feat(hindsight): cut over LLM model to Gemma-4-26B-A4B-it-AWQ (t_gemma4_swap)

astro-orbiter's vLLM primary model changed 2026-09-01 from
DeepSeek-R1-Distill-Qwen-32B-AWQ to Gemma-4-26B-A4B-it-AWQ (Google,
Apache 2.0, US-origin). Same endpoint/API key — only the served model
name changed. max_model_len also bumped to 65536 (was 32768).

Gemma 4 does not emit an always-on <think> reasoning trace like
DeepSeek-R1 did, so HINDSIGHT_API_RETAIN_MAX_COMPLETION_TOKENS=4096
should have more effective headroom for real extraction output than
before, not less.

cluster/applications/hindsight/values.yaml

@@ -75,23 +75,29 @@ api:
     # MODEL SWAP (t_r1d32b_swap, 2026-09-01): astro-orbiter's vLLM primary
     # model changed from Qwen2.5-32B-Instruct-AWQ to
     # DeepSeek-R1-Distill-Qwen-32B-AWQ (single-model deployment; nomic-embed
-    # and Qwen3-8B-AWQ both disabled on that host). Same endpoint
-    # (http://astro-orbiter:8000/v1), same API key (vLLM instance/port
-    # unchanged) -- only the served model name changed. DeepSeek-R1 emits
-    # a <think>...</think> reasoning trace before its final answer;
-    # Hindsight's fact-extraction prompt parsing has not been specifically
-    # audited against this reasoning-model output shape -- flag for
-    # follow-up if extraction quality regresses (see README note below).
+    # and Qwen3-8B-AWQ both disabled on that host). Superseded below.
+    #
+    # MODEL SWAP #2 (t_gemma4_swap, 2026-09-01): DeepSeek-R1-Distill-Qwen-32B
+    # retired after confirming its tool_choice="auto" reliability is a
+    # known, documented DeepSeek-R1-distillation limitation (upstream
+    # GitHub-confirmed: trained on pure reasoning traces, no function-
+    # calling data) -- not relevant to Hindsight's pure-text extraction
+    # use case, but disqualifying for agent-facing Hermes profiles, which
+    # drove the swap. Replaced with Gemma 4 26B A4B (Google, Apache 2.0,
+    # US-origin). Same endpoint (http://astro-orbiter:8000/v1), same API
+    # key -- only the served model name changed. Gemma 4 does NOT emit a
+    # reasoning trace by default (confirmed live) -- simpler completion
+    # parsing than DeepSeek-R1's always-on <think> blocks.
     HINDSIGHT_API_LLM_BASE_URL: "http://astro-orbiter:8000/v1"
     HINDSIGHT_API_LLM_PROVIDER: "openai"
-    HINDSIGHT_API_LLM_MODEL: "DeepSeek-R1-Distill-Qwen-32B-AWQ"
-    # vLLM's max_model_len is now 32768 (up from the prior 8192 role
-    # default) -- more headroom than before, but DeepSeek-R1 reasoning
-    # traces consume completion tokens before the final answer even
-    # appears, so the completion cap is left unchanged pending live
-    # verification that 4096 is enough for a real retain to complete
-    # (rather than truncating mid-<think>). Raise if extraction responses
-    # come back truncated/empty.
+    HINDSIGHT_API_LLM_MODEL: "Gemma-4-26B-A4B-it-AWQ"
+    # vLLM's max_model_len is now 65536 (up from DeepSeek's 32768, up from
+    # the original 8192 role default). Gemma 4's native context is 256K;
+    # 65536 is astro-orbiter's configured ceiling, comfortably above
+    # Hermes's 64K floor. Completion cap left at 4096 pending live
+    # verification -- Gemma 4 doesn't burn tokens on unwanted reasoning
+    # traces the way DeepSeek-R1 did, so 4096 should have MORE effective
+    # headroom for the actual extraction output than it did before.
     HINDSIGHT_API_RETAIN_MAX_COMPLETION_TOKENS: "4096"
     # DO NOT set HINDSIGHT_API_EMBEDDINGS_* here (t_e6facb19, 2026-08-31
     # attempted this, reverted after a production incident — see below).
@@ -130,8 +136,9 @@ api:
     # Kept at 1 post-cutover: vLLM's single-process-per-model design is also
     # effectively serial for a single generative model instance under this
     # GPU's VRAM budget (KV cache sized tight against the 24GB card at
-    # max_model_len=32768 for DeepSeek-R1-Distill-Qwen-32B-AWQ, t_r1d32b_swap
-    # 2026-09-01 — previously 8192 for Qwen2.5-32B-Instruct-AWQ).
+    # max_model_len=65536 for Gemma-4-26B-A4B-it-AWQ, t_gemma4_swap
+    # 2026-09-01 — previously 32768 for DeepSeek-R1-Distill-Qwen-32B-AWQ,
+    # previously 8192 for Qwen2.5-32B-Instruct-AWQ).
     HINDSIGHT_API_LLM_MAX_CONCURRENT: "1"
     HINDSIGHT_API_RETAIN_LLM_MAX_CONCURRENT: "1"
     # Client + per-request timeout. Default is 120s; a 29K-token retain runs