use supported TEI 1.9 options

servers/unraid/hindsight-retrieval/README.md

@@ -8,8 +8,8 @@ Arcane-managed TEI services on unRaid GPU0 for the production Hindsight VM.
 | Reranker | `Alibaba-NLP/gte-multilingual-reranker-base` | `http://192.168.50.100:18084` |
 
 The BGE-M3 cache is reused from the retired unRaid Hindsight stack. The GTE
-reranker has a separate cache. Both services use physical GPU0 and conservative
+reranker has a separate cache. Both services use physical GPU0 and the defaults
-TEI concurrency/tokenizer settings.
+provided by the pinned TEI 1.9 image.
 
 Hindsight uses the local embedding directly. Hindsight has no native embedding
 provider fallback chain, so OpenRouter BGE-M3 remains a manual rollback rather

servers/unraid/hindsight-retrieval/compose.yaml

@@ -9,17 +9,6 @@ services:
       - BAAI/bge-m3
       - --dtype
       - float16
-      - --max-input-length
-      - "8192"
-      - --max-client-batch-size
-      - "32"
-      - --max-concurrent-requests
-      - "64"
-      - --max-batch-tokens
-      - "8192"
-      - --tokenization-workers
-      - "4"
-      - --auto-truncate
     environment:
       HF_HUB_CACHE: /data
       TZ: America/Los_Angeles
@@ -55,17 +44,6 @@ services:
       - Alibaba-NLP/gte-multilingual-reranker-base
       - --dtype
       - float16
-      - --max-input-length
-      - "8192"
-      - --max-client-batch-size
-      - "32"
-      - --max-concurrent-requests
-      - "32"
-      - --max-batch-tokens
-      - "8192"
-      - --tokenization-workers
-      - "4"
-      - --auto-truncate
     environment:
       HF_HUB_CACHE: /data
       TZ: America/Los_Angeles