Skip to content
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

Add stream_options support according to OpenAI API #1552

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Add stream_options support according to OpenAI API #1552

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
aa397ee
Adding stream_options and include_usage to server. Extracting Updatin…
tpfau Oct 1, 2024
cbaaea3
fix linting
tpfau Oct 15, 2024
ffc47e5
Clearer if clauses
tpfau Oct 17, 2024
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
220 changes: 142 additions & 78 deletions llama_cpp/llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1116,6 +1116,56 @@ def decode_batch(seq_sizes: List[int]):
else:
return output

def _create_chunk(
self,
completion_id: str,
created: int,
model_name: str,
text: str,
logprobs_or_none: Union[Optional[CompletionLogprobs], None],
include_usage: bool,
index: int,
finish_reason: Union[str, None],
usage: Union[Dict[str, Any], None] = None,
) -> CreateChatCompletionStreamResponse:
"""
Create chunks for streaming API, depending on whether usage is requested or
not they need (or don't need) an additional field
"""

if include_usage:
token = {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": text,
"index": index,
"logprobs": logprobs_or_none,
"finish_reason": finish_reason,
},
],
"usage": usage,
}
else:
token = {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": text,
"index": index,
"logprobs": logprobs_or_none,
"finish_reason": finish_reason,
}
],
}
return token

def _create_completion(
self,
prompt: Union[str, List[int]],
Expand All @@ -1133,6 +1183,7 @@ def _create_completion(
repeat_penalty: float = 1.0,
top_k: int = 40,
stream: bool = False,
stream_options: Optional[StreamOptions] = None,
seed: Optional[int] = None,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
Expand Down Expand Up @@ -1363,6 +1414,11 @@ def logit_bias_processor(
break

if stream:
if stream_options is not None and "include_usage" in stream_options:
include_usage = True if stream_options["include_usage"] else False
else:
include_usage = False
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of the double nested if block, presumably you could do: if stream and stream_options and "include_usage" in stream_options:? @tpfau

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No you can't since all the following code has to be run in both instances, and only the include stream usage needs to be adapted based on the props.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ahh, of course. I guess I was meant something like this:

if stream_options and "include_usage" in stream_options:
    include_usage = stream_options["include_usage"]
else:
    include_usage = False

It relies on a None stream_options being falsey, and stream_options.include_usage being a boolean.
Just a little more readable.


remaining_tokens = completion_tokens[returned_tokens:]
remaining_text = self.detokenize(
remaining_tokens,
Expand Down Expand Up @@ -1442,24 +1498,23 @@ def logit_bias_processor(
"top_logprobs": [top_logprob],
}
returned_tokens += 1
yield {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": self.detokenize(
[token],
prev_tokens=prompt_tokens
+ completion_tokens[:returned_tokens],
).decode("utf-8", errors="ignore"),
"index": 0,
"logprobs": logprobs_or_none,
"finish_reason": None,
}
],
}
text = (
self.detokenize(
[token],
prev_tokens=prompt_tokens
+ completion_tokens[:returned_tokens],
).decode("utf-8", errors="ignore"),
)
yield self._create_chunk(
completion_id=completion_id,
created=created,
model_name=model_name,
text=text,
finish_reason=None,
index=0,
logprobs_or_none=logprobs_or_none,
include_usage=include_usage,
)
else:
while len(remaining_tokens) > 0:
decode_success = False
Expand Down Expand Up @@ -1488,20 +1543,16 @@ def logit_bias_processor(
remaining_tokens = remaining_tokens[i:]
returned_tokens += i

yield {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": ts,
"index": 0,
"logprobs": None,
"finish_reason": None,
}
],
}
yield self._create_chunk(
index=0,
finish_reason=None,
completion_id=completion_id,
created=created,
model_name=model_name,
text=ts,
logprobs_or_none=None,
include_usage=include_usage,
)

if len(completion_tokens) >= max_tokens:
text = self.detokenize(completion_tokens, prev_tokens=prompt_tokens)
Expand Down Expand Up @@ -1580,54 +1631,60 @@ def logit_bias_processor(
if token_end_position == end - 1:
break
returned_tokens += 1
yield {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": last_text[
: len(last_text) - (token_end_position - end)
].decode("utf-8", errors="ignore"),
"index": 0,
"logprobs": logprobs_or_none,
"finish_reason": None,
}
],
}
text = last_text[
: len(last_text) - (token_end_position - end)
].decode("utf-8", errors="ignore")

yield self._create_chunk(
completion_id=completion_id,
created=created,
model_name=model_name,
text=text,
logprobs_or_none=logprobs_or_none,
include_usage=include_usage,
index=0,
finish_reason=None,
)
break
returned_tokens += 1
yield {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": self.detokenize([token]).decode(
"utf-8", errors="ignore"
),
"index": 0,
"logprobs": logprobs_or_none,
"finish_reason": None,
}
],
}
yield {
"id": completion_id,
"object": "text_completion",
"created": created,
"model": model_name,
"choices": [
{
"text": "",
"index": 0,
"logprobs": None,
"finish_reason": finish_reason,
}
],
}
text = self.detokenize([token]).decode("utf-8", errors="ignore")
yield self._create_chunk(
completion_id=completion_id,
created=created,
model_name=model_name,
text=text,
logprobs_or_none=logprobs_or_none,
include_usage=include_usage,
index=0,
finish_reason=None,
)
yield self._create_chunk(
completion_id= completion_id,
created= created,
model_name=model_name,
text="",
index=0,
logprobs_or_none= None,
include_usage=include_usage,
usage=None,
finish_reason=finish_reason)

if include_usage:
yield self._create_chunk(
completion_id=completion_id,
created=created,
model_name=model_name,
text="",
logprobs_or_none=None,
include_usage=include_usage,
index=0,
finish_reason=None,
usage={
"prompt_tokens": len(prompt_tokens),
"completion_tokens": returned_tokens,
"total_tokens": len(prompt_tokens) + returned_tokens,
},
)
if self.cache:
if self.verbose:
print("Llama._create_completion: cache save", file=sys.stderr)
Expand Down Expand Up @@ -1736,6 +1793,7 @@ def logit_bias_processor(
},
}


def create_completion(
self,
prompt: Union[str, List[int]],
Expand All @@ -1753,6 +1811,7 @@ def create_completion(
repeat_penalty: float = 1.0,
top_k: int = 40,
stream: bool = False,
stream_options: Optional[StreamOptions] = None,
seed: Optional[int] = None,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
Expand Down Expand Up @@ -1816,6 +1875,7 @@ def create_completion(
repeat_penalty=repeat_penalty,
top_k=top_k,
stream=stream,
stream_options=stream_options,
seed=seed,
tfs_z=tfs_z,
mirostat_mode=mirostat_mode,
Expand Down Expand Up @@ -1850,6 +1910,7 @@ def __call__(
repeat_penalty: float = 1.0,
top_k: int = 40,
stream: bool = False,
stream_options: Optional[StreamOptions] = None,
seed: Optional[int] = None,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
Expand Down Expand Up @@ -1913,6 +1974,7 @@ def __call__(
repeat_penalty=repeat_penalty,
top_k=top_k,
stream=stream,
stream_options=stream_options,
seed=seed,
tfs_z=tfs_z,
mirostat_mode=mirostat_mode,
Expand All @@ -1938,6 +2000,7 @@ def create_chat_completion(
min_p: float = 0.05,
typical_p: float = 1.0,
stream: bool = False,
stream_options: Optional[StreamOptions] = False,
stop: Optional[Union[str, List[str]]] = [],
seed: Optional[int] = None,
response_format: Optional[ChatCompletionRequestResponseFormat] = None,
Expand Down Expand Up @@ -2011,6 +2074,7 @@ def create_chat_completion(
logprobs=logprobs,
top_logprobs=top_logprobs,
stream=stream,
stream_options=stream_options,
stop=stop,
seed=seed,
response_format=response_format,
Expand Down
Loading