From ffa86d48cdd2b7eeb516a216526cf592ce3433a3 Mon Sep 17 00:00:00 2001 From: jenkins Date: Wed, 12 Aug 2026 01:54:30 -0300 Subject: [PATCH] hermes: remove auto response ceiling --- services/hermes/switchyard-configmap.yaml | 5 ----- testing/tests/test_hermes_chat_quality.py | 2 ++ 2 files changed, 2 insertions(+), 5 deletions(-) diff --git a/services/hermes/switchyard-configmap.yaml b/services/hermes/switchyard-configmap.yaml index 6e8e4c045..4f401d205 100644 --- a/services/hermes/switchyard-configmap.yaml +++ b/services/hermes/switchyard-configmap.yaml @@ -190,7 +190,6 @@ data: # Keep both xhigh providers first so recovery can escalate, never downgrade. targets = ["codex_sol_xhigh", "claude_opus_xhigh", "codex_sol_high", "claude_sonnet_high", "claude_opus_high", "codex_terra_high", "codex_sol_medium", "claude_sonnet_medium", "codex_terra_medium", "local_qwen_medium", "codex_terra_low", "codex_luna_low", "claude_haiku_low", "local_qwen_low"] default_target = "codex_terra_medium" - max_output_tokens = 96 session_affinity = false recent_turn_window = 8 context_window = 272000 @@ -248,7 +247,6 @@ data: # Keep both xhigh providers first so recovery can escalate, never downgrade. targets = ["codex_sol_xhigh", "claude_opus_xhigh", "codex_sol_high", "claude_sonnet_high", "claude_opus_high", "codex_terra_high", "codex_sol_medium", "claude_sonnet_medium", "codex_terra_medium", "local_qwen_medium", "codex_terra_low", "codex_luna_low", "claude_haiku_low", "local_qwen_low"] default_target = "codex_terra_medium" - max_output_tokens = 96 session_affinity = false recent_turn_window = 8 context_window = 272000 @@ -307,7 +305,6 @@ data: # Keep both xhigh providers first so recovery can escalate, never downgrade. targets = ["claude_opus_xhigh", "codex_sol_xhigh", "claude_sonnet_high", "codex_sol_high", "claude_opus_high", "codex_terra_high", "claude_sonnet_medium", "codex_sol_medium", "codex_terra_medium", "local_qwen_medium", "claude_haiku_low", "codex_luna_low", "codex_terra_low", "local_qwen_low"] default_target = "claude_sonnet_high" - max_output_tokens = 96 session_affinity = false recent_turn_window = 12 context_window = 272000 @@ -366,7 +363,6 @@ data: # Keep both xhigh providers first so recovery can escalate, never downgrade. targets = ["codex_sol_xhigh", "claude_opus_xhigh", "codex_sol_high", "claude_opus_high", "claude_sonnet_high", "codex_terra_high", "codex_sol_medium", "claude_sonnet_medium", "codex_terra_medium", "local_qwen_medium", "codex_luna_low", "claude_haiku_low", "codex_terra_low", "local_qwen_low"] default_target = "codex_sol_high" - max_output_tokens = 96 session_affinity = false recent_turn_window = 16 context_window = 272000 @@ -423,7 +419,6 @@ data: classifier_target = "classifier" targets = ["worker_codex_sol_high", "worker_claude_sonnet_high", "worker_codex_sol_xhigh", "worker_claude_opus_xhigh", "worker_codex_terra_medium", "worker_claude_sonnet_medium", "worker_codex_luna_low", "worker_claude_haiku_low"] default_target = "worker_codex_sol_high" - max_output_tokens = 96 session_affinity = false recent_turn_window = 16 context_window = 272000 diff --git a/testing/tests/test_hermes_chat_quality.py b/testing/tests/test_hermes_chat_quality.py index fe4223d62..da808acc3 100644 --- a/testing/tests/test_hermes_chat_quality.py +++ b/testing/tests/test_hermes_chat_quality.py @@ -854,6 +854,8 @@ def test_local_flux_runtime_and_gpu_handoff_are_flux_managed(): for route_name in ("auto_fast", "auto_balanced", "auto_deep", "auto_maximum"): leading_targets = set(routes[route_name]["targets"][:2]) assert leading_targets == {"codex_sol_xhigh", "claude_opus_xhigh"} + assert "max_output_tokens" not in routes[route_name] + assert "max_output_tokens" not in routes["worker_auto_maximum"] model_gate = _documents(HERMES / "model-gate-configmap.yaml")[0]["data"][ "model_gate.py" ]