🌐 AI搜索 & 代理 主页
Skip to content

Working streaming tokenizer #1210

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
SilasMarvin merged 2 commits into from
Dec 1, 2023
Merged

Working streaming tokenizer #1210

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a1dda88
General clean ups
SilasMarvin Dec 1, 2023
2f26eda
Made the text iterator streamer return spaces correctly
SilasMarvin Dec 1, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Made the text iterator streamer return spaces correctly
  • Loading branch information
@SilasMarvin
SilasMarvin committed Dec 1, 2023
commit 2f26eda585ee3a22b7874d06415ca60795efd2d0
41 changes: 33 additions & 8 deletions pgml-extension/src/bindings/transformers/transformers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,19 +104,42 @@ def __init__(self, tokenizer, skip_prompt=False, timeout=None, **decode_kwargs):
self.next_tokens_are_prompt = True
self.stop_signal = None
self.text_queue = queue.Queue()
self.token_cache = []
self.text_index_cache = []

def put(self, value):
def put(self, values):
if self.skip_prompt and self.next_tokens_are_prompt:
self.next_tokens_are_prompt = False
return
# Can't batch this decode
decoded_values = []
for v in value:
decoded_values.append(self.tokenizer.decode(v, **self.decode_kwargs))
self.text_queue.put(decoded_values, self.timeout)
output = []
for i, v in enumerate(values):
if len(self.token_cache) <= i:
self.token_cache.append([])
self.text_index_cache.append(0)
token = v.tolist() # Returns a list or number
if type(token) == list:
self.token_cache[i].extend(token)
else:
self.token_cache[i].append(token)
text = self.tokenizer.decode(self.token_cache[i], **self.decode_kwargs)
if text.endswith("\n"):
output.append(text[self.text_index_cache[i] :])
self.token_cache[i] = []
self.text_index_cache[i] = 0
else:
printable_text = text[self.text_index_cache[i] : text.rfind(" ") + 1]
self.text_index_cache[i] += len(printable_text)
output.append(printable_text)
if any(output):
self.text_queue.put(output, self.timeout)

def end(self):
self.next_tokens_are_prompt = True
output = []
for i, tokens in enumerate(self.token_cache):
text = self.tokenizer.decode(tokens, **self.decode_kwargs)
output.append(text[self.text_index_cache[i] :])
self.text_queue.put(output, self.timeout)
self.text_queue.put(self.stop_signal, self.timeout)

def __iter__(self):
Expand All @@ -127,6 +150,7 @@ def __next__(self):
if value != self.stop_signal:
return value


class GGMLPipeline(object):
def __init__(self, model_name, **task):
import ctransformers
Expand Down Expand Up @@ -245,7 +269,8 @@ def stream(self, input, **kwargs):
generation_kwargs = None
if self.task == "conversational":
streamer = TextIteratorStreamer(
self.tokenizer, skip_prompt=True, skip_special_tokens=True
self.tokenizer,
skip_prompt=True,
)
if "chat_template" in kwargs:
input = self.tokenizer.apply_chat_template(
Expand All @@ -261,7 +286,7 @@ def stream(self, input, **kwargs):
input = self.tokenizer(input, return_tensors="pt").to(self.model.device)
generation_kwargs = dict(input, streamer=streamer, **kwargs)
else:
streamer = TextIteratorStreamer(self.tokenizer, skip_special_tokens=True)
streamer = TextIteratorStreamer(self.tokenizer)
input = self.tokenizer(input, return_tensors="pt", padding=True).to(
self.model.device
)
Expand Down