chore(llm): qwen2.5 model tiers, recency + retrieval knobs (refs #654)

docs/superpowers/specs/2026-07-16-llm-module-design.md

@@ -1,6 +1,6 @@
 # llm module — local RAG over rulemaking comments + Zotero corpus
 
-**Status:** Designed — awaiting implementation (milestones P33–P35)
+**Status:** Designed — awaiting implementation (milestones P33–P35). Superseded in part by `2026-09-03-llm-corpus-recency-links-gpu-design.md` (retrieval scope, ranking, evidence links, GPU routing).
 **Date:** 2026-07-16
 **Related:** #253 (extraction, closed), #254/#255/#256 (P30, open — consumers of
 this module), `.claude/specs/2026-04-23-comments-extraction-design.md`

stack.toml

@@ -104,7 +104,10 @@ timeout = 120.0
 [llm]
 ollama = "http://127.0.0.1:11434"     # host-side default; containers override via LLM_OLLAMA_HOSTS
 embed_model = "nomic-embed-text"       # 768-dim; bake-off (P33 #564) may revise
-instruct_model = "llama3.1:8b"         # bake-off may revise
+instruct_model = "qwen2.5:14b"         # baseline, pulled on every fleet host
+instruct_model_large = "qwen2.5:32b"   # used when the chosen host declares >= large_min_vram_gb
+large_min_vram_gb = 20
+chat_num_ctx = 8192                    # passed per request; no Modelfile ctx variants
 embed_dim = 768
 # HNSW index build. Safe since the postgres image ships an AVX-512-free
 # pgvector rebuild (infra/images/postgresql.Dockerfile, #580); before that it
@@ -114,6 +117,15 @@ pg_host = "127.0.0.1"                  # host-side default; containers set LLM_P
 pg_port = 5432
 pg_db = "llm"
 pg_user = "llm"
+# Retrieval: similarity × recency blend (half-life in days), sources kept
+recency_half_life_days = 365
+recency_weight = 0.3
+top_n = 8
+
+[llm.k_per_kind]                       # over-fetched ×3 per kind, then re-ranked
+comment = 8
+rule = 4
+corpus = 4
 
 [s3]
 endpoint = "http://rustfs:9000"