Getting triton compiler error while running the inference code mentioned in the README.

[sahiljoshi515]

triton.compiler.errors.CompilationError: at 114:14:
else:
if EVEN_HEADDIM:
k = tl.load(k_ptrs + start_n * stride_kn,
mask=(start_n + offs_n)[:, None] < seqlen_k,
other=0.0)
else:
k = tl.load(k_ptrs + start_n * stride_kn,
mask=((start_n + offs_n)[:, None] < seqlen_k) &
(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)

Hi, I am not sure why I get this, I am simply running the code below:

`import torch
from transformers import AutoTokenizer, AutoModel
from transformers.models.bert.configuration_bert import BertConfig

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

config = BertConfig.from_pretrained("zhihan1996/DNABERT-2-117M")
tokenizer = AutoTokenizer.from_pretrained("zhihan1996/DNABERT-2-117M", trust_remote_code=True, config=config)
model = AutoModel.from_pretrained("zhihan1996/DNABERT-2-117M", trust_remote_code=True, config=config)
model.to(device)
model.eval()

dna = "ACGTAGCATCGGATCTATCTATCGACACTTGGTTATCGATCTACGAGCATCTCGTTAGC"
inputs = tokenizer(dna, return_tensors = 'pt')["input_ids"].to(device)
hidden_states = model(inputs)[0] # [1, sequence_length, 768]

embedding with mean pooling

embedding_mean = torch.mean(hidden_states[0], dim=0)
print(embedding_mean.shape) # expect to be 768

embedding with max pooling

embedding_max = torch.max(hidden_states[0], dim=0)[0]
print(embedding_max.shape) # expect to be 768
`

If anybody has solved this please let me know. Thank you for your awesome work!