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

Jump to bottom

Create a new worker implementation for Apple MLX #2937

Merged
merged 2 commits into from
Jan 24, 2024
Merged

Create a new worker implementation for Apple MLX #2937

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
18d221e
Create a new worker for Apple MLX
aliasaria Jan 19, 2024
0211a4e
format
aliasaria Jan 19, 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
Prev Previous commit
format
  • Loading branch information
@aliasaria
aliasaria committed Jan 19, 2024
commit 0211a4e1dd76ecdb7180b84d425ded1989f99867
30 changes: 14 additions & 16 deletions fastchat/serve/mlx_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,7 +119,10 @@ async def generate_stream(self, params):

finish_reason = "length"

for token, _ in zip(generate_step(context_mlx, self.mlx_model, temperature), range(max_new_tokens)):
for token, _ in zip(
generate_step(context_mlx, self.mlx_model, temperature),
range(max_new_tokens),
):
if token == self.mlx_tokenizer.eos_token_id:
finish_reason = "stop"
break
Expand All @@ -128,8 +131,7 @@ async def generate_stream(self, params):
last_token_decoded = self.mlx_tokenizer.decode([token.item()])
skip = len(tokens_decoded)

partial_stop = any(is_partial_stop(tokens_decoded, i)
for i in stop)
partial_stop = any(is_partial_stop(tokens_decoded, i) for i in stop)

if partial_stop:
finish_reason = "stop"
Expand All @@ -143,20 +145,17 @@ async def generate_stream(self, params):
"completion_tokens": len(tokens),
"total_tokens": len(context) + len(tokens),
},
"cumulative_logprob": [
],
"finish_reason": None # hard code for now
"cumulative_logprob": [],
"finish_reason": None, # hard code for now
}
# print(ret)
yield (json.dumps(ret) + "\0").encode()
ret = {
"text": self.mlx_tokenizer.decode(tokens),
"error_code": 0,
"usage": {
},
"cumulative_logprob": [
],
"finish_reason": finish_reason
"usage": {},
"cumulative_logprob": [],
"finish_reason": finish_reason,
}
yield (json.dumps(obj={**ret, **{"finish_reason": None}}) + "\0").encode()
yield (json.dumps(ret) + "\0").encode()
Expand Down Expand Up @@ -231,6 +230,7 @@ async def api_get_conv(request: Request):
async def api_model_details(request: Request):
return {"context_length": worker.context_len}


worker = None


Expand All @@ -246,13 +246,11 @@ def cleanup_at_exit():
parser = argparse.ArgumentParser()
parser.add_argument("--host", type=str, default="localhost")
parser.add_argument("--port", type=int, default=21002)
parser.add_argument("--worker-address", type=str,
default="http://localhost:21002")
parser.add_argument("--worker-address", type=str, default="http://localhost:21002")
parser.add_argument(
"--controller-address", type=str, default="http://localhost:21001"
)
parser.add_argument("--model-path", type=str,
default="microsoft/phi-2")
parser.add_argument("--model-path", type=str, default="microsoft/phi-2")
parser.add_argument(
"--model-names",
type=lambda s: s.split(","),
Expand Down Expand Up @@ -285,4 +283,4 @@ def cleanup_at_exit():
"MLX",
args.conv_template,
)
uvicorn.run(app, host=args.host, port=args.port, log_level="info")
uvicorn.run(app, host=args.host, port=args.port, log_level="info")