Repository navigation
fix(highlight): run inference under torch.inference_mode and cap context size - #2
Merged
Merged
Conversation
…ext size The model's batch inference path is not wrapped in no_grad, so autograd kept every layer's activations and memory scaled with input size. Measured on M1 (CPU, 8 GB sandbox): 12k chars peaked 7.9 GB, 24k was OOM-killed. On 2026-10-06 the live service reached ~31 GB and was OOM-killed 11 times; the memory storm left M1's network links in systemd-networkd 'failed' state and the box dropped offline two days later when the DHCP leases expired. With inference_mode: typical 2.0 GB, 12k 2.4 GB, 24k 3.4 GB, 192k 5.5 GB, settling back to ~2 GB after each request (150-request soak flat). Contexts over HIGHLIGHT_MAX_CONTEXT_CHARS (default 200000) are refused with 413, and each request's size is logged so the next incident names its input.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The model's batch inference path is not wrapped in no_grad, so autograd kept every
layer's activations and memory scaled with input size. Measured on M1 (CPU, 8 GB
sandbox): 12k chars peaked 7.9 GB, 24k was OOM-killed. On 2026-10-06 the live service
reached ~31 GB and was OOM-killed 11 times; the memory storm left M1's network links
in systemd-networkd 'failed' state and the box dropped offline two days later when the
DHCP leases expired.
With inference_mode: typical 2.0 GB, 12k 2.4 GB, 24k 3.4 GB, 192k 5.5 GB, settling back
to ~2 GB after each request (150-request soak flat). Contexts over
HIGHLIGHT_MAX_CONTEXT_CHARS (default 200000) are refused with 413, and each request's
size is logged so the next incident names its input.