Added MAX_PASSAGE_LENGTH and missing deps mlx and psutil

.env.example

@@ -30,6 +30,7 @@ BATCH_SIZE=32
 MAX_BATCH_SIZE=128
 MAX_TEXTS_PER_REQUEST=100
 MAX_PASSAGES_PER_RERANK=1000
+MAX_PASSAGE_LENGTH=4096
 MAX_SEQUENCE_LENGTH=512
 DEVICE_MEMORY_FRACTION=0.8
 REQUEST_TIMEOUT=300

app/models/requests.py

@@ -2,6 +2,7 @@
 Pydantic models for API requests.
 """
 
+import os
 from typing import List, Literal, Optional
 
 from pydantic import BaseModel, Field, field_validator
@@ -30,9 +31,7 @@ class EmbedRequest(BaseModel):
         description="Whether to automatically truncate texts exceeding token limits",
         json_schema_extra={"example": True},
     )
-    truncation_strategy: Optional[
-        Literal["smart_truncate", "truncate", "extract", "error"]
-    ] = Field(
+    truncation_strategy: Optional[Literal["smart_truncate", "truncate", "extract", "error"]] = Field(
         "smart_truncate",
         description=(
             "Strategy for handling long texts: smart_truncate (preserve sentences), "
@@ -53,7 +52,7 @@ class EmbedRequest(BaseModel):
         json_schema_extra={"example": False},
     )
 
-    @field_validator('texts')
+    @field_validator("texts")
     @classmethod
     def validate_texts(cls, v, info):
         """Validate texts array with enhanced error handling."""
@@ -64,7 +63,7 @@ def validate_texts(cls, v, info):
             raise ValueError("texts cannot be empty")
 
         # Get auto_truncate setting from the same data
-        auto_truncate = info.data.get('auto_truncate', True) if info.data else True
+        auto_truncate = info.data.get("auto_truncate", True) if info.data else True
 
         for i, text in enumerate(v):
             if not isinstance(text, str):
@@ -79,12 +78,12 @@ def validate_texts(cls, v, info):
 
         return v
 
-    @field_validator('batch_size')
+    @field_validator("batch_size")
     @classmethod
     def validate_batch_size(cls, v, info):
         """Validate batch size relative to number of texts."""
-        if info.data and 'texts' in info.data and info.data['texts']:
-            num_texts = len(info.data['texts'])
+        if info.data and "texts" in info.data and info.data["texts"]:
+            num_texts = len(info.data["texts"])
             if v > num_texts:
                 # Adjust batch size to not exceed number of texts
                 return num_texts
@@ -121,7 +120,7 @@ class RerankRequest(BaseModel):
         True, description="Whether to return the original passage texts", json_schema_extra={"example": True}
     )
 
-    @field_validator('query')
+    @field_validator("query")
     @classmethod
     def validate_query(cls, v):
         """Validate query text."""
@@ -133,7 +132,7 @@ def validate_query(cls, v):
 
         return v.strip()
 
-    @field_validator('passages')
+    @field_validator("passages")
     @classmethod
     def validate_passages(cls, v):
         """Validate passage inputs."""
@@ -147,17 +146,26 @@ def validate_passages(cls, v):
             if not passage.strip():
                 raise ValueError(f"Passage at index {i} cannot be empty or whitespace only")
 
-            if len(passage) > 4096:  # Reasonable character limit for passages
-                raise ValueError(f"Passage at index {i} too long: {len(passage)} > 4096 characters")
+            max_passage_length = int(os.getenv("MAX_PASSAGE_LENGTH", "4096"))
+
+        for i, passage in enumerate(v):
+            if not isinstance(passage, str):
+                raise ValueError(f"Passage at index {i} must be a string")
+
+            if not passage.strip():
+                raise ValueError(f"Passage at index {i} cannot be empty or whitespace only")
+
+            if len(passage) > max_passage_length:
+                raise ValueError(f"Passage at index {i} too long: {len(passage)} > {max_passage_length} characters")
 
         return v
 
-    @field_validator('top_k')
+    @field_validator("top_k")
     @classmethod
     def validate_top_k(cls, v, info):
         """Validate top_k relative to number of passages."""
-        if info.data and 'passages' in info.data and info.data['passages']:
-            num_passages = len(info.data['passages'])
+        if info.data and "passages" in info.data and info.data["passages"]:
+            num_passages = len(info.data["passages"])
             if v > num_passages:
                 # Adjust top_k to not exceed number of passages
                 return num_passages

pyproject.toml

@@ -36,6 +36,8 @@ dependencies = [
   "python-dotenv",
   "structlog",
   "prometheus-client",
+  "mlx>=0.29.2",
+  "psutil>=7.1.0",
 ]
 
 [project.optional-dependencies]