Introduce context compaction: optimize conversation context persistence, reduce redundant tokens, and enhance summary handling.
This commit is contained in:
@@ -10,3 +10,9 @@ UPLOAD_MAX_PIXELS=30000000
|
||||
VISION_MODEL=gemma3:4b
|
||||
VISION_DEBUG=false
|
||||
VISION_DEBUG_SHA256_ENABLED=false
|
||||
# Required to enable automatic context compaction; use the exact hosted Ollama tag.
|
||||
CONTEXT_MODEL=
|
||||
CONTEXT_COMPACTION_TRIGGER_TOKENS=6000
|
||||
CONTEXT_COMPACTION_RECENT_TOKENS=2500
|
||||
CONTEXT_COMPACTION_MIN_MESSAGES=4
|
||||
CONTEXT_COMPACTION_MAX_CHARS=8000
|
||||
|
||||
Reference in New Issue
Block a user