Update README.md

README.md

@@ -14,7 +14,7 @@ license: apache-2.0
 # DeepSignal-4B-V1 (GGUF)
 
 This repository provides a GGUF model file for local inference (e.g., `llama.cpp` / LM Studio). It is intended for traffic-signal-control analysis and related text-generation workflows.
-For details, check our repository at [`AIMSLaboratory/DeepSignal`](https://github.com/AIMSLaboratory/DeepSignal/settings).
+For details, check our repository at [`AIMSLaboratory/DeepSignal`](https://github.com/AIMSLaboratory/DeepSignal).
 
 
 ## Files
@@ -35,28 +35,18 @@ where the number is the phase index (starting from 0) and the seconds is the dur
 
 ## Evaluation (Traffic Simulation)
 
-### Performance Metrics Comparison by Model
-
-| Model | Avg Saturation | Avg Queue Length | Max Saturation | Max Queue Length | Avg Congestion Index |
-|---|---:|---:|---:|---:|---:|
-| [`Qwen3-30B-A3B`](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct) | 0.1550 | 5.5000 | 0.1550 | 5.4995 | 0.1500 |
-| DeepSignal-4B (Ours) | 0.1580 | 5.5500 | 0.1580 | 5.5498 | 0.1550 |
-| [`LightGPT-8B-Llama3`](https://huggingface.co/lightgpt/LightGPT-8B-Llama3) | 0.1720 | 6.1000 | 0.1720 | 6.1000 | 0.1950 |
-| SFT | 0.1780 | 6.2500 | 0.1780 | 6.2500 | 0.2050 |
-| Last Round GRPO | 0.1850 | 6.4500 | 0.1850 | 6.4500 | 0.2150 |
-| [`Qwen3-4B`](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507) | 0.1980 | 7.2000 | 0.1980 | 7.1989 | 0.2450 |
-| Max Pressure | 0.2050 | 7.8000 | 0.2049 | 7.7968 | 0.2550 |
-| [`GPT-OSS-20B`](https://huggingface.co/openai/gpt-oss-20b) | 0.2250 | 8.5001 | 0.2250 | 8.4933 | 0.3050 |
-
-### Congestion Level Distribution by Model (%)
-
-| Model | Light congestion | Smooth | Very smooth |
-|---|---:|---:|---:|
-| DeepSignal-4B (Ours) | 0.00 | 12.00 | 88.00 |
-| [`GPT-OSS-20B`](https://huggingface.co/openai/gpt-oss-20b) | 2.00 | 53.33 | 44.67 |
-| [`LightGPT-8B-Llama3`](https://huggingface.co/lightgpt/LightGPT-8B-Llama3) | 0.00 | 21.00 | 79.00 |
-| Max Pressure | 0.00 | 36.44 | 63.56 |
-| [`Qwen3-30B-A3B`](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct) | 0.00 | 10.00 | 90.00 |
-| [`Qwen3-4B`](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507) | 2.33 | 32.00 | 65.67 |
-| Qwen3-4B-SFT | 0.00 | 23.33 | 76.67 |
+### Performance Metrics Comparison by Model $^{*}$
+
+| Model | Avg Saturation | Avg Queue Length (veh/s) | Avg Throughput (veh/5min) | Avg Response Time (s) |
+|:---:|:---:|:---:|:---:|:---:|
+| [`GPT-OSS-20B (thinking)`](https://huggingface.co/openai/gpt-oss-20b) | 0.3801 | 0.476210 | 77.910075 | 6.768 |
+| **DeepSignal-4B (Ours)** | 0.4219 | 0.498338 | 79.883430 | 2.131 |
+| [`Qwen3-30B-A3B`](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct) | 0.4314 | 0.580256 | 79.059117 | 2.727 |
+| [`Qwen3-4B`](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507) | 0.4655 | 2.453933 | 75.711907 | 1.994 |
+| Max Pressure | 0.4647 | 0.639584 | 77.235637 | ** |
+| [`LightGPT-8B-Llama3`](https://huggingface.co/lightgpt/LightGPT-8B-Llama3) | 0.5230 | 1.258782 | 75.512073 | 3.025*** |
+
+`*`: Each simulation scenario runs for 60 minutes. We discard the first **5 minutes** as warm-up, then compute metrics over the next **20 minutes** (minute 5 to 25). We cap the evaluation window because, when an LLM controls signal timing for only a single intersection, spillback from neighboring intersections may occur after ~20+ minutes and destabilize the scenario. All evaluations are conducted on a **Mac Studio M3 Ultra**.  
+`**`: Max Pressure is a fixed signal-timing optimization algorithm (not an LLM), so we omit its Avg Response Time; this metric is only defined for LLM-based signal-timing optimization.  
+`***`: For LightGPT-8B-Llama3, Avg Response Time is computed using only the successful responses.