初始化项目，由ModelHub XC社区提供模型

Model: pathcosmos/frankenstallm
Source: Original Platform

source/eval/results/gemma3-4b/ollama_benchmark_results.json

@@ -0,0 +1,22 @@
+{
+  "metadata": {
+    "date": "2026-03-09 22:51:21",
+    "models": [
+      "gemma3:4b"
+    ],
+    "total_tests": 35,
+    "categories": [
+      "code",
+      "instruction_following",
+      "knowledge",
+      "korean_generation",
+      "korean_nlu",
+      "multilingual",
+      "reasoning",
+      "repetition_resistance",
+      "safety"
+    ]
+  },
+  "results": {},
+  "summary": {}
+}

source/eval/results/gemma3-4b/ollama_benchmark_summary.md

@@ -0,0 +1,29 @@
+# FRANKENSTALLM Ollama Benchmark Results
+
+- **Date**: 2026-03-09 22:51:21
+- **Models**: 
+- **Total test cases**: 35
+
+## Overall Auto-Scored Average
+
+| Model | Auto Avg |
+|-------|----------|
+
+## Auto-Scored Results by Category
+
+| Category |  |
+|----------||
+
+## Latency Comparison
+
+| Model | Avg TTFT (ms) | P50 TTFT | P95 TTFT | Avg TPS | P50 TPS | P95 TPS |
+|-------|--------------|----------|----------|---------|---------|---------|
+
+## Repetition Analysis Detail
+
+| Model | Test ID | Rep Rate | Unique/Total N-grams | Score |
+|-------|---------|----------|---------------------|-------|
+
+## Manual Review Needed
+
+The following prompts require human evaluation: