O gráfico aparece a partir do segundo dia de coleta.
✓ Llama 3.1 8B Instruct (4.9 GB) ✓ Llama 3.2 3B Instruct (2 GB) ✗ Llama 3.3 70B Instruct (42.5 GB) ✓ Qwen3 4B (2.5 GB) ✓ Qwen3 8B (5 GB) ✓ Qwen3 14B (9 GB) ✗ Qwen3 32B (19.8 GB) ✗ Qwen3 30B-A3B (18.6 GB) ✓ Qwen2.5 Coder 14B (9 GB) ✗ Qwen2.5 Coder 32B (19.9 GB) ✓ Gemma 3 4B (2.5 GB) ✓ Gemma 3 12B (7.3 GB) ✗ Gemma 3 27B (16.5 GB) ✓ Mistral Nemo 12B (7.5 GB) ✓ Mistral Small 3.2 24B (14.3 GB) ✓ DeepSeek R1 Distill Qwen 7B (4.7 GB) ✓ DeepSeek R1 Distill Qwen 14B (9 GB) ✗ DeepSeek R1 Distill Qwen 32B (19.9 GB) ✗ DeepSeek R1 Distill Llama 70B (42.5 GB) ✓ gpt-oss 20B (11.6 GB) ✗ gpt-oss 120B (62.8 GB) ✓ Phi-4 14B (9.1 GB)
✓ Llama 3.1 8B Instruct (8.5 GB) ✓ Llama 3.2 3B Instruct (3.4 GB) ✗ Llama 3.3 70B Instruct (75 GB) ✓ Qwen3 4B (4.3 GB) ✓ Qwen3 8B (8.7 GB) ✗ Qwen3 14B (15.7 GB) ✗ Qwen3 32B (34.8 GB) ✗ Qwen3 30B-A3B (32.5 GB) ✗ Qwen2.5 Coder 14B (15.7 GB) ✗ Qwen2.5 Coder 32B (34.8 GB) ✓ Gemma 3 4B (4.1 GB) ✓ Gemma 3 12B (12.5 GB) ✗ Gemma 3 27B (28.7 GB) ✓ Mistral Nemo 12B (13 GB) ✗ Mistral Small 3.2 24B (25.1 GB) ✓ DeepSeek R1 Distill Qwen 7B (8.1 GB) ✗ DeepSeek R1 Distill Qwen 14B (15.7 GB) ✗ DeepSeek R1 Distill Qwen 32B (34.8 GB) ✗ DeepSeek R1 Distill Llama 70B (75 GB) ✓ gpt-oss 20B (12.1 GB) ✗ gpt-oss 120B (63.4 GB) ✗ Phi-4 14B (15.6 GB)
Modelos maiores que a VRAM ainda rodam dividindo com a RAM (offload), só que bem mais devagar.