Text Generation
GGUF
PyTorch
English
nvidia
conversational

πŸš€ Starting Agent Loop Tool Efficiency Test

#1
by bukit - opened

πŸ“Š Configuration:
Base URL: http://localhost:9099/v1
Model: NVIDIA-Nemotron-3-Nano-4B-UD-Q8_K_XL
Test Cases: 17
Output: results/agent_test_results_NVIDIA-Nemotron-3-Nano-4B-UD-Q8_K_XL_20260319_011353.json
Log File: logs/agent_test_logs_NVIDIA-Nemotron-3-Nano-4B-UD-Q8_K_XL_20260319_011353.log

πŸ”„ Running agent tests...
Starting agent test suite with 17 test cases
Running agent test: zero_weather_question
Running agent test: zero_greeting
Running agent test: zero_general_question
Running agent test: zero_capabilities
Running agent test: simple_search_electronics
Running agent test: simple_add_iphone
Running agent test: simple_view_cart
Running agent test: simple_remove_product
Running agent test: simple_checkout
Running agent test: medium_search_and_add
Running agent test: medium_search_category_and_add
Running agent test: medium_remove_and_add
Running agent test: medium_view_and_add
Running agent test: complex_shopping_workflow
Running agent test: complex_cart_management
Running agent test: complex_gift_shopping
Running agent test: zero_thank_you
βœ… Tests completed in 1m9.120009s

πŸ“ˆ Agent Test Results

Total Tests: 17
βœ… Passed: 15
❌ Failed: 2
⏱️ Total LLM Time: 11m59.7020832s
⏱️ Average Time per Request: 17.553709346s

πŸ“‹ Test Case Results:

Test Case: zero_greeting
Status: βœ… PASSED
Matched Path: no_tools
Response Time: 15.4117354s
Tool Calls: 0

Test Case: simple_checkout
Status: βœ… PASSED
Matched Path: direct_checkout
Response Time: 21.2737574s
Tool Calls: 1
Tools Used: checkout

Test Case: simple_add_iphone
Status: βœ… PASSED
Matched Path: direct_add
Response Time: 22.0804369s
Tool Calls: 1
Tools Used: add_to_cart

Test Case: zero_general_question
Status: βœ… PASSED
Matched Path: no_tools
Response Time: 24.1708517s
Tool Calls: 0

Test Case: simple_view_cart
Status: βœ… PASSED
Matched Path: view_cart
Response Time: 27.6753334s
Tool Calls: 1
Tools Used: view_cart

Test Case: simple_remove_product
Status: βœ… PASSED
Matched Path: direct_remove
Response Time: 30.6650127s
Tool Calls: 1
Tools Used: remove_from_cart

Test Case: medium_search_category_and_add
Status: βœ… PASSED
Matched Path: search_then_add
Response Time: 34.8005152s
Tool Calls: 2
Tools Used: search_products, add_to_cart

Test Case: simple_search_electronics
Status: βœ… PASSED
Matched Path: search_by_category
Response Time: 37.5446965s
Tool Calls: 1
Tools Used: search_products

Test Case: zero_capabilities
Status: βœ… PASSED
Matched Path: no_tools
Response Time: 43.082385s
Tool Calls: 0

Test Case: medium_remove_and_add
Status: βœ… PASSED
Matched Path: remove_then_add
Response Time: 43.5885379s
Tool Calls: 2
Tools Used: remove_from_cart, add_to_cart

Test Case: zero_weather_question
Status: βœ… PASSED
Matched Path: no_tools
Response Time: 51.0073694s
Tool Calls: 0

Test Case: zero_thank_you
Status: βœ… PASSED
Matched Path: no_tools
Response Time: 53.7771323s
Tool Calls: 0

Test Case: complex_gift_shopping
Status: ❌ FAILED
Response Time: 57.4760189s
Tool Calls: 5
Tools Used: search_products, search_products, add_to_cart, add_to_cart, view_cart

Test Case: complex_shopping_workflow
Status: ❌ FAILED
Response Time: 1m0.7459407s
Tool Calls: 5
Tools Used: search_products, add_to_cart, add_to_cart, view_cart, checkout

Test Case: medium_view_and_add
Status: βœ… PASSED
Matched Path: view_then_add
Response Time: 1m1.6214938s
Tool Calls: 2
Tools Used: view_cart, add_to_cart

Test Case: medium_search_and_add
Status: βœ… PASSED
Matched Path: search_by_query
Response Time: 1m5.8858487s
Tool Calls: 2
Tools Used: search_products, add_to_cart

Test Case: complex_cart_management
Status: βœ… PASSED
Matched Path: cart_organization
Response Time: 1m9.118951s
Tool Calls: 3
Tools Used: view_cart, remove_from_cart, add_to_cart

❌ Failed Tests Details:

Test Case: complex_gift_shopping
Expected Tool Variants: 2
Variant 1 (gift_shopping_workflow): 5 tools
Variant 2 (gift_shopping_workflow): 5 tools
Actual Tool Calls: 5
1. search_products
2. search_products
3. add_to_cart
4. add_to_cart
5. view_cart
Response Time: 57.4760189s

Test Case: complex_shopping_workflow
Expected Tool Variants: 4
Variant 1 (full_workflow_with_iphone): 4 tools
Variant 2 (full_workflow_with_headphones): 4 tools
Variant 3 (full_workflow_with_headphones_and_iphone): 5 tools
Variant 4 (full_workflow_with_iphone_and_headphones): 5 tools
Actual Tool Calls: 5
1. search_products
2. add_to_cart
3. add_to_cart
4. view_cart
5. checkout
Response Time: 1m0.7459407s

πŸ“Š Overall Success Rate: 88.24%

llama-server --port 9099 -ngl 99 -fa on -c 16000 --temp 0.1 --jinja -m X:\path\to\NVIDIA-Nemotron-3-Nano-4B-UD-Q8_K_XL.gguf

49 tokens/s on RTX 3060 12GB // https://github.com/docker/model-test/

A very fast and accurate small reasoning model for function calling.

Sign up or log in to comment