refac

backend/open_webui/main.py

@@ -1065,6 +1065,7 @@ async def chat_completion(
         # Chat Params
         stream_delta_chunk_size = form_data.get('params', {}).get('stream_delta_chunk_size')
         reasoning_tags = form_data.get('params', {}).get('reasoning_tags')
+        compact_token_threshold = form_data.get('params', {}).get('compact_token_threshold')
 
         # Model Params
         if model_info_params.get('stream_response') is not None:
@@ -1076,6 +1077,9 @@ async def chat_completion(
         if model_info_params.get('reasoning_tags') is not None:
             reasoning_tags = model_info_params.get('reasoning_tags')
 
+        if model_info_params.get('compact_token_threshold') is not None:
+            compact_token_threshold = model_info_params.get('compact_token_threshold')
+
         # parent_id signals intent:
         #   null   → new chat (root message, no parent)
         #   value  → follow-up (user message's parentId = prev assistant)
@@ -1133,6 +1137,7 @@ async def chat_completion(
             'params': {
                 'stream_delta_chunk_size': stream_delta_chunk_size,
                 'reasoning_tags': reasoning_tags,
+                'compact_token_threshold': compact_token_threshold,
                 'function_calling': (
                     form_data.get('params', {}).get('function_calling')
                     or model_info_params.get('function_calling')

backend/open_webui/utils/context_compaction.py

@@ -53,8 +53,9 @@ async def compact_messages_for_request(
         return messages, None, False
 
     messages, previous_summary = _apply_latest_summary_checkpoint(messages)
+    token_threshold = _resolve_token_threshold(config['token_threshold'], metadata)
     if (
-        not _exceeds_token_threshold(messages, system_prompt, previous_summary, config['token_threshold'])
+        not _exceeds_token_threshold(messages, system_prompt, previous_summary, token_threshold)
         or len(messages) <= 3
     ):
         return messages, previous_summary, False
@@ -147,6 +148,21 @@ async def _load_config() -> dict:
     }
 
 
+def _parse_positive_int(value: Any) -> int | None:
+    try:
+        parsed = int(value)
+    except (TypeError, ValueError):
+        return None
+    return parsed if parsed > 0 else None
+
+
+def _resolve_token_threshold(global_threshold: int, metadata: dict) -> int:
+    configured_threshold = _parse_positive_int((metadata.get('params') or {}).get('compact_token_threshold'))
+    if configured_threshold is None:
+        return global_threshold
+    return min(configured_threshold, global_threshold)
+
+
 def _apply_latest_summary_checkpoint(messages: list[dict]) -> tuple[list[dict], str | None]:
     summary = None
     summary_idx = None

backend/open_webui/utils/middleware.py

@@ -2076,6 +2076,7 @@ def apply_params_to_form_data(form_data, model):
         'stream_delta_chunk_size': int,
         'function_calling': str,
         'reasoning_tags': list,
+        'compact_token_threshold': int,
         'system': str,
     }
 

backend/open_webui/utils/payload.py

@@ -72,6 +72,7 @@ def remove_open_webui_params(params: dict) -> dict:
         'stream_delta_chunk_size': int,
         'function_calling': str,
         'reasoning_tags': list,
+        'compact_token_threshold': int,
         'system': str,
     }
 

src/lib/components/chat/Settings/Advanced/AdvancedParams.svelte

@@ -16,6 +16,7 @@
 		// Advanced
 		stream_response: null, // Set stream responses for this model individually
 		stream_delta_chunk_size: null, // Set the chunk size for streaming responses
+		compact_token_threshold: null,
 		function_calling: null,
 		reasoning_tags: null,
 		seed: null,
@@ -145,6 +146,52 @@
 				</div>
 			{/if}
 		</div>
+
+		<div>
+			<Tooltip
+				content={$i18n.t(
+					'Lower the context compaction token threshold for this model. The global context compaction threshold remains the maximum.'
+				)}
+				placement="top-start"
+				className="inline-tooltip"
+			>
+				<div class="flex w-full justify-between">
+					<div class=" self-center text-xs">
+						{$i18n.t('Context Compaction Threshold')}
+					</div>
+					<button
+						class="p-1 px-3 text-xs flex rounded-sm transition shrink-0 outline-hidden"
+						type="button"
+						on:click={() => {
+							params.compact_token_threshold =
+								(params?.compact_token_threshold ?? null) === null ? 80000 : null;
+						}}
+					>
+						{#if (params?.compact_token_threshold ?? null) === null}
+							<span class="ml-2 self-center"> {$i18n.t('Default')} </span>
+						{:else}
+							<span class="ml-2 self-center"> {$i18n.t('Custom')} </span>
+						{/if}
+					</button>
+				</div>
+			</Tooltip>
+
+			{#if (params?.compact_token_threshold ?? null) !== null}
+				<div class="flex mt-0.5 space-x-2">
+					<div class=" flex-1">
+						<input
+							class="text-sm w-full bg-transparent outline-hidden outline-none"
+							type="number"
+							placeholder={$i18n.t('Enter token threshold')}
+							bind:value={params.compact_token_threshold}
+							autocomplete="off"
+							min="1"
+							step="1"
+						/>
+					</div>
+				</div>
+			{/if}
+		</div>
 	{/if}
 
 	<div>