@miller-tech/uap 2.36.4 → 2.39.1

package/config/model-profiles/qwen38.json

@@ -3,8 +3,8 @@
   "_description": "Qwen3.8 Flash-Next — local Strata serve-layer deployment for agentic tool-calling workloads. Successor to qwen36 and to the 2026-09-20 GSQ-RCO llama.cpp deployment; the default local executor in UAP routing. Routes through the anthropic-proxy for the tool/finalize guardrails, NOT the inference server directly.",
   "_engine": "strata serve layer 0.1.44.1 (build_info 'Strata 0.1.44.1', compiled locally with the user-space CUDA 13.4 toolkit) on :8080, launched by the systemd user unit uap-strata-server.service with /home/cogtek/dev/strata/strata-swift-iq3_s.json since the 2026-10-17 Swift 1.5 A/B trial (revert: repoint the unit's --config at strata-iq3_s.json, daemon-reload, restart; the unit file carries the dated note). Serves llama.cpp-shaped /props, /slots and /v1, plus a JSON /metrics (engine/live/requests — NOT Prometheus text; `uap inference health` auto-detects it). Weights: strata-data pack swift-iq3_s (Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S, the UkisAI reasoning-efficient post-train, since the 2026-10-17 trial; predecessor base Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S since the 2026-10-07 swap per the IQ3_S speed≈IQ3_XXS reports; earlier predecessors Swift 1.5 IQ3_XXS 2026-10-06, Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS, native experts + PLE shard) with MTP speculative drafts (spec 8 since the 2026-10-06 tuning; spec-min-p 0.70 after the 0.1.44.1-era calibration, previously 0.8, before that 0.5; previously spec 4, mtp_max 4) and an auto-sized expert cache. Model id advertised at /v1/models: swift-1.5-iq3_s; the server ignores requested model ids entirely, which is why stale client pins (e.g. a resumed session still naming qwen36 or qwen3.8-flash-next ids) keep working.",
   "_engine_history": "SUPERSEDED 2026-10-04: this profile described buun-llama-cpp behind uap-gsq-rco-server.service (2 rails sharing a 229376-cell VBR pool, DFlash2 draft). That unit is now INACTIVE — the box serves the Strata layer on :8080 instead (tuned 2026-10-03: +15% over the llama.cpp fork on agent workloads). The old description was itself a 2026-09-20 correction of a ninfer-serve description. Verify against /metrics before trusting engine claims here; the dead unit's ExecStart still reports -c 229376 to systemctl, and on 2026-10-04 `uap inference health` printed that stale pool size for the live (131072) backend.",
-  "model": "swift-1.5-iq3_s",
-  "_model_comment": "The alias the live backend advertises at /v1/models; the rails suite drift-checks this field against the serving process (test/model-profiles-qwen38-rails.test.ts). 2026-10-17: the Swift 1.5 IQ3_S A/B trial swapped the unit from the base pack, which advertised qwen3.8-flash-next-iq3_s. The server ignores requested model ids, so clients pinning the old id are unaffected; on revert, restore this field alongside the unit's --config.",
+  "model": "abliterated-iq2_xs",
+  "_model_comment": "The alias the live backend advertises at /v1/models; the rails suite drift-checks this field against the serving process (test/model-profiles-qwen38-rails.test.ts). 2026-10-17: the Swift 1.5 IQ3_S A/B trial swapped the unit from the base pack, which advertised qwen3.8-flash-next-iq3_s. The server ignores requested model ids, so clients pinning the old id are unaffected; on revert, restore this field alongside the unit's --config. 2026-10-28: the unit default moved to the DavidAU abliterated IQ2_XS arm (uap-strata-server -> strata-abliterated-iq2_xs.json), so the live alias is abliterated-iq2_xs; updated here the same day the rails drift-check caught it (the _engine description above still documents the Swift 1.5 A/B trial era).",
   "max_tokens": 57344,
   "_max_tokens_comment": "Deliberately generous: 50% of the 114688 window, the same ratio the previous 65536/131072 pairing used. The engine returns chain-of-thought in a SEPARATE reasoning_content field, so reasoning spends completion budget without ever appearing in content. A tight cap therefore does not produce a short answer, it produces an EMPTY one with finish_reason=length — measured previously as ~90% first-turn decode failure that read as bad output format and was really budget truncation. Do not lower this without re-measuring decode compliance. (The proxy additionally clamps tool turns to PROXY_TOOL_TURN_MAX_TOKENS=32768 regardless.)",
   "temperature": 0.3,
@@ -75,4 +75,4 @@
     "endpoint": "http://127.0.0.1:4000/v1",
     "_comment": "Route through the anthropic-proxy, NOT the strata port directly — the proxy applies the tool/finalize guardrails and, via this profile, the per-session context window that clients size their history to. Clients select this profile with the header 'x-uap-model-profile: qwen38' (or a uap_model_profile body param)."
   }
-}
+}

package/config/model-registry.json

@@ -54,6 +54,28 @@
           "engine": "strata",
           "unit": "uap-strata-server",
           "note": "Swift 1.5 (UkisAI reasoning-efficient Flash-Next post-train) GSQ-RCO IQ3_S on the strata engine; A/B trial vs strata-iq3_s since 2026-10-17 with identical tuning (spec 8/0.70, vram-reserve 3072, int8 KV resident 32768, 131k). Weights: ukisai Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S, sha256-verified, PLE table in shard 1 (setup-detected); pack strata-data/packs/swift-iq3_s; config strata-swift-iq3_s.json behind uap-strata-server.service (revert: point the unit --config back at strata-iq3_s.json, daemon-reload, restart). Run `uap models measure` to record this machine's footprint before it can load (fail-closed). Trial evidence: /home/cogtek/dev/strata/bench/results/2026-10-17-swift-iq3s-trial/."
+        },
+        "strata-abliterated-iq2_xs": {
+          "status": "unmeasured",
+          "engine": "strata",
+          "unit": "uap-strata-server",
+          "note": "The LIVE :8080 default arm since 2026-10-28 (DavidAU abliterated Flash-Next at IQ2_XS, full expert pinning without the IQ3_S RAM-OOM risk; engine 0.1.52, config strata-abliterated-iq2_xs.json). Raw GGUF shards declared here: the pack (strata-data/packs/abliterated-iq2_xs) is derived from them by tools/iq_pack.py and rebuilds from these files; shard 2 is byte-identical across the repo's tiers (verified 2026-10-28).",
+          "files": [
+            {
+              "path": "/home/cogtek/models/strata-data/models/abliterated-IQ2_XS/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-IQ2_XS-00001-of-00002.gguf",
+              "repo": "SC117/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF",
+              "file": "Qwen3.8-Flash-Next-GSQ-RCO-abliterated-IQ2_XS-00001-of-00002.gguf",
+              "bytes": 39665015072,
+              "sha256": "4acc3ebb5050d20117bdc77fc39200e709dfe2b2367f3522c95b9112a30aff81"
+            },
+            {
+              "path": "/home/cogtek/models/strata-data/models/abliterated-IQ2_XS/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-IQ2_XS-00002-of-00002.gguf",
+              "repo": "SC117/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF",
+              "file": "Qwen3.8-Flash-Next-GSQ-RCO-abliterated-IQ2_XS-00002-of-00002.gguf",
+              "bytes": 28800138432,
+              "sha256": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"
+            }
+          ]
         }
       },
       "advertises": ["qwen3.8-flash-next-iq3_s", "swift-1.5-iq3_s", "gpt-5.1-codex"]
@@ -64,13 +86,21 @@
       "task_affinity": ["code", "general"],
       "affinity": { "device": ["gpu0"] },
       "endpoint": "http://127.0.0.1:8080",
-      "advertises": ["Qwen3.8-27B"],
+      "advertises": ["Qwen3.8-27B", "qwen38-ampere-27b", "qwen38-swift-27b", "Qwen3.8-27B-Q4_K_M.gguf"],
       "configs": {
         "llama-mtp": {
           "status": "unmeasured",
           "engine": "llama.cpp",
           "unit": "uap-llama-server",
-          "profile": "config/llama-profiles/qwen38-27b-mtp.env"
+          "profile": "config/llama-profiles/qwen38-27b-mtp.env",
+          "files": [
+            {
+              "path": "/home/cogtek/Downloads/Qwen3.8-27B-UD-IQ4_XS.gguf",
+              "repo": "unsloth/Qwen3.8-27B-GGUF",
+              "file": "Qwen3.8-27B-UD-IQ4_XS.gguf",
+              "bytes": 14252845984
+            }
+          ]
         },
         "exl3-4bit": {
           "status": "unmeasured",
@@ -86,23 +116,83 @@
         "signal": {
           "status": "unmeasured",
           "engine": "llama.cpp",
-          "unit": "uap-signal-server"
+          "unit": "uap-signal-server",
+          "files": [
+            {
+              "path": "/home/cogtek/models/signal-27b/AP-IQ4_XS/Signal-3.8-27B-AP-IQ4_XS.gguf",
+              "bytes": 14252844608
+            },
+            {
+              "path": "/home/cogtek/models/signal-27b/mmproj-BF16.gguf",
+              "bytes": 931146432
+            }
+          ]
         },
         "gsq-rco": {
           "status": "unmeasured",
           "engine": "llama.cpp",
           "unit": "uap-gsq-rco-server",
-          "service": "gsq-rco-server"
+          "service": "gsq-rco-server",
+          "files": [
+            {
+              "path": "/home/cogtek/models/qwen3.8-27b-gsq-rco/Qwen3.8-27B-GSQ-RCO-IQ3_S.gguf",
+              "bytes": 11771546784
+            },
+            {
+              "path": "/home/cogtek/models/qwen3.8-27b-gsq-rco/Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp.gguf",
+              "bytes": 12120016960
+            },
+            {
+              "path": "/home/cogtek/models/qwen3.8-27b-gsq-rco/mmproj-Qwen3.8-27B-BF16.gguf",
+              "bytes": 931146528
+            }
+          ]
         },
         "ampere": {
           "status": "unmeasured",
           "engine": "llama.cpp",
-          "unit": "uap-ampere-server"
+          "unit": "uap-ampere-server",
+          "rails": 2,
+          "note": "ATX-IQ4_XS-M (4.56 bpw, llamAmpere-tuned recipe: bulk IQ4_XS, Q5_0/Q6_K attention K/V, Q8_0 GDN vectors) on llamAmpere v0.3, the engine+quant pair measured on this GPU class (3090/3090 Ti): 99.4 tok/s clean decode, 93.2 at 100K depth, exact p/q draft verification (output distribution = target's exactly). The 2-rail arm since 2026-10-30, measured live: -c 86016 total / -np 2 = 43008 per rail, q8_0 K + turbo3 V KV (the fused SM86 MMA path), MTP-3 spec decode (70.4% acceptance measured), engine 18,400 MiB with 3,072-3,156 MiB free beside the live desktop — ~52 tok/s solo decode engine-side at full 318W draw (2 rails buy concurrency, not aggregate). Unit ExecStartPre runs `uap models ensure-by-path` so a deleted GGUF re-downloads before the start fails.",
+          "files": [
+            {
+              "path": "/home/cogtek/models/atx-iq4xs-m/Qwen3.8-27B-ATX-4-XS.gguf",
+              "repo": "jakeatx/Qwen3.8-27B-ATX-IQ4_XS-M-GGUF",
+              "file": "Qwen3.8-27B-ATX-4-XS.gguf",
+              "bytes": 15588551712,
+              "sha256": "5cf05ad901dcaa76f41db13a5629146ed882219339377a80e37b12a8528d963b"
+            }
+          ]
         },
         "swift": {
           "status": "unmeasured",
           "engine": "llama.cpp",
-          "unit": "uap-swift-server"
+          "unit": "uap-swift-server",
+          "files": [
+            {
+              "path": "/home/cogtek/models/swift-27b/Swift-Qwen3.8-27B-IQ4_XS.gguf",
+              "repo": "ukisai/Swift-Qwen3.8-27B-GGUF",
+              "file": "Swift-Qwen3.8-27B-IQ4_XS.gguf",
+              "bytes": 15688288352,
+              "sha256": "27f5531a012a4bb3c3e379318dc62216f057131d34347498fabfc23186a4ffda"
+            }
+          ]
+        },
+        "openjet": {
+          "status": "unmeasured",
+          "engine": "openjet",
+          "unit": "uap-openjet-server",
+          "note": "OpenJet megakernel arm (:8080 default 2026-10-28 daytime, stood down evening for the strata switch-back). The Q4_K_M file under models/qwen3.8-27b-q4km went missing from disk after the arm was stood down — restored-on-demand via this manifest. unsloth deleted the file from main on 2026-08-19, so the source is the OpenJet catalog's pinned revision (open-jet src/config.py QWEN38_27B_URL).",
+          "files": [
+            {
+              "path": "/home/cogtek/models/qwen3.8-27b-q4km/Qwen3.8-27B-Q4_K_M.gguf",
+              "repo": "unsloth/Qwen3.8-27B-GGUF",
+              "file": "Qwen3.8-27B-Q4_K_M.gguf",
+              "rev": "566347e99d832d31f6245a577198d10d5bdd389e",
+              "bytes": 17106775008,
+              "sha256": "7e78da5d7e3ae28d178121f58646953305f3e5bd3cb46f4a75584e8b6c6fe169"
+            }
+          ]
         }
       }
     }