prose-contracts/inference-optimization.prose.md

name, kind, description, id

name	kind	description	id
inference-optimization	responsibility	Optimizes the full Syslog inference stack — LiteLLM routing weights, GPU model assignments, agent context management, and prompt caching — to reduce response times to sub-15s average. All GPUs now at 128K context (stable ceiling).	067NC6KP02RG60S50M40E30928

Goal

Syslog inference response times reduced to sub-15s average by optimizing the full stack: LiteLLM routing weights, GPU model assignments, Hermes agent context management, and prompt caching — without sacrificing agent capability.

Requires

• inference-metrics: current SpendLogs from CT116 LiteLLM Postgres — avg request_duration_ms, prompt_tokens, completion_tokens, model_group breakdown, cache_hit rate over the last 3 hours
• agent-configs: current config.yaml from each active Hermes agent (Mumuni .123, any others on .129/.122) including compression, model, context_window, prompt_caching, memory settings
• gpu-health: health check response from all 3 GPU backends (strix-moe .15:8080, gpu-dense .8:8080, gpu-vision .110:8080)

Maintains

The optimized inference stack configuration — every change is applied and verified end-to-end. Postcondition: avg request_duration_ms ≤ 15000 for 90% of inference calls.

liteLLM-routing

The syslog-auto routing weights, model-specific timeouts, RPM limits, and model_list entries on CT116 /opt/inference-harness/litellm_config.yaml.

agent-compression

Each Hermes agent's ~/.hermes/config.yaml compression, context_window, prompt_caching, and model sections.

prompt-caching

LiteLLM cache configuration and llama.cpp --cache-prompt flag on GPU hosts.

verification

End-to-end latency measurements after changes applied — at least 3 test inference calls per model path measuring ttft (time-to-first-token) and total duration.

Continuity

• input-driven

Strategies

Context is the root cause. Every ~46K prompt token costs ~87s of prefill time at 532 tok/s. Fix context first, routing second.

• Route by task: gpu-dense for code/standard queries; gpu-vision for vision/web-auxiliary; syslog-auto for compression.
• Compress aggressively: threshold at 40% (not 65%) — a 128K window should compact at 51K, not 85K. Target 15% tail (not 30%).
• Cache everything repeated: system prompts, skill docs, AGENTS.md — these never change between turns. Single-digit cache hit rate is unacceptable.
• Lower context ceiling: 128K window is the stable ceiling for agent conversations. GPUs reduced from 256K to 128K (2026-07-17). 128K window should compact at 85K (0.65 threshold). For larger contexts, route to external providers.

Shape

• self: analyze metrics, compute optimal configs, apply changes, verify
• delegates:
  • apply-liteLLM: update litellm_config.yaml and reload
  • apply-agent-config: update hermes config.yaml per agent
  • verify-latency: run test inference calls and measure response

Execution

-- Phase 1: Analyze current state (already complete)
-- Phase 2: Apply LiteLLM routing optimization

call apply-liteLLM-routing
  config_path: /opt/inference-harness/litellm_config.yaml
  host: 192.168.68.116

-- Phase 3: Apply agent context compression optimization  

call apply-agent-compression
  agent: mumuni
  host: 192.168.68.14
  config_path: /home/hermes/.hermes/config.yaml

-- Phase 4: Enable llama.cpp prompt caching on GPU hosts

call enable-prompt-caching
  hosts: [192.168.68.15, 192.168.68.8, 192.168.68.110]

-- Phase 5: Verify end-to-end latency
-- `models` is a literal verification parameter (a snapshot only): the authoritative registry is
-- CT 116 /opt/inference-harness/litellm_config.yaml; re-read it before use.

call verify-latency
  host: 192.168.68.116
  models: [syslog-auto, gpu-dense, gpu-vision, strix-moe]