model-library 0.1.6__py3-none-any.whl → 0.1.8__py3-none-any.whl

model_library/base/utils.py

@@ -1,18 +1,36 @@
-from typing import Sequence, TypeVar, cast
+import json
+import re
+from datetime import datetime, timedelta
+from typing import Any, Sequence, TypeVar
+
+from pydantic import BaseModel
 
 from model_library.base.input import (
     FileBase,
     InputItem,
-    RawInputItem,
+    RawInput,
+    RawResponse,
     TextInput,
     ToolResult,
 )
 from model_library.utils import truncate_str
-from pydantic import BaseModel
 
 T = TypeVar("T", bound=BaseModel)
 
 
+def serialize_for_tokenizing(content: Any) -> str:
+    """
+    Serialize parsed content into a string for tokenization
+    """
+    parts: list[str] = []
+    if content:
+        if isinstance(content, str):
+            parts.append(content)
+        else:
+            parts.append(json.dumps(content, default=str))
+    return "\n".join(parts)
+
+
 def add_optional(
     a: int | float | T | None, b: int | float | T | None
 ) -> int | float | T | None:
@@ -54,12 +72,43 @@ def get_pretty_input_types(input: Sequence["InputItem"], verbose: bool = False)
                 return repr(item)
             case ToolResult():
                 return repr(item)
-            case dict():
-                item = cast(RawInputItem, item)
+            case RawInput():
                 return repr(item)
-            case _:
-                # RawResponse
+            case RawResponse():
                 return repr(item)
 
     processed_items = [f"  {process_item(item)}" for item in input]
     return "\n" + "\n".join(processed_items) if processed_items else ""
+
+
+TIME_PATTERN = re.compile(r"^(\d+(?:\.\d+)?)([a-zA-Z]+)$")
+UNIT_TO_SECONDS = {
+    "ms": 0.001,
+    "s": 1,
+    "m": 60,
+    "h": 3600,
+}
+
+
+def to_timestamp(input_str: str, server_now: datetime) -> int:
+    """Converts a header string into a server-relative Unix timestamp in ms."""
+    input_str = input_str.strip()
+
+    # ISO Timestamp (e.g. 2026-01-09T21:58:01Z)
+    if "T" in input_str and "-" in input_str:
+        try:
+            dt = datetime.fromisoformat(input_str.replace("Z", "+00:00"))
+            return int(dt.timestamp() * 1000)
+        except ValueError:
+            pass
+
+    # Duration (e.g. 10s, 6ms)
+    match = TIME_PATTERN.match(input_str)
+    if match:
+        value, unit = match.groups()
+        offset_seconds = float(value) * UNIT_TO_SECONDS.get(unit.lower(), 0)
+        # Add duration to the SERVER'S provided date
+        dt = server_now + timedelta(seconds=offset_seconds)
+        return int(dt.timestamp() * 1000)
+
+    raise ValueError(f"Unsupported time format: {input_str}")

model_library/config/alibaba_models.yaml

@@ -1,17 +1,51 @@
-qwen-models:
-  base-config:
-    company: Alibaba
-    open_source: false
+base-config:
+  company: Alibaba
+  open_source: false
+  supports:
+    temperature: true
+  metadata:
+    available_for_everyone: false
+    available_as_evaluator: false
+  default_parameters:
+    temperature: 0.7
+  properties:
+    reasoning_model: false
 
+qwen-3-vl-models:
+  base-config:
     supports:
-      temperature: true
+      images: true
+
+  alibaba/qwen3-vl-plus-2025-09-23:
+    label: Qwen 3 VL Plus
+    open_source: true
+    description: Qwen 3 VL Plus (2025-09-23)
+    release_date: 2025-09-23
     metadata:
-      available_for_everyone: false
-      available_as_evaluator: false
-    default_parameters:
-      temperature: 0.7
+      deprecated: true
     properties:
+      context_window: 262_144
+      max_tokens: 32_768
+      training_cutoff: ""
       reasoning_model: false
+    costs_per_million_token:
+      input: 0.2
+      output: 1.6
+
+qwen-3-max-models:
+  base-config:
+    supports:
+      tools: true
+      images: false
+
+  alibaba/qwen3-max-2026-01-23:
+    label: Qwen 3 Max Thinking
+    description: Qwen 3 Max with enhanced reasoning capabilities
+    release_date: 2026-01-23
+    properties:
+      context_window: 256_000
+      max_tokens: 32_000
+      reasoning_model: true
 
   alibaba/qwen3-max-preview:
     label: Qwen 3 Max Preview
@@ -20,15 +54,7 @@ qwen-models:
     properties:
       context_window: 262_144
       max_tokens: 65_536
-      training_cutoff: ""
-    costs_per_million_token:
-      input: 1.2
-      output: 6
-    supports:
-      images: false
-      tools: true
-    metadata:
-      available_for_everyone: false
+      reasoning_model: true
 
   alibaba/qwen3-max-2025-09-23:
     label: Qwen 3 Max 2025-09-23
@@ -39,14 +65,6 @@ qwen-models:
       max_tokens: 65_536
       training_cutoff: ""
       reasoning_model: true
-    costs_per_million_token:
-      input: 1.2
-      output: 6
-    supports:
-      images: false
-      tools: true
-    metadata:
-      available_for_everyone: false
 
   alibaba/qwen3-max:
     label: Qwen 3 Max
@@ -57,34 +75,3 @@ qwen-models:
       max_tokens: 65_536
       training_cutoff: ""
       reasoning_model: false
-    costs_per_million_token:
-      input: 1.2
-      output: 6
-      cache:
-        read_discount: 0.8
-        write_markup: 1
-      context:
-        threshold: 32_000
-        input: 2.4
-        output: 12
-    supports:
-      images: false
-      tools: true
-    metadata:
-      available_for_everyone: false
-
-  alibaba/qwen3-vl-plus-2025-09-23:
-    label: Qwen 3 VL Plus
-    open_source: true
-    description: Qwen 3 VL Plus (2025-09-23)
-    release_date: 2025-09-23
-    properties:
-      context_window: 262_144
-      max_tokens: 32_768
-      training_cutoff: ""
-      reasoning_model: false
-    costs_per_million_token:
-      input: 0.2
-      output: 1.6
-    supports:
-      images: true