447444fd06594e531ae1141afac78051481e4468,catalyst/rl/offpolicy/algorithms/td3.py,TD3,_base_loss,#TD3#Any#Any#Any#Any#Any#,107

Before Change

            for x in self.target_critics
        ], dim=-1)
        // B x num_heads x 1
        q_values_tp1 = q_values_tp1.min(dim=-1, keepdim=True)[0].detach()

        // B x num_heads x 1
        q_target_t = rewards_t + (1 - done_t) * gammas * q_values_tp1

After Change

        // {num_critics} * [bs; num_heads; 1] -> concat
        // [bs; num_heads; num_critics] -> min over all critics
        // [bs; num_heads; 1]
        q_values_tp1 = (
            torch.cat(q_values_tp1, dim=-1)
            .min(dim=-1, keepdim=True)[0]
        )

        // [bs; num_heads; 1] -> many-heads view transform
        // [{bs * num_heads}; 1]
        q_target_t = (

Instances