Add model-index with instruction benchmark evaluations

Added structured evaluation results from README benchmark table:

**Instruction Model Benchmarks (No Extended Thinking):**
- AIME 2025 (High school math): 9.3
- GSM-Plus (Math problem-solving): 72.8
- LiveCodeBench v4 (Competitive programming): 15.2
- GPQA Diamond (Graduate-level reasoning): 35.7
- IFEval (Instruction following): 76.7
- MixEval Hard (Alignment): 26.9
- BFCL (Tool Calling): 92.3
- Global MMLU (Multilingual Q&A): 53.5

Total: 8 benchmarks covering reasoning, math, coding, instruction-following, alignment, tool use, and multilingual capabilities.

This enables the model to appear in leaderboards and makes it easier to compare with other models.

Note: This PR adds benchmark metadata to the model card frontmatter and should not conflict with existing PRs #43, #32, and #16 which only modify the chat template.

README.md

@@ -11,7 +11,43 @@ language:
 - ar
 - ru
 base_model:
-  - HuggingFaceTB/SmolLM3-3B-Base
+- HuggingFaceTB/SmolLM3-3B-Base
+model-index:
+- name: SmolLM3-3B
+  results:
+  - task:
+      type: text-generation
+    dataset:
+      name: Instruction Benchmarks
+      type: benchmark
+    metrics:
+    - name: AIME 2025
+      type: aime_2025
+      value: 9.3
+    - name: GSM-Plus
+      type: gsm_plus
+      value: 72.8
+    - name: LiveCodeBench v4
+      type: live_code_bench_v4
+      value: 15.2
+    - name: GPQA Diamond
+      type: gpqa_diamond
+      value: 35.7
+    - name: IFEval
+      type: ifeval
+      value: 76.7
+    - name: MixEval Hard
+      type: mixeval_hard
+      value: 26.9
+    - name: BFCL
+      type: bfcl
+      value: 92.3
+    - name: Global MMLU
+      type: global_mmlu
+      value: 53.5
+    source:
+      name: Model README - Instruction Benchmarks
+      url: https://huggingface.co/HuggingFaceTB/SmolLM3-3B
 ---