huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-11 04:14:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove pdb comments.

Browse Source
This commit is contained in:
Nicolas Patry 2023-12-04 14:43:29 +00:00
parent 79f9afba90
commit d99f281050
2 changed files with 0 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
server/text_generation_server/models/flash_causal_lm.py
View File

@ -825,8 +825,6 @@ class FlashCausalLM(Model):
next_token_logits = out next_token_logits = out
# if next_token_logits.shape[0] == 3:
# import ipdb;ipdb.set_trace()
from text_generation_server.models import SPECULATE from text_generation_server.models import SPECULATE
next_input_ids, next_token_logprobs, logprobs, accepted_ids, speculative_ids = batch.next_token_chooser( next_input_ids, next_token_logprobs, logprobs, accepted_ids, speculative_ids = batch.next_token_chooser(
batch.all_input_ids_tensor[:, : batch.max_seqlen], next_token_logits, SPECULATE, batch.speculative_ids, speculative_logits batch.all_input_ids_tensor[:, : batch.max_seqlen], next_token_logits, SPECULATE, batch.speculative_ids, speculative_logits
@ -850,7 +848,6 @@ class FlashCausalLM(Model):
batch.cu_seqlen_prefill = None batch.cu_seqlen_prefill = None
else: else:
prefill_logprobs = None prefill_logprobs = None
# import ipdb;ipdb.set_trace()
next_position_ids = batch.position_ids next_position_ids = batch.position_ids
# Cumulative length # Cumulative length
@ -912,15 +909,6 @@ class FlashCausalLM(Model):
cumulative_length += input_length cumulative_length += input_length
# if accepted_ids[0] > 1:
# import ipdb;ipdb.set_trace()
# if len(accepted_ids) > 1:
# raise Exception("Implemtent the batched behavior")
# Set values in batch
# batch.input_ids = torch.cat([next_input_ids.unsqueeze(-1), speculative_ids], dim=1).view(-1)
batch.input_ids = next_input_ids[accepted_ids.cumsum(dim=-1) - 1] batch.input_ids = next_input_ids[accepted_ids.cumsum(dim=-1) - 1]
batch.speculative_ids = speculative_ids batch.speculative_ids = speculative_ids
batch.position_ids = next_position_ids + accepted_ids batch.position_ids = next_position_ids + accepted_ids

2
server/text_generation_server/utils/tokens.py
View File

@ -308,8 +308,6 @@ class HeterogeneousNextTokenChooser:
break break
accepted_ids.append(accepted) accepted_ids.append(accepted)
from loguru import logger
logger.info(f"ACCEPTED IDS {accepted_ids}")
accepted_ids = torch.tensor(accepted_ids, device=input_ids.device, dtype=input_ids.dtype) accepted_ids = torch.tensor(accepted_ids, device=input_ids.device, dtype=input_ids.dtype)
next_ids = next_ids[indices] next_ids = next_ids[indices]
scores = scores[indices] scores = scores[indices]