Make compatible with recent versions of triton

Many people are complaining about not being able to run this model with the latest versions of triton, e.g. [here](https://github.com/Zhihan1996/DNABERT_2/issues/19#issuecomment-1641944919). This patch solves this issue.

flash_attn_triton.py

@@ -188,7 +188,7 @@ def _fwd_kernel(
                             (offs_d[None, :] < headdim),
                             other=0.0)
         qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32)
-        qk += tl.dot(q, k, trans_b=True)
+        qk += tl.dot(q, tl.trans(k))
         # Trying to combine the two masks seem to make the result wrong
         if not EVEN_N:  # Need to mask out otherwise the softmax is wrong
             qk += tl.where((start_n + offs_n)[None, :] < seqlen_k, 0,
@@ -431,7 +431,7 @@ def _bwd_kernel_one_col_block(
                             (offs_d[None, :] < headdim),
                             other=0.0)
         # recompute p = softmax(qk, dim=-1).T
-        qk = tl.dot(q, k, trans_b=True)
+        qk = tl.dot(q, tl.trans(k))
         # Trying to combine the two masks seem to make the result wrong
         if not EVEN_N:  # Need to mask out otherwise the softmax is wrong
             qk = tl.where(offs_n[None, :] < seqlen_k, qk, float('-inf'))