homelab

rblundon/homelab

Fork 0

Commit Graph

Author	SHA1	Message	Date
Hermes Agent service account	7b44a41da3	feat(llm): swap astro-orbiter primary model Qwen3.6 -> Qwen3.8-27B-Q4_K_M Ryan-directed model swap (kanban t_f5f7e9ad, 2026-08-16). Changes: - Replace [Qwen3.6-35B-A3B-UD-Q4_K_S] with [Qwen3.8-27B-Q4_K_M] in llama-server-router-preset.ini.j2 (production model slot). - Qwen3.8-27B: dense 27B VLM, Apache-2.0, Alibaba Aug 2026. Unsloth Dynamic V3.0 GGUF quantization. Q4_K_M chosen: 17,106,775,008 bytes, 17.1GB. Measured VRAM: 17,068 MiB at ctx=32768 (q4_0 KV cache). - ctx-size set to 32768 (32K) via new variable llm_router_qwen38_ctx_size. Native context is 262K; 32K chosen to maintain eviction headroom on 24GB RTX 3090. - models-max reduced 4 -> 2 in host_vars. Qwen3.8 (17.6GB) + nomic-embed (558MB) exhaust the 24GB card; no auxiliary model can co-reside with Qwen3.8. LRU eviction handles model switching with ~30-60s cold-load latency. - llm_router_expected_model_id updated to Qwen3.8-27B-Q4_K_M. - Qwen3.6 GGUF retained at /opt/models/Qwen3.6-35B-A3B-UD-Q4_K_S.gguf (not deleted — pending stable period and explicit cleanup task). - day2_swap_qwen38.yml playbook added for Ansible idempotent redeployment. Architecture note: Qwen3.8 uses Gated DeltaNet; llama.cpp 6ea215d logs 'fused Gated Delta Net (chunked) not supported, set to disabled'. Inference works correctly on the non-fused fallback. A llama.cpp update may improve throughput on the GDN layers. Smoke test passed: model responded via router endpoint (http://10.1.71.130:8002). VRAM: 17,630 MiB (Qwen3.8) + 5,928 MiB (Llama-8B concurrent) = 23,558 MiB. Also commits accumulated but unpushed changes: - nomic-embed batch-size/rope-scaling fix (t_openviking_embed_batch) - per-model ctx-size day2 playbook (day2_per_model_ctx_size.yml) - llama-server-router.service.j2 minor update	2026-08-16 20:40:31 -05:00
Hermes Agent service account	ad70b3439c	feat(llm): add nomic-embed-text-v1.5-Q4_K_M to astro-orbiter router OpenViking Phase 1b (t_34b96e83) — Ryan-approved implementation. Changes: - roles/llm-inference-multimodel/templates/llama-server-router-preset.ini.j2: Add [nomic-embed-text-v1.5] section with embedding=true, n-gpu-layers=99, ctx-size=8192, load-on-startup=true, sleep-idle-seconds=-1. No flash-attn or KV cache params (embedding models use bidirectional forward pass, not autoregressive KV cache). Var: llm_router_nomic_ctx_size. - roles/llm-inference-multimodel/defaults/main.yml: Add llm_router_nomic_ctx_size: 8192. - host_vars/astro-orbiter/vars.yml: Add nomic-embed-text-v1.5-Q4_K_M.gguf to llm_staged_models list (size_bytes: 84106624, source: nomic-ai/nomic-embed-text-v1.5-GGUF). Update VRAM note to reflect 5 registered models (nomic adds ~84MB, negligible given sleep-idle-seconds=-1 / load-on-startup=true pinning). - playbooks/day2_add_nomic_embed.yml: New day2 playbook following the coder-alias pattern: Phase 1: idempotent GGUF download (exact size check) Phase 2: redeploy preset INI Phase 3: redeploy + restart systemd unit Phase 4: /v1/models gate (all 5 models present) Phase 5: /v1/embeddings smoke test (vector returned, not empty) VRAM: ~84MB, always pinned. No impact on generative model LRU behavior. peter-parker Helm values already point at :8002 for the embedding endpoint.	2026-08-13 23:18:27 -05:00

Author

SHA1

Message

Date

Hermes Agent service account

7b44a41da3

feat(llm): swap astro-orbiter primary model Qwen3.6 -> Qwen3.8-27B-Q4_K_M

Ryan-directed model swap (kanban t_f5f7e9ad, 2026-08-16).

Changes:
- Replace [Qwen3.6-35B-A3B-UD-Q4_K_S] with [Qwen3.8-27B-Q4_K_M] in
  llama-server-router-preset.ini.j2 (production model slot).
- Qwen3.8-27B: dense 27B VLM, Apache-2.0, Alibaba Aug 2026.
  Unsloth Dynamic V3.0 GGUF quantization.
  Q4_K_M chosen: 17,106,775,008 bytes, 17.1GB. Measured VRAM: 17,068 MiB
  at ctx=32768 (q4_0 KV cache).
- ctx-size set to 32768 (32K) via new variable llm_router_qwen38_ctx_size.
  Native context is 262K; 32K chosen to maintain eviction headroom on 24GB RTX 3090.
- models-max reduced 4 -> 2 in host_vars. Qwen3.8 (17.6GB) + nomic-embed
  (558MB) exhaust the 24GB card; no auxiliary model can co-reside with Qwen3.8.
  LRU eviction handles model switching with ~30-60s cold-load latency.
- llm_router_expected_model_id updated to Qwen3.8-27B-Q4_K_M.
- Qwen3.6 GGUF retained at /opt/models/Qwen3.6-35B-A3B-UD-Q4_K_S.gguf
  (not deleted — pending stable period and explicit cleanup task).
- day2_swap_qwen38.yml playbook added for Ansible idempotent redeployment.

Architecture note: Qwen3.8 uses Gated DeltaNet; llama.cpp 6ea215d logs
'fused Gated Delta Net (chunked) not supported, set to disabled'. Inference
works correctly on the non-fused fallback. A llama.cpp update may improve
throughput on the GDN layers.

Smoke test passed: model responded via router endpoint (http://10.1.71.130:8002).
VRAM: 17,630 MiB (Qwen3.8) + 5,928 MiB (Llama-8B concurrent) = 23,558 MiB.

Also commits accumulated but unpushed changes:
- nomic-embed batch-size/rope-scaling fix (t_openviking_embed_batch)
- per-model ctx-size day2 playbook (day2_per_model_ctx_size.yml)
- llama-server-router.service.j2 minor update

2026-08-16 20:40:31 -05:00

Hermes Agent service account

ad70b3439c

feat(llm): add nomic-embed-text-v1.5-Q4_K_M to astro-orbiter router

OpenViking Phase 1b (t_34b96e83) — Ryan-approved implementation.

Changes:
- roles/llm-inference-multimodel/templates/llama-server-router-preset.ini.j2:
  Add [nomic-embed-text-v1.5] section with embedding=true, n-gpu-layers=99,
  ctx-size=8192, load-on-startup=true, sleep-idle-seconds=-1. No flash-attn
  or KV cache params (embedding models use bidirectional forward pass, not
  autoregressive KV cache). Var: llm_router_nomic_ctx_size.

- roles/llm-inference-multimodel/defaults/main.yml:
  Add llm_router_nomic_ctx_size: 8192.

- host_vars/astro-orbiter/vars.yml:
  Add nomic-embed-text-v1.5-Q4_K_M.gguf to llm_staged_models list
  (size_bytes: 84106624, source: nomic-ai/nomic-embed-text-v1.5-GGUF).
  Update VRAM note to reflect 5 registered models (nomic adds ~84MB,
  negligible given sleep-idle-seconds=-1 / load-on-startup=true pinning).

- playbooks/day2_add_nomic_embed.yml:
  New day2 playbook following the coder-alias pattern:
  Phase 1: idempotent GGUF download (exact size check)
  Phase 2: redeploy preset INI
  Phase 3: redeploy + restart systemd unit
  Phase 4: /v1/models gate (all 5 models present)
  Phase 5: /v1/embeddings smoke test (vector returned, not empty)

VRAM: ~84MB, always pinned. No impact on generative model LRU behavior.
peter-parker Helm values already point at :8002 for the embedding endpoint.

2026-08-13 23:18:27 -05:00

2 Commits