Keep on to blur preview images; turn off to show them clearly
nice that it's open weight, but comparing dense vs moe models and only looking at total params is pretty unfair, if you look at active params instead of total params it's a different story: - GLM 4.6 (32B): 74% fewer - Minimax M2 (10B): 92% fewer - K2 thinking (32B): 74% fewer - V3.2 (37B): 70% fewer size (both total or active!) is not the right metric here, we should have the same graph with speed on vllm / sglang
Training llm's (now: @huggingface) anon feedback: https://t.co/JmMh7Sfvxd
nice that it's open weight, but comparing dense vs moe models and only looking at total params is pretty unfair, if you look at active params instead of total params it's a different story: - GLM 4.6 (32B): 74% fewer - Minimax M2 (10B): 92% fewer - K2 thinking (32B): 74% fewer - V3.2 (37B): 70% fewer size (both total or active!) is not the right metric here, we should have the same graph with speed on vllm / sglang
Training llm's (now: @huggingface) anon feedback: https://t.co/JmMh7Sfvxd
nice that it's open weight, but comparing dense vs moe models and only looking at total params is pretty unfair, if you look at active params instead of total params it's a different story: - GLM 4.6 (32B): 74% fewer - Minimax M2 (10B): 92% fewer - K2 thinking (32B): 74% fewer - V3.2 (37B): 70% fewer size (both total or active!) is not the right metric here, we should have the same graph with speed on vllm / sglang
Training llm's (now: @huggingface) anon feedback: https://t.co/JmMh7Sfvxd