Merged PR #50: fix/gpu-dense-docs

gpu-fleet.prose.md

@@ -14,9 +14,6 @@ description: >
   Instability observed near 100K at 256K (now all GPUs at 128K). 128K is the stable ceiling.
   For larger context needs → fall back to external providers (deepseek).
   VRAM headroom improved: RTX 3090 ~70%, RTX 5070 ~65%.
-  UPDATED 2026-08-15: gpu-dense swapped to Qwen3.8-27B-Uncensored-Q4_K_M
-  (~16.8GB, 128K ctx, --spec-type draft-mtp (v2)). Served on .8:8080 under the legacy
-  alias qwen3.6-27B-code for LiteLLM routing continuity. Replaces SmartCode-Fable-5-27B.
 agent: abiba
 triggers:
   - on model add/remove
@@ -74,7 +71,7 @@ triggers:
 │ RTX 3090 │  │ RTX 5070    │  │ Strix Halo│  │ GPU Monitor  │
 │ 24GB     │  │ 12GB        │  │ 64GB UMA  │  │ :9100        │
 │ 128K ctx │  │ 128K ctx    │  │ 128K ctx  │  │ Watchdog     │
-│ qwen3.6  │  │ gemma-4-12b │  │ qwen3.6   │  │ Prometheus   │
+│ qwen3.6  │  │ Qwen3.5-9B  │  │ qwen3.5   │  │ Prometheus   │
 │ 27B-code │  │ :8080       │  │ -35B-udq4 │  │ exporter     │
 │ :8080    │  │ :9400 (exp) │  │ :9400(exp)│  │ :9401        │
 │ :9400    │  └─────────────┘  └───────────┘  └──────────────┘
@@ -92,16 +89,13 @@ When a model is swapped on a GPU, ONLY the infrastructure layer changes — agen
 | `gpu-dense` | RTX 3090 (.8) | Qwen3.8-27B-Uncensored-Q4_K_M | Whatever runs on RTX 3090 |
 | `gpu-light` | RTX 5070 (.110) | gemma-4-12b | Whatever runs on RTX 5070 |
 
-**Backward compatibility**: Old model-specific names (qwen3.6-27B-code, gemma-4-12b, qwen3.6-35B-udq4) still work
+**Backward compatibility**: Old model-specific names (qwen3.6-27B-code, gemma-4-12b, qwen3.5-9b-it) still work
 but are deprecated for agent configs. Only the stable aliases survive model swaps.
 
 ## Current Model Assignments (2026-07-15)
 
 | Model | GPU | Host | VRAM | Ctx | KV Cache | Parallel | Batch/Ubatch | Status |
 |-------|-----|------|------|-----|----------|----------|-------------|--------|
-| Qwen3.8-27B-Uncensored-Q4_K_M | RTX 3090 | .8 (llm-gpu) | ~16.8/24.6GB | **128K** | turbo4 | 1 | 2048/1024 | ✅ healthy |
-| gemma-4-12b | RTX 5070 | .110 (ocu-llm) | ~7.8/12.2GB (65%) | 128K | q4_0 | 2 | 2048/1024 | ✅ healthy |
-| qwen3.6-35B-udq4 | Strix Halo Vulkan | .15 (amdpve) | ~22GB/64GB | 128K | q4_0 | 1 | 4096/1024 | ✅ 65 tok/s |
 
 ## Routing Configuration (LiteLLM — July 2026)
 
@@ -110,8 +104,6 @@ but are deprecated for agent configs. Only the stable aliases survive model swap
 | Model | GPU | Weight | RPM Cap | Timeout |
 |-------|-----|--------|---------|---------|
 | Qwen3.8-27B-Uncensored-Q4_K_M | RTX 3090 (.8:8080) | **0.55** | 500 | **300s** |
-| qwen3.6-35B-udq4 | Strix Halo (.15:8080) | **0.30** | 60 | **300s** |
-| gemma-4-12b | RTX 5070 (.110:8080) | **0.15** | 200 | **120s** |
 
 Note: All syslog-auto entries route directly to GPUs with `api_key: not-needed`. The router (port 9000) is NOT in the inference path.
 
@@ -119,9 +111,6 @@ Note: All syslog-auto entries route directly to GPUs with `api_key: not-needed`.
 
 | Model | RPM Cap | Notes |
 |-------|---------|-------|
-| strix-moe (qwen3.6-35B-udq4) | 40 | Tight cap — prevents Strix overload |
-| Qwen3.8-27B-Uncensored-Q4_K_M | 500 | High cap — primary workhorse (replaces qwen3.6-27B-code) |
-| gemma-4-12b | 500 | High cap — IQ4_NL+MTP, 122 tok/s |
 
 ### Stable Aliases (for agent configs — never change)
 
@@ -192,7 +181,7 @@ Show full fleet status: GPUs, models, VRAM, context windows, parallel slots, act
 3. Check LiteLLM: `curl http://192.168.68.116/health` (expect "I'm alive!")
 4. Check LiteLLM models: `curl -H "Authorization: Bearer $MASTER_KEY" http://192.168.68.116/v1/models`
 5. Check LiteLLM timeouts: `grep -n 'timeout:' /opt/inference-harness/litellm_config.yaml`
-   - gemma-4-12b: 120s, qwen3.6-27B-code: 300s, qwen3.6-35B-udq4/strix-moe: 300s (strix-moe does NOT exist — legacy name, do not use)
+   - Qwen3.5-9B: 120s, qwen3.6-27B-code: 300s, Carnice-Qwen3.6-MoE-35B-A3B/strix-moe: 300s (strix-moe alias retained, legacy name qwen3.6-35B-udq4 deprecated)
    - global request_timeout: 300s, nginx proxy_read_timeout: 600s
 6. Check AMD metrics: `curl http://192.168.68.15:9400/metrics` (Radeon 8060S, util%, VRAM, temp, power)
 7. Check port conflicts: verify only one llama-server on :8080 per host
@@ -249,14 +238,6 @@ If no SSH access, send Zulip DM via abiba-bot with vault update instructions.
 - **Router startup race**: Compose router.py doesn't call load_roster(). Reload thread sleeps 30s first.
   Fix: trigger roster reload via SSH after restart, or rebuild image with startup load_roster().
 - **LiteLLM /metrics**: Requires auth. Prometheus uses `/health/liveliness` as workaround.
-- **VRAM (2026-07-15)**: RTX 3090 at ~17/24.6GB (~70%) with **128K context** (reduced from 256K 2026-07-17). RTX 5070 at ~7.8/12.2GB (~65%) with 128K context + MTP. Strix Halo at ~7GB/64GB.
-- **RTX 3090 (2026-08-15)**: Swapped to Qwen3.8-27B-Uncensored-Q4_K_M (~16.8GB, replaced
-  SmartCode-Fable-5-27B-UD-Q3_K_XL). Service: `/home/llmuser/llama-fable-wrapper.sh`.
-  Served under alias `qwen3.6-27B-code` for LiteLLM routing continuity.
-- **RTX 5070 config (2026-07-15)**: Switched to IQ4_NL + MTP draft (Q8_0) at 128K context. Gen speed: 122 tok/s. VRAM: ~7.8/12.2GB (~65%). Service: `/home/llmuser/llama-wrapper.sh`. Config: `--model gemma-4-12b-it-IQ4_NL.gguf --spec-draft-model gemma-4-12b-it-Q8_0-MTP.gguf --spec-type draft-mtp --spec-draft-n-max 4 --ctx-size 131072`.
-- **LiteLLM timeout tuning (verified 2026-08-16)**: Qwen3.8-27B (alias qwen3.6-27B-code)
-  300s, gemma-4-12b 120s, qwen3.6-35B-udq4 300s, strix-moe 300s, syslog-auto routes all
-  300s. Nginx proxy_read_timeout: 600s. Global request_timeout: 300s.
 - **Strix Halo GPU**: Vulkan is the working backend (ROCm/HIP path abandoned — HSA runtime blocked on Debian 13). Build at `/root/llama.cpp/build-vk/`, commit `4fc4ec5` (2026-07-01), ggml 0.15.3 shared-lib arch. Mesa RADV 25.0.7, KHR_coopmat fast path active. ~70 tok/s gen, 532 tok/s prompt. Service: `strix-server.service` on port 8080, model: `qwen3.6-35B-udq4`, alias `strix-moe`, 128K context, flash-attn + q4 KV, multimodal (mmproj loaded).
 - **Port conflict detection (2026-07-05)**: All 3 GPU wrappers now detect ghost processes squatting port 8080 before starting. `.8` and `.110` use inline pre-start check in `llama-wrapper.sh`; `.15` uses `/usr/local/bin/port-cleanup.sh` ExecStartPre. Replaces the blanket `pkill -9 -x llama-server` on .15 which would kill ALL llama-server instances regardless of port. Ghost detection was the root cause of .8 crash-looping for 27+ restarts (stale pid 25836 squatting 8080 after OOM kill).
 - **Strix Halo thermal safeguard (2026-07-02)**: `strix-server.service` has `-n 8192` (hard generation cap per request). Without it, `--predict` defaults to -1 (infinity) — a runaway request from .123 (old Mumuni CT114 — now inside Abiba CT100 at .24) decoded 39,868 tokens over 24 min, pushing Tctl to 98°C (crit 89.8°C) and throttling 70→29 t/s. The cap bounds worst-case generation to ~5 min. Do NOT remove `-n` without a replacement ceiling. Sustained load hits ~84°C even at 92s; the APU is fanless/low-flow. Clients MUST also set `max_tokens`.
@@ -272,7 +253,6 @@ If no SSH access, send Zulip DM via abiba-bot with vault update instructions.
 | GPU | Model | Gen tok/s | Prompt tok/s | Baseline | Context |
 |-----|-------|-----------|--------------|----------|---------|
 | RTX 3090 (.8) | Qwen3.8-27B-Uncensored-Q4_K_M | **TBD** | — | — | **128K** |
-| RTX 5070 (.110) | gemma-4-12b (IQ4_NL+MTP) | **191** | — | — | **128K** |
 | Strix Halo (.15) | qwen3.6-35B-udq4 | **65** | 140 | — | **128K** |
 
 Benchmarks from 2026-07-17. Strix Halo model: qwen3.6-35B-udq4. RTX 5070 MTP provides 2.7x speedup over pre-upgrade 70 tok/s.
@@ -299,7 +279,8 @@ When the underlying model is swapped, only the LiteLLM config changes — agent
 ### Context Windows
 - RTX 3090: **128K** (reduced from 256K 2026-07-17) | RTX 5070: **128K** (reduced from 256K) | Strix Halo: **128K**
 - **All agents**: 128K ceiling — stable margin. For >128K workloads, use external providers (deepseek)
-- Compression threshold 0.65: fires at ~85K (~43K headroom before 128K ceiling)
+- Compression threshold 0.60: fires at ~77K (~51K headroom before 128K ceiling)
+- **Pi agents (Abiba)**: `compaction.reserveTokens: 52739` (≈60% of 128K)
 - Mumuni compression model alias: `strix-moe` with 300s timeout
 
 ### Mumuni Agent Profile
@@ -316,7 +297,7 @@ Mumuni (kagentz CT105, 192.168.68.14 — migrated from CT100 2026-08-29) is the
 | `aux.web_extract.model` | `gpu-light` | Web extraction |
 | `delegation.model` | `gpu-dense` | Sub-agent reasoning (RTX 3090) |
 | `context.max_context_window` | 131072 (128K) | Reduced from 256K 2026-07-17 — stable 128K ceiling |
-| `compression.threshold` | 0.65 | Triggers at ~85K |
+| `compression.threshold` | 0.60 | Triggers at ~77K (~60% of 128K) — optimized for 128K context |
 | `compression.target_ratio` | 0.3 | Compresses to ~38K |
 | `compression.protect_last_n` | 40 | Preserves last 40 messages |
 | `memory.memory_char_limit` | 800 | Brief memory entries |