Update InternVL3_5 1B Q8 vs InternVL3_5 2B Q4.md

InternVL3_5 1B Q8 vs InternVL3_5 2B Q4.md

@@ -7,8 +7,8 @@ llama-server --host 0.0.0.0 --port 8000 --no-mmap -c 32768 -ub 4096 --temp 0.6 -
 
 | Model | Parameters | Quantization | MMLU-PRO (Overall) | GPQA (Accuracy) | GPQA (Refusal Fraction) |
 | :--- | :--- | :--- | :--- | :--- | :--- |
-| **InternVL3_5** | 1B | Q8_0 | 28.11% | 14.51% | 49.33% |
-| **InternVL3_5** | 2B | Q4_K_M | 38.34% | 15.85% | 42.86% |
+| **InternVL3_5** | 1B (0.6B for text + 0.3 for vision) | Q8_0 | 28.11% | 14.51% | 49.33% |
+| **InternVL3_5** | 2B (1.7B for text + 0.3 for vision) | Q4_K_M | 38.34% | 15.85% | 42.86% |
 
 ### MMLU-PRO by Subject