huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-04-26 12:32:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

cleanup comment

Browse Source
This commit is contained in:
Mohit Sharma 2025-03-21 11:28:40 +00:00
parent ac6fc70c75
commit b41faae318
2 changed files with 1 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
server/text_generation_server/models/globals.py
View File

@ -6,7 +6,6 @@ from typing import Dict, Optional
from text_generation_server.utils.log import log_master
REQUEST_LOGPROBS = os.getenv("REQUEST_LOGPROBS", "0").lower() in {"1", "true"}
ATTENTION = os.environ["ATTENTION"]
# default_prefix_caching = "1" if ATTENTION in {"flashinfer", "flashdecoding"} else "0"
PREFIX_CACHING = os.environ["PREFIX_CACHING"].lower() in {

6
server/text_generation_server/models/transformers_flash_causal_lm.py
View File

@ -36,12 +36,11 @@ def tgi_flash_attention_forward(
softcap: Optional[float] = None,
**kwargs, # This is needed to "absorb" other args passed by Transformers modeling
):
# from pdb import set_trace; set_trace()
kv_cache = kv_cache[module.layer_idx]
query_states = query_states.transpose(1, 2).squeeze(dim=0)
key_states = key_states.transpose(1, 2).squeeze(dim=0)
value_states = value_states.transpose(1, 2).squeeze(dim=0)
# from pdb import set_trace; set_trace()
# Take care of updating the cache in-place
kv_cache.store(key=key_states, value=value_states, slots=slots, kv_scales=kv_scales)
@ -49,7 +48,6 @@ def tgi_flash_attention_forward(
_, num_heads, head_dim = query_states.shape
softmax_scale = 1 / math.sqrt(head_dim) if softmax_scale is None else softmax_scale
sliding_window = -1 if sliding_window is None else sliding_window
# from pdb import set_trace; set_trace()
if cu_seqlen_prefill is not None:
attn_output = attention(
@ -108,7 +106,6 @@ class TransformersFlashCausalLM(FlashCausalLM):
tokenizer_class=AutoTokenizer,
kv_cache_dtype: Optional[torch.dtype] = None,
):
# # from pdb import set_trace; set_trace()
self.quantize = quantize
self.process_group, rank, world_size = initialize_torch_distributed()
@ -266,7 +263,6 @@ class TransformersFlashCausalLM(FlashCausalLM):
prefill_cache_indices=None, # not used, but passed to match original signature
adapter_data=None, # not supported, but passed to match original signature
):
# from pdb import set_trace; set_trace()
# A value of `None` (i.e. no logit slicing) translates to `0` in Transformers
logits_to_keep = lm_head_indices if lm_head_indices is not None else 0