↓ 3 callersFunctionprepare_cumsum(query, key, value, g_key, g_value, normalizer_gk=8, normalizer_gv=8, clamp_min=-3)
based/models/triton/kernels/inter_chunk_contribution/preprocess_cumsum_gk.py:260
↓ 3 callersFunctionupdate_graph_cache(
model,
cache,
batch_size,
seqlen_og,
max_seqlen,
decoding_seqlens=(1,),
tensor_p
train/src/generation.py:698
↓ 2 callersMethod__init__(self, config: GPT2Config, process_group=None, device=None, dtype=None, multiple_of=256)
based/models/transformer/gpt.py:384
↓ 2 callersMethod_apply_rotary_update_kvcache_attention Fast path that combine 3 steps: apply rotary to Q and K, update kv cache, and apply attention. q: (batch_size, seqlen_q, nheads, head
based/models/mixers/mha.py:512
↓ 2 callersMethod_apply_rotary_update_kvcache_attention Fast path that combine 3 steps: apply rotary to Q and K, update kv cache, and apply attention. q: (batch_size, seqlen_q, nheads, head
based/models/mixers/mha.py:865
↓ 2 callersMethod_apply_rotary_update_kvcache_attention Fast path that combine 3 steps: apply rotary to Q and K, update kv cache, and apply attention. q: (batch_size, seqlen_q, nheads, head
based/models/mixers/slide_attention.py:390
↓ 2 callersFunction_layer_norm_bwd(
dy,
x,
weight,
bias,
eps,
mean,
rstd,
dresidual=None,
dy1=None,
weig
based/ops/triton/layer_norm.py:560
↓ 2 callersFunction_layer_norm_bwd(
dy,
x,
weight,
bias,
eps,
mean,
rstd,
dresidual=None,
has_residual=False
based/models/mixers/mamba/ops/triton/layernorm.py:293
↓ 2 callersFunction_layer_norm_bwd(
dy,
x,
weight,
bias,
eps,
mean,
rstd,
dresidual=None,
has_residual=False
synthetic/zoology/mixers/mamba_ssm/triton/layernorm.py:293
↓ 2 callersFunction_layer_norm_bwd(
dy,
x,
weight,
bias,
eps,
mean,
rstd,
dresidual=None,
dy1=None,
weig
train/src/ops/layer_norm.py:560
↓ 2 callersFunction_layer_norm_fwd(
x,
weight,
bias,
eps,
residual=None,
x1=None,
weight1=None,
bias1=None,
based/ops/triton/layer_norm.py:255
↓ 2 callersFunction_layer_norm_fwd(
x, weight, bias, eps, residual=None, out_dtype=None, residual_dtype=None, is_rms_norm=False
)
based/models/mixers/mamba/ops/triton/layernorm.py:123
↓ 2 callersFunction_layer_norm_fwd(
x, weight, bias, eps, residual=None, out_dtype=None, residual_dtype=None, is_rms_norm=False
)
synthetic/zoology/mixers/mamba_ssm/triton/layernorm.py:123
↓ 2 callersFunction_layer_norm_fwd(
x,
weight,
bias,
eps,
residual=None,
x1=None,
weight1=None,
bias1=None,
train/src/ops/layer_norm.py:255
↓ 2 callersMethod_loglikelihood_tokens(
self,
requests: List[Tuple[Tuple[str, str], List[int], List[int]]],
disable_tqdm: bo
lm-eval-harness/lm_eval/models/vllm_causallms.py:342
↓ 2 callersMethod_loglikelihood_tokens(
self,
requests: List[Tuple[Tuple[str, str], List[int], List[int]]],
disable_tqdm: bo
lm-eval-harness/lm_eval/models/huggingface.py:952
↓ 2 callersFunction_update_kv_cachekv: (batch_size, seqlen, 2, nheads, head_dim) or (batch_size, 1, 2, nheads, head_dim)
based/models/mixers/mha.py:347