Skip to content

Draft: Add caching to MLX model #999

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Draft: Add caching to MLX model #999

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e870436
Add caching to MLX model
kilianyp Mar 16, 2025
796ebb2
Reset num_processed token
kilianyp Mar 17, 2025
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
11 changes: 10 additions & 1 deletion src/smolagents/models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -648,9 +648,11 @@ def __init__(
self.model_id = model_id
self.model, self.tokenizer = mlx_lm.load(model_id, tokenizer_config={"trust_remote_code": trust_remote_code})
self.stream_generate = mlx_lm.stream_generate
self.prompt_cache = mlx_lm.models.cache.make_prompt_cache(self.model)
self.tool_name_key = tool_name_key
self.tool_arguments_key = tool_arguments_key
self.is_vlm = False # mlx-lm doesn't support vision models
self._num_processed_token = 0

def __call__(
self,
Expand All @@ -677,12 +679,19 @@ def __call__(
tools=tools,
add_generation_prompt=True,
)
if self._num_processed_token > len(prompt_ids):
# TODO: this only works if the new prmopt_ids is shorter than
# what was previously processed.
self._num_processed_token = 0
prompt_ids = prompt_ids[self._num_processed_token:]
self._num_processed_token += len(prompt_ids)

self.last_input_token_count = len(prompt_ids)
self.last_output_token_count = 0
text = ""

for _ in self.stream_generate(self.model, self.tokenizer, prompt=prompt_ids, **completion_kwargs):
found_stop_sequence = False
for _ in self.stream_generate(self.model, self.tokenizer, prompt=prompt_ids, prompt_cache=self.prompt_cache, **completion_kwargs):
self.last_output_token_count += 1
text += _.text
for stop_sequence in prepared_stop_sequences:
Expand Down