{ "scope": "Pinned deployment components and source-defined request protocols; not new inference. Audio JSON is the saved deployment configuration, not a request-time runtime dump.", "components": [ { "name": "llama-swap", "version": "262", "role": "Backend process management and API proxy", "source": { "path": "jarvis-flash174/llama-swap/jarvis-llama-swap.service", "bytes": 1296, "sha256": "09d723312009b7c849589ebfe1ffe10c0795148cd37a4bcd5d311df31a68c1f7" }, "argv": [ "/llama-swap", "--config", "/config.yaml", "--listen", "127.0.0.1:14003" ], "launch_command": "'/llama-swap' \\\n --config '/config.yaml' \\\n --listen 127.0.0.1:14003", "unset_environment": [ "GGML_CUDA_REGISTER_HOST", "GGML_SCHED_PREFETCH_EXPERTS", "GGML_CUDA_ENABLE_UNIFIED_MEMORY" ] }, { "name": "Nginx", "role": "Streaming HTTPS/API proxy; system package", "argv": [ "/usr/sbin/nginx", "-c", "/nginx.conf", "-g", "daemon off;" ], "launch_command": "/usr/sbin/nginx \\\n -c '/nginx.conf' \\\n -g 'daemon off;'", "service_source": { "path": "jarvis-flash174/switching/systemd/jarvis-api-proxy.service", "bytes": 854, "sha256": "9d571ff2411b6086a947c5b1b49abff3494e243cde07ccb286f8297a0ae18a1c" }, "source": { "path": "jarvis-flash174/switching/nginx/api-inference.conf", "bytes": 4570, "sha256": "50f00248eeb720ba8019cb841985d8fa450b125f8ab9d04952dd5b1b17e8211b" }, "proxy_settings": { "proxy_read_timeout": [ "1860s" ], "proxy_send_timeout": [ "60s" ], "send_timeout": [ "60s" ], "proxy_buffering": [ "off" ], "proxy_request_buffering": [ "on" ], "proxy_http_version": [ "1.1" ], "proxy_next_upstream": [ "off" ], "client_body_timeout": [ "30s" ], "client_header_timeout": [ "15s" ], "client_max_body_size": [ "5m", "25m", "1m" ] }, "settings_scope": "Values from the saved switching configuration; body limits differ by route; not a full request-time configuration dump" }, { "name": "LiteLLM", "version": "1.104.0", "role": "OpenAI Chat Completions / Responses gateway", "image": "ghcr.io/berriai/litellm@sha256:625981c83410a3ea68eb0697590a57ec1d764d634514d54fa5db0591077ee839", "source": { "path": "jarvis-api168/compose.yaml", "bytes": 1845, "sha256": "340f464733051bd4e44b4f3ee7efb490df8511d8dbf638eceec65ad6dda4a396" }, "container_arguments": [ "--config", "/app/jarvis-config.yaml", "--host", "127.0.0.1", "--port", "14000" ], "launch_scope": "Image default entrypoint plus recorded container arguments; not a replacement executable", "limits": { "cpus": 2, "memory_GiB": 3, "pids": 512 }, "configuration_source": { "path": "jarvis-api168/config.yaml", "bytes": 2511, "sha256": "4f787cd766fb83edd4510e9b47de6244f8083b4819ac292466c28b1bf38704b6" }, "settings": { "num_retries": 0, "request_timeout_seconds": 1860, "drop_params": true, "turn_off_message_logging": true, "router_num_retries": 0, "router_timeout_seconds": 1860 }, "functional_test_settings": { "use_chat_completions_api": true, "sse_keepalive_ping_interval_seconds": 15 }, "functional_settings_source": { "path": "jarvis-fixes195/install_keepalive.py", "bytes": 4374, "sha256": "e1937aee89dbfd0b1a2c680e0db6983757756d0641ad6a01887eee05a492c591" } }, { "name": "Codex CLI", "version": "0.160.1", "role": "Long-history, native compaction and tool-continuation client" }, { "name": "Pithagoras", "repository": "https://github.com/thecodacus/Pithagoras", "commit": "a14763d7f6172d632ebe8577ee2cb581d8e1ae2e", "image": "ghcr.io/thecodacus/pithagoras@sha256:9c899f8675392e9b74005e09d27d9b4071efa1a5fbb18c7bba61e2a196e4ef07", "pi_sdk_lock_version": "0.82.1", "source_patches": false, "role": "Agent / tool / voice integration; EXECUTOR=host inside an unprivileged workspace container", "argv": [ "/usr/bin/docker", "compose", "--env-file", "portal.env", "-f", "compose.yml", "up", "-d", "--pull", "never" ], "launch_command": "/usr/bin/docker compose \\\n --env-file portal.env \\\n -f compose.yml \\\n up \\\n -d \\\n --pull never", "service_source": { "path": "jarvis-flash174/jarvis-pithagoras.service", "bytes": 680, "sha256": "a6c35ac063f39c25b9b40416496784bf31096b4a0c997422b20f80df4d381b03" }, "compose_source": { "path": "jarvis-flash174/voice/portal-compose.yml", "bytes": 3953, "sha256": "061f6330c8065540880199d163723dcdf2eb02b046cc34d853a6be2892bfcb69" }, "working_directory": "/pithagoras", "settings": { "VOICE_PIPELINE_MODE": "parallel", "VOICE_SKIP_FIRST_THINKING": true, "VOICE_STATUS_SPEECH": false } }, { "name": "pi-web-access", "repository": "https://github.com/nicobailon/pi-web-access", "version": "0.35.0", "commit": "72c6e67787d67d8a7d01bf0abf30c072a6112de6", "sha256": "42e48533218d038d483cbd59c323aa2d7f50300b47f7c65954e9b9bcb40ee09e", "role": "Stock search and document retrieval tools" }, { "name": "Chromium", "repository": "https://github.com/chromedp/docker-headless-shell", "image": "docker.io/chromedp/headless-shell@sha256:a6beb6f169686744e05912a3ae25d97ee808510c9370142b00c3eb38c983db0e", "version_observed": "155.0.8059.26", "role": "Headless browser; CPU/software rendering", "argv": [ "/headless-shell/headless-shell", "--remote-debugging-address=0.0.0.0", "--remote-debugging-port=9223", "--disable-gpu", "--enable-unsafe-swiftshader", "--headless", "--no-first-run", "--no-default-browser-check", "--user-data-dir=/profile", "about:blank" ], "launch_command": "/headless-shell/headless-shell \\\n --remote-debugging-address=0.0.0.0 \\\n --remote-debugging-port=9223 \\\n --disable-gpu \\\n --enable-unsafe-swiftshader \\\n --headless \\\n --no-first-run \\\n --no-default-browser-check \\\n --user-data-dir=/profile about:blank", "source": { "path": "jarvis-flash174/voice/portal-compose.yml", "bytes": 3953, "sha256": "061f6330c8065540880199d163723dcdf2eb02b046cc34d853a6be2892bfcb69" }, "limits": { "cpus": 2, "memory_GiB": 2, "shm_GiB": 1, "pids": 256 }, "CDP_transport": { "argv": [ "socat", "TCP4-LISTEN:9222,fork,reuseaddr", "TCP4:127.0.0.1:9223" ], "cpus": 0.5, "memory_MiB": 128 } }, { "name": "audio.cpp", "repository": "https://github.com/0xShug0/audio.cpp", "release": "v0.9.0", "commit": "795c45fbde0a7d29c93b22199728ff5caaec02e5", "cuda_architecture": "86", "source_patches": false, "build_commands": [ "bash scripts/build_linux.sh --native-model-manager --system-openssl --cuda on --cuda-arch 86 --build-dir \"$source_dir/build-cuda\" --build-type Release --model-set custom --models chatterbox --target audiocpp_server --jobs 4", "bash scripts/build_linux.sh --cuda off --build-dir \"$source_dir/build-cpu\" --build-type Release --model-set custom --models qwen3_asr --target audiocpp_server --jobs 4" ], "models": { "repository": "audio-cpp/audio.cpp-gguf", "revision": "6d5436fc85f7a20c2e9f4e472b7f3a532f686444", "chatterbox": { "file": "Chatterbox-GGUF/chatterbox-q8_0.gguf", "bytes": 2088393668, "sha256": "d586dd1aa59613cab8046176fb7ca5ba191c02a9b10ffa5b0d892ed22b470656" }, "qwen3_asr": { "file": "Qwen3-ASR-0.6B-GGUF/qwen3-asr-0.6b-q8_0.gguf", "bytes": 1151272416, "sha256": "6c44ec2fb4cee513892d7863c1fcc3ea6b699ffa4d899b0ef4ab19956d9544f7" } }, "tts": { "source": { "path": "jarvis-flash174/voice/jarvis-voice-tts.service", "bytes": 713, "sha256": "87090ca25a213dd16a2fd961c7cad6a68b1ff8fedfa54ceb782a70b271a930e7" }, "argv": [ "/audiocpp_server", "--config", "/audio-tts.json", "--no-ui" ], "launch_command": "'/audiocpp_server' \\\n --config '/audio-tts.json' \\\n --no-ui", "environment": [ "CUDA_DEVICE_ORDER=PCI_BUS_ID", "LD_LIBRARY_PATH=/lib", "CUDA_VISIBLE_DEVICES=0" ], "config_source": { "path": "jarvis-flash174/voice/audio-tts.json", "bytes": 607, "sha256": "2705a142c2bf22fe4909f51f0db6df09f438c4cbc554f9949870c3fdcb363cef" }, "configuration": { "host": "127.0.0.1", "port": 7862, "backend": "cuda", "device": 0, "threads": 4, "lazy_load": false, "idle_unload_ms": 0, "min_free_memory_mb": 2048, "max_loaded_models": 1, "ui": false, "ui_management": true, "log_request_body": false, "models": [ { "id": "chatterbox", "family": "chatterbox", "path": "/chatterbox-q8_0.gguf", "task": "clon", "mode": "offline", "session_options": { "chatterbox.multilingual_t3": "v3", "chatterbox.conditionals_cache_slots": "2" } } ] } }, "asr": { "source": { "path": "jarvis-flash174/voice/jarvis-voice-asr.service", "bytes": 624, "sha256": "9f06274b1d4b83976c88594bcc0af376ae1c4bbbdd7d2e013b517f5d381373b9" }, "argv": [ "/audiocpp_server", "--config", "/audio-asr.json", "--no-ui" ], "launch_command": "'/audiocpp_server' \\\n --config '/audio-asr.json' \\\n --no-ui", "environment": [ "CUDA_VISIBLE_DEVICES=" ], "config_source": { "path": "jarvis-flash174/voice/audio-asr.json", "bytes": 428, "sha256": "92f8250f4b5b623e734343d3d419eadebf411ed58523ef331d32e8311e1d7142" }, "configuration": { "host": "127.0.0.1", "port": 7863, "backend": "cpu", "device": 0, "threads": 4, "lazy_load": true, "idle_unload_ms": 90000, "max_loaded_models": 1, "ui": false, "log_request_body": false, "models": [ { "id": "qwen3-asr", "family": "qwen3_asr", "path": "/qwen3-asr-0.6b-q8_0.gguf", "task": "asr", "mode": "offline" } ] } } } ], "request_protocols": [ { "name": "Upstream text throughput / C053 and C057", "source": { "path": "jarvis-flash174/upstream178_screen.py", "bytes": 13617, "sha256": "9041198db263ad5230e52e9c6d82eda65ebd1b7a6f6a3ebad1f8218f30767252" }, "endpoint": "/completion", "input": "Exact token array, complete native chat template; token digest recorded per observation", "parameters": { "n_predict": 512, "stream": true, "cache_prompt": false, "temperature": 1.0, "top_p": 0.95, "top_k": 20, "min_p": 0, "presence_penalty": 0, "repeat_penalty": 1.0, "seed": 174, "ignore_eos": true }, "thinking": "Native default ON; no forced thought budget", "retrieval_changes": "Natural EOS; ignore_eos omitted; output budget min(4096, context - input - 32); exact values, JSON and closed reasoning checked" }, { "name": "Historical thecodacus throughput / C043", "source": { "path": "jarvis-flash174/benchmark_http.py", "bytes": 25314, "sha256": "0c3a53256f1cc3490f6f21c6ddf2bfd9f40c092dba010adb704704d176844dd8" }, "endpoint": "/completion", "parameters": { "n_predict": 512, "stream": true, "temperature": 1.0, "top_p": 0.95, "top_k": 20, "min_p": 0, "presence_penalty": 0, "repeat_penalty": 1.0, "ignore_eos": true }, "seed": "174 + repetition index", "cache_prompt": "false for fresh requests; true for the separate reuse-requested series; actual cache_n verified", "retrieval_changes": "Natural EOS, output budget min(16384, context - input - 32); cache_prompt false" }, { "name": "Integrated public Responses / M0402", "source": { "path": "jarvis-fixes195/test_long_stream.py", "bytes": 6907, "sha256": "2eb9a3dfe9d4147941ee6d23ec4bec61e538bb25311bdc5a88ee8138df8cfe43" }, "endpoint": "/v1/responses", "parameters": { "model": "jarvis-flash-next", "stream": true, "max_output_tokens": 2048, "reasoning": { "effort": "xhigh" } }, "input": "Repetitive synthetic history; actual 230100 tokens, no reused prefix, audio inactive", "client_timeout_seconds": 1860, "SSE_keepalive_seconds": 15, "scope": "Single transport/completion check, not a semantic-quality or concurrent-voice benchmark" }, { "name": "Twenty-phrase Polish voice transport", "source": { "path": "jarvis-flash174/voice/voice-acceptance182.py", "bytes": 11357, "sha256": "38a418c22dbb4e1d3c03b221adca2bd41af30ce7b455e884a4258ace64b700e7" }, "parameters": { "phrases": 20, "language": "pl", "voice_kind": "clone", "LLM_stream": true, "LLM_max_tokens": 1024, "voice_first_variant_enable_thinking": false, "TTS_client_timeout_seconds": 130, "ASR_client_timeout_seconds": 130, "LLM_client_timeout_seconds": 600, "resource_sample_interval_seconds": 0.5, "VRAM_baseline_guard_MiB": 4656, "VRAM_during_test_guard_MiB": 2048 }, "thinking_policy": "The later --voice-first variant explicitly disables thinking; the default path leaves native thinking enabled. This is not a claim that all earlier runs used the later first-response setting.", "input": "Synthetic Polish phrases; no user microphone recording redistributed", "timing": "TTS request to complete WAV receipt; not first audible audio. ASR uses the generated WAV.", "percentiles": "p50 = sorted values[n//2] (upper middle for n=20); p95 = nearest-rank sorted values[ceil(0.95*n)-1]", "concurrency": "Per-phrase llm_generating_at_tts_start is recorded; not every phrase overlaps active LLM decode", "scope": "Transport and memory checks, not 20 live microphone turns or subjective voice quality" } ], "cache_conditions": { "OS_file_cache": "Not routinely flushed; process restart does not establish cold NVMe", "prefix_cache": "Reported separately from fresh tokens; throughput comparisons use fresh input", "RAM_disk": "No RAM-disk A/B experiment; standard NVMe backing and OS page cache" }, "build_invocations": [ { "engine": "llama.cpp upstream", "source": { "path": "jarvis-flash174/upstream178-build.sh", "bytes": 2186, "sha256": "b46624daa26e715a97f149d9aaf39d700d2a794d342f242b157595a0ae1d2eba" }, "build_command": "cmake --build \"$stage/build-cuda86\" --parallel 4 --target llama-server llama-bench" }, { "engine": "ik_llama.cpp", "source": { "path": "jarvis-flash174/ik178-build.sh", "bytes": 2793, "sha256": "28656824d4842236b9924952fd2efda768269114c93bd0b6e1940031d6fd6f4a" }, "build_command": "cmake --build \"$stage/build-cuda86\" --parallel 4 --target llama-server llama-bench test-chat-template test-chat-auto-parser" }, { "engine": "OptLlama", "source": { "path": "jarvis-flash174/opt181-build.sh", "bytes": 2109, "sha256": "71dfb27ad7df6d0f2f08ec6ee6c70b6cad4faaddb273238edb0d6dfcf3378b9d" }, "build_command": "cmake --build \"$stage/build-cuda86\" --parallel 4 \\\n --target llama-server llama-bench test-chat-template test-chat-auto-parser" }, { "engine": "Strata", "source": { "path": "jarvis-flash174/strata188/build.sh", "bytes": 727, "sha256": "62760e503314258134a5f84b010f605d8bba723a8465964fc505d2f5061c175b" }, "build_command": "cmake --build /output/build-cuda86 --target strata --parallel 4" } ], "source_versions": { "path": "jarvis-flash174/voice/versions.lock.json", "bytes": 7284, "sha256": "ef61da98b25027e84b1ff3330779bb8feb762a9bd7485bd77ac9a2a86a9c1e17" } }