fix: correct max_tokens and test file paths

Address Greptile review feedback:

1. Fix max_tokens: Set to 8192 (max_output_tokens) instead of 32768
   - Matches pattern of other featherless_ai models
   - Prevents users from requesting more tokens than model supports

2. Fix test file paths: Use absolute paths relative to test file
   - Prevents test failures when run from different directories
   - Matches pattern used by other model-price tests

3. Update test assertions to verify correct max_tokens value

model_prices_and_context_window.json

@@ -14724,14 +14724,14 @@
         "litellm_provider": "featherless_ai",
         "max_input_tokens": 32768,
         "max_output_tokens": 8192,
-        "max_tokens": 32768,
+        "max_tokens": 8192,
         "mode": "chat"
     },
     "featherless_ai/WeiboAI/VibeThinker-3B": {
         "litellm_provider": "featherless_ai",
         "max_input_tokens": 32768,
         "max_output_tokens": 8192,
-        "max_tokens": 32768,
+        "max_tokens": 8192,
         "mode": "chat"
     },
     "fireworks-ai-4.1b-to-16b": {

tests/test_litellm/test_vibethinker_models.py

@@ -2,12 +2,18 @@
 Test VibeThinker models configuration.
 """
 import json
+import os
 import pytest
 
 
 def test_vibethinker_models_in_model_prices():
     """Verify VibeThinker models are correctly configured in model_prices_and_context_window.json."""
-    with open("model_prices_and_context_window.json", "r") as f:
+    # Use absolute path relative to test file
+    model_prices_path = os.path.join(
+        os.path.dirname(os.path.dirname(os.path.dirname(__file__))),
+        "model_prices_and_context_window.json"
+    )
+    with open(model_prices_path, "r") as f:
         model_prices = json.load(f)
     
     models = [
@@ -27,14 +33,19 @@ def test_vibethinker_models_in_model_prices():
         assert "max_input_tokens" in config, f"{model}: missing max_input_tokens"
         assert "max_output_tokens" in config, f"{model}: missing max_output_tokens"
         
-        # Verify context window (Qwen base models have 32K context)
+        # Verify context window (Qwen base models have 32K input, 8K output)
         assert config["max_input_tokens"] == 32768, f"{model}: incorrect max_input_tokens"
-        assert config["max_tokens"] == 32768, f"{model}: incorrect max_tokens"
+        assert config["max_tokens"] == 8192, f"{model}: incorrect max_tokens (should match max_output_tokens)"
 
 
 def test_vibethinker_provider_consistency():
     """Verify VibeThinker models use the same provider pattern as other Featherless AI models."""
-    with open("model_prices_and_context_window.json", "r") as f:
+    # Use absolute path relative to test file
+    model_prices_path = os.path.join(
+        os.path.dirname(os.path.dirname(os.path.dirname(__file__))),
+        "model_prices_and_context_window.json"
+    )
+    with open(model_prices_path, "r") as f:
         model_prices = json.load(f)
     
     # Get all featherless_ai models

verify_uk_pii_proxy.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# Manual verification script for UK PII entity types in litellm proxy
+# Tests that UK entities are properly recognized and masked by Presidio
+
+set -e
+
+echo "=== UK PII Entity Types Verification Script ==="
+echo ""
+echo "This script verifies that UK_PASSPORT, UK_POSTCODE, and UK_VEHICLE_REGISTRATION"
+echo "entity types are properly recognized by the Presidio guardrail."
+echo ""
+
+CONFIG_FILE="/tmp/uk_pii_test_config.yaml"
+
+cat > "$CONFIG_FILE" <<'EOF'
+model_list:
+  - model_name: gpt-3.5-turbo
+    litellm_params:
+      model: gpt-3.5-turbo
+      api_key: os.environ/OPENAI_API_KEY
+
+guardrails:
+  - guardrail_name: "uk-pii-test"
+    litellm_params:
+      guardrail: presidio
+      mode: pre_call
+      default_on: true
+      pii_entities:
+        UK_NHS: MASK
+        UK_NINO: MASK
+        UK_PASSPORT: MASK
+        UK_POSTCODE: MASK
+        UK_VEHICLE_REGISTRATION: MASK
+EOF
+
+echo "Created test config at: $CONFIG_FILE"
+echo ""
+echo "Starting litellm proxy on port 4000 with UK PII guardrail enabled..."
+echo ""
+echo "Run the following commands in separate terminals to test:"
+echo ""
+echo "# Terminal 1: Start proxy"
+echo "python litellm/proxy/proxy_cli.py --config $CONFIG_FILE --detailed_debug"
+echo ""
+echo "# Terminal 2: Test UK_PASSPORT"
+echo "curl -X POST http://localhost:4000/chat/completions \\"
+echo "  -H 'Content-Type: application/json' \\"
+echo "  -d '{\"model\": \"gpt-3.5-turbo\", \"messages\": [{\"role\": \"user\", \"content\": \"My passport is 012345678\"}]}' | jq"
+echo ""
+echo "# Terminal 2: Test UK_POSTCODE"
+echo "curl -X POST http://localhost:4000/chat/completions \\"
+echo "  -H 'Content-Type: application/json' \\"
+echo "  -d '{\"model\": \"gpt-3.5-turbo\", \"messages\": [{\"role\": \"user\", \"content\": \"I live at SW1A 1AA\"}]}' | jq"
+echo ""
+echo "# Terminal 2: Test UK_VEHICLE_REGISTRATION"
+echo "curl -X POST http://localhost:4000/chat/completions \\"
+echo "  -H 'Content-Type: application/json' \\"
+echo "  -d '{\"model\": \"gpt-3.5-turbo\", \"messages\": [{\"role\": \"user\", \"content\": \"My car registration is AB12 CDE\"}]}' | jq"
+echo ""
+echo "Expected: The sensitive UK data should be masked (replaced with <UK_PASSPORT>, <UK_POSTCODE>, <UK_VEHICLE_REGISTRATION>)"
+echo "          in the request before being sent to OpenAI."