiku-4.5')}")
-        print(f"  Evaluation: {os.getenv('EVALUATION_MODEL', 'qwen/qwen3.5-397b-a17b')}")
+        print(f"  Evaluation: {os.getenv('EVALUATION_MODEL', 'qwen/qwen3.5-122b-a10b')}")
         print(f"  Provider (annotation): {os.getenv('ANNOTATION_PROVIDER', 'anthropic')}")
-        print(f"  Provider (eval): {os.getenv('EVALUATION_PROVIDER', '') or '(auto-routed)'}")
+        print(f"  Provider (eval): {os.getenv('EVALUATION_PROVIDER', 'alibaba')}")
 
         workflow = create_openrouter_workflow(
             api_key=openrouter_api_key,
@@ -395,8 +395,8 @@ async def lifespan(app: FastAPI):
 
     # Initialize vision agent (only for OpenRouter)
     if llm_provider == "openrouter":
-        vision_model = os.getenv("VISION_MODEL", "qwen/qwen3-vl-32b-instruct")
-        vision_provider = os.getenv("VISION_PROVIDER", "novita")
+        vision_model = os.getenv("VISION_MODEL", "qwen/qwen3.5-122b-a10b")
+        vision_provider = os.getenv("VISION_PROVIDER", "alibaba")
 
         print(f"Initializing vision model: {vision_model} (provider: {vision_provider})")
 
@@ -486,6 +486,7 @@ app.add_middleware(
         "X-OpenRouter-Key",  # BYOK mode
         "X-OpenRouter-Model",  # BYOK model override
         "X-OpenRouter-Vision-Model",  # BYOK vision model override
+        "X-OpenRouter-Vision-Provider",  # BYOK vision provider override
         "X-OpenRouter-Provider",  # BYOK provider preference
         "X-OpenRouter-Temperature",  # BYOK temperature override
         "X-OpenRouter-Eval-Model",  # BYOK eval model override
@@ -754,6 +755,9 @@ async def annotate_from_image(
     # Check for model override headers (from frontend dropdown or CLI)
     model_override = request.model or req.headers.get("x-openrouter-model")
     vision_model_override = request.vision_model or req.headers.get("x-openrouter-vision-model")
+    vision_provider_override = request.vision_provider or req.headers.get(
+        "x-openrouter-vision-provider"
+    )
     provider_override = request.provider or req.headers.get("x-openrouter-provider")
     eval_model_override = req.headers.get("x-openrouter-eval-model")
     eval_provider_override = req.headers.get("x-openrouter-eval-provider")
@@ -782,14 +786,20 @@ async def annotate_from_image(
                 temperature=temperature,
                 user_id_override=user_id_override,
             )
+            # Vision uses its own provider; fall back to annotation provider only if
+            # a custom vision_model was specified without an explicit vision_provider
+            vision_provider = vision_provider_override or (
+                provider_override if vision_model_override else None
+            )
             active_vision_agent = create_byok_vision_agent(
                 openrouter_key,
                 vision_model=vision_model_override,
-                provider=provider_override,
+                provider=vision_provider,
                 temperature=temperature,
                 user_id_override=user_id_override,
             )
         except Exception as e:
+            logging.exception("Failed to initialize BYOK agents")
             raise HTTPException(
                 status_code=500, detail=f"Failed to initialize BYOK agents: {str(e)}"
             ) from e
@@ -815,9 +825,10 @@ async def annotate_from_image(
                 validator_path=_byok_config.get("validator_path"),
                 use_js_validator=_byok_config.get("use_js_validator", True),
             )
-            # Note: Vision agent uses its own provider (novita for qwen-vl)
-            # Only pass provider_override to vision if a custom vision_model was specified
-            vision_provider = provider_override if vision_model_override else None
+            # Note: Vision agent uses the vision-specific provider, not the annotation provider
+            vision_provider = vision_provider_override or (
+                provider_override if vision_model_override else None
+            )
             active_vision_agent = create_byok_vision_agent(
                 server_api_key,
                 vision_model=vision_model_override,
@@ -1329,6 +1340,9 @@ async def annotate_from_image_stream(
     # Determine which workflow and vision agent to use (same logic as /annotate-from-image)
     model_override = request.model or req.headers.get("x-openrouter-model")
     vision_model_override = request.vision_model or req.headers.get("x-openrouter-vision-model")
+    vision_provider_override = request.vision_provider or req.headers.get(
+        "x-openrouter-vision-provider"
+    )
     provider_override = request.provider or req.headers.get("x-openrouter-provider")
     eval_model_override = req.headers.get("x-openrouter-eval-model")
     eval_provider_override = req.headers.get("x-openrouter-eval-provider")
@@ -1355,14 +1369,20 @@ async def annotate_from_image_stream(
                 temperature=temperature,
                 user_id_override=user_id_override,
             )
+            # Vision uses its own provider; fall back to annotation provider only if
+            # a custom vision_model was specified without an explicit vision_provider
+            vision_provider = vision_provider_override or (
+                provider_override if vision_model_override else None
+            )
             active_vision_agent = create_byok_vision_agent(
                 openrouter_key,
                 vision_model=vision_model_override,
-                provider=provider_override,
+                provider=vision_provider,
                 temperature=temperature,
                 user_id_override=user_id_override,
             )
         except Exception as e:
+            logging.exception("Failed to initialize BYOK agents")
             raise HTTPException(
                 status_code=500, detail=f"Failed to initialize BYOK agents: {str(e)}"
             ) from e
@@ -1386,9 +1406,10 @@ async def annotate_from_image_stream(
                 validator_path=_byok_config.get("validator_path"),
                 use_js_validator=_byok_config.get("use_js_validator", True),
             )
-            # Note: Vision agent uses its own provider (novita for qwen-vl)
-            # Only pass provider_override to vision if a custom vision_model was specified
-            vision_provider = provider_override if vision_model_override else None
+            # Note: Vision agent uses the vision-specific provider, not the annotation provider
+            vision_provider = vision_provider_override or (
+                provider_override if vision_model_override else None
+            )
             active_vision_agent = create_byok_vision_agent(
                 server_api_key,
                 vision_model=vision_model_override,

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/api/models.py

@@ -41,12 +41,12 @@ class AnnotationRequest(BaseModel):
     model: str | None = Field(
         default=None,
         description="Override model for annotation (BYOK mode only, e.g., 'openai/gpt-4o')",
-        examples=["anthropic/claude-haiku-4.5", "qwen/qwen3.5-122b-a10b", "qwen/qwen3.5-397b-a17b"],
+        examples=["anthropic/claude-haiku-4.5", "qwen/qwen3.5-122b-a10b"],
     )
     provider: str | None = Field(
         default=None,
-        description="Override provider preference (BYOK mode only, e.g., 'Cerebras')",
-        examples=["Cerebras", "deepinfra/fp8", None],
+        description="Override provider preference (BYOK mode only, e.g., 'anthropic')",
+        examples=["anthropic", "alibaba", None],
     )
     temperature: float | None = Field(
         default=None,
@@ -177,12 +177,17 @@ class ImageAnnotationRequest(BaseModel):
     vision_model: str | None = Field(
         default=None,
         description="Override vision model for image description (BYOK mode only)",
-        examples=["qwen/qwen3-vl-32b-instruct", "qwen/qwen3-vl-235b-a22b-instruct"],
+        examples=["qwen/qwen3.5-122b-a10b", "qwen/qwen3-vl-235b-a22b-instruct"],
+    )
+    vision_provider: str | None = Field(
+        default=None,
+        description="Override vision model provider (BYOK mode only, e.g., 'alibaba')",
+        examples=["alibaba", "novita", None],
     )
     provider: str | None = Field(
         default=None,
-        description="Override provider preference (BYOK mode only, e.g., 'Cerebras')",
-        examples=["Cerebras", "deepinfra/fp8", None],
+        description="Override annotation provider preference (BYOK mode only, e.g., 'anthropic')",
+        examples=["anthropic", "alibaba", None],
     )
     temperature: float | None = Field(
         default=None,

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/cli/client.py

@@ -139,7 +139,14 @@ class HEDitClient:
             raise APIError(
                 "Service unavailable",
                 status_code=503,
-                detail="The API is temporarily unavailable. Please try again later.",
+                detail=detail or "The API is temporarily unavailable. Please try again later.",
+            )
+        elif response.status_code == 504:
+            raise APIError(
+                "Gateway timeout",
+                status_code=504,
+                detail="The server took too long to respond. Try a faster model/provider "
+                "or use --standalone mode.",
             )
         else:
             raise APIError(

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/cli/config.py

@@ -41,13 +41,13 @@ DEFAULT_DEV_API_URL = "https://api.annotation.garden/hedit-dev"
 DEFAULT_MODEL = "anthropic/claude-haiku-4.5"
 DEFAULT_PROVIDER = "anthropic"
 
-# Evaluation model: Qwen3.5-397B MoE (most capable Qwen, high throughput)
-DEFAULT_EVAL_MODEL = "qwen/qwen3.5-397b-a17b"
-DEFAULT_EVAL_PROVIDER = None
+# Evaluation model: Qwen3.5-122B MoE (fast, capable, cost-effective via Alibaba)
+DEFAULT_EVAL_MODEL = "qwen/qwen3.5-122b-a10b"
+DEFAULT_EVAL_PROVIDER = "alibaba"
 
-# Vision model: Qwen3-VL-32B for image descriptions
-DEFAULT_VISION_MODEL = "qwen/qwen3-vl-32b-instruct"
-DEFAULT_VISION_PROVIDER = None
+# Vision model: Qwen3.5-122B MoE (accepts vision, fast via Alibaba)
+DEFAULT_VISION_MODEL = "qwen/qwen3.5-122b-a10b"
+DEFAULT_VISION_PROVIDER = "alibaba"
 
 
 class CredentialsConfig(BaseModel):
@@ -69,12 +69,12 @@ class ModelsConfig(BaseModel):
     )
     eval_provider: str | None = Field(
         default=DEFAULT_EVAL_PROVIDER,
-        description="Provider for evaluation model (None = OpenRouter auto-routes)",
+        description="Provider for evaluation model (default: alibaba)",
     )
     vision: str = Field(default=DEFAULT_VISION_MODEL, description="Vision model for images")
     vision_provider: str | None = Field(
         default=DEFAULT_VISION_PROVIDER,
-        description="Provider for vision model (deepinfra/fp8 for qwen-vl)",
+        description="Provider for vision model (alibaba for qwen)",
     )
     temperature: float = Field(default=0.1, ge=0.0, le=1.0, description="Model temperature")
 
@@ -237,8 +237,8 @@ def get_effective_config(
         api_url: Override API URL
         model: Override model (if non-default, clears provider unless explicitly set)
         eval_model: Override evaluation model (for consistent benchmarking)
-        eval_provider: Override provider for evaluation model (e.g., "Cerebras")
-        provider: Override provider preference (e.g., "Cerebras")
+        eval_provider: Override provider for evaluation model (e.g., "alibaba")
+        provider: Override provider preference (e.g., "anthropic")
         temperature: Override temperature
         schema_version: Override schema version
         output_format: Override output format

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/cli/local_executor.py

@@ -68,6 +68,7 @@ class LocalExecutionBackend(ExecutionBackend):
         eval_model: str | None = None,
         eval_provider: str | None = None,
         vision_model: str | None = None,
+        vision_provider: str | None = None,
         provider: str | None = None,
         temperature: float = 0.1,
         schema_dir: Path | str | None = None,
@@ -78,9 +79,10 @@ class LocalExecutionBackend(ExecutionBackend):
         Args:
             api_key: OpenRouter API key (required for LLM operations, optional for health/validate)
             model: Model for text annotation (default: anthropic/claude-haiku-4.5)
-            eval_model: Model for evaluation/assessment agents (default: qwen/qwen3.5-397b-a17b)
-            eval_provider: Provider for evaluation model (None = OpenRouter auto-routes)
-            vision_model: Model for image annotation (default: qwen/qwen3-vl-32b-instruct)
+            eval_model: Model for evaluation/assessment agents (default: qwen/qwen3.5-122b-a10b)
+            eval_provider: Provider for evaluation model (default: alibaba)
+            vision_model: Model for image annotation (default: qwen/qwen3.5-122b-a10b)
+            vision_provider: Provider for vision model (default: alibaba)
             provider: Provider preference (cleared if custom model specified)
             temperature: LLM temperature (0.0-1.0)
             schema_dir: Optional directory with JSON schemas (None = fetch from GitHub)
@@ -102,7 +104,7 @@ class LocalExecutionBackend(ExecutionBackend):
         self._eval_model = eval_model or DEFAULT_EVAL_MODEL
         self._eval_provider = eval_provider or DEFAULT_EVAL_PROVIDER
         self._vision_model = vision_model or DEFAULT_VISION_MODEL
-        self._vision_provider = DEFAULT_VISION_PROVIDER
+        self._vision_provider = vision_provider or DEFAULT_VISION_PROVIDER
         self._temperature = temperature
         self._schema_dir = Path(schema_dir) if schema_dir else None
         self._user_id = user_id  # Custom user ID (None = use auto-generated machine ID)

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/cli/main.py

@@ -204,6 +204,7 @@ def get_executor(
             eval_model=config.models.evaluation,
             eval_provider=config.models.eval_provider,
             vision_model=config.models.vision,
+            vision_provider=config.models.vision_provider,
             provider=config.models.provider,
             temperature=config.models.temperature,
             user_id=user_id,

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/utils/openrouter_llm.py

@@ -28,11 +28,11 @@ def create_openrouter_llm(
     to include cache_control markers for 90% cost reduction on cache hits.
 
     Args:
-        model: Model identifier (e.g., "anthropic/claude-haiku-4.5", "qwen/qwen3.5-397b-a17b")
+        model: Model identifier (e.g., "anthropic/claude-haiku-4.5", "qwen/qwen3.5-122b-a10b")
         api_key: OpenRouter API key (defaults to OPENROUTER_API_KEY env var)
         temperature: Sampling temperature (0.0-1.0)
         max_tokens: Maximum tokens to generate
-        provider: Specific provider to use (e.g., "Cerebras", "Anthropic")
+        provider: Specific provider to use (e.g., "anthropic", "alibaba")
         user_id: User identifier for cache optimization (sticky routing)
         enable_caching: Enable Anthropic prompt caching. If None (default),
             auto-enables for Anthropic Claude models.

{hedit-0.7.8.dev1 → hedit-0.7.9.dev1}/src/version.py

@@ -1,7 +1,7 @@
 """Version information for HEDit."""
 
-__version__ = "0.7.8.dev1"
-__version_info__ = (0, 7, 8, "dev")
+__version__ = "0.7.9.dev1"
+__version_info__ = (0, 7, 9, "dev")
 
 
 def get_version() -> str: