fix: exclude reasoning tokens from compression trigger threshold#12481
Closed
Sanjays2402 wants to merge 1 commit into
Closed
fix: exclude reasoning tokens from compression trigger threshold#12481Sanjays2402 wants to merge 1 commit into
Sanjays2402 wants to merge 1 commit into
Conversation
teknium1
pushed a commit
that referenced
this pull request
Apr 20, 2026
…12026) Cherry-picked from PR #12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes #12026
teknium1
pushed a commit
that referenced
this pull request
Apr 20, 2026
…12026) Cherry-picked from PR #12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes #12026
Contributor
|
Merged via PR #13006 #13006 — your commit was cherry-picked onto current main with your authorship preserved. Cleanest fix of the four submissions. Thanks @Sanjays2402! |
ulasbilgen
pushed a commit
to ulasbilgen/hermes-adhd-agent
that referenced
this pull request
May 1, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
aj-nt
pushed a commit
to aj-nt/hermes-agent
that referenced
this pull request
May 1, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
Luminet2023
pushed a commit
to Luminet2023/hermes-agent
that referenced
this pull request
May 1, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
02356abc
pushed a commit
to 02356abc/hermes-agent
that referenced
this pull request
May 14, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
gweeteve
pushed a commit
to gweeteve/hermes-agent
that referenced
this pull request
Jun 2, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
Egavasyug
pushed a commit
to Egavasyug/hermes-agent
that referenced
this pull request
Jun 10, 2026
…ousResearch#12026) Cherry-picked from PR NousResearch#12481 by @Sanjays2402. Reasoning models (GLM-5.1, QwQ, DeepSeek R1) inflate completion_tokens with internal thinking tokens. The compression trigger summed prompt_tokens + completion_tokens, causing premature compression at ~42% actual context usage instead of the configured 50% threshold. Now uses only prompt_tokens — completion tokens don't consume context window space for the next API call. - 3 new regression tests - Added AUTHOR_MAP entry for @Sanjays2402 Closes NousResearch#12026
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
2 participants
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.
Problem
Fixes #12026
The compression trigger at
run_agent.py:11274sumslast_prompt_tokens + last_completion_tokensto determine whether to compress context. However, thinking models (GLM-5.1, QwQ, DeepSeek R1) inflatecompletion_tokenswith reasoning/thinking tokens that do not consume context window space.This causes premature compression — e.g. 40k prompt + 80k reasoning = 120k exceeds a 100k threshold, even though the actual context window usage is only 40k.
Fix
Use only
last_prompt_tokensfor the compression threshold check. Completion tokens are the model's output and don't contribute to context window pressure. The existing fallback path (estimate when prompt_tokens is 0) is preserved.Tests
Added
tests/run_agent/test_compression_trigger_excludes_reasoning.pywith 3 tests: