feat(mistral-small-4): pin v0.22.0 for working VISION baseline + reasoning entry

Operator needs a verified-working vision tower as the abliteration/tuning
baseline. vLLM 0.23.0 crashes Mistral multimodal at startup (#44911
fetch_images regression, ~0.22.1+). Pinned the Mistral container to
v0.22.0 — the last pre-regression release — which loads the NVFP4
(compressed-tensors) AND serves vision: verified a half-blue/half-red
image read correctly ('left blue, right red'). Dropped --limit-mm
(vision re-enabled). qwen36 stays on 0.23.0 (separate container; needs it
for its ModelOpt NVFP4).

- gateway: add mistral-small-4-reasoning. Operator asked for effort=medium
  but Mistral's reasoning_effort is BINARY (none/high only — medium 400s);
  set to 'high' (sole reasoning-ON level). NOTE: reasoning fires but
  reasoning_content-splitting is unreliable on v0.22.0 (lands in content);
  clean split would need 0.23.0, which breaks vision — vision prioritized.
- mistral-small-4 (instant) + mistral-small-4-reasoning both gateway-live.

stacks/litellm/conf/config.yaml

@@ -80,6 +80,21 @@ model_list:
     model_info:
       mode: chat
 
+  # mistral-small-4-reasoning: same upstream checkpoint, reasoning ON (2026-06-15,
+  #   operator wanted "medium" — but Mistral's reasoning_effort is BINARY, only
+  #   'none' or 'high' (a medium/low request 400s). 'high' is the sole reasoning-ON
+  #   level, so it carries the -reasoning intent. LiteLLM forwards extra_body to
+  #   vLLM; the mistral reasoning-parser splits [THINK]…[/THINK] into reasoning_content.
+  - model_name: mistral-small-4-reasoning
+    litellm_params:
+      model: hosted_vllm/mistral-small-4
+      api_base: http://10.250.50.54:8010/v1
+      api_key: os.environ/VLLM_API_KEY
+      extra_body:
+        reasoning_effort: high
+    model_info:
+      mode: chat
+
   # --- Qwen3 embeddings ---
   - model_name: qwen3-embedding
     litellm_params:

stacks/mistral-small-4/.env.example

@@ -3,11 +3,15 @@
 #
 # See compose.yaml header for the NVFP4/TP=1/MLA rationale and the vLLM>=0.20 floor.
 
-# PINNED by digest, not :latest — this model is version-sensitive (needs vLLM
-# >= 0.20 for day-0 support; the related nvidia-ModelOpt NVFP4 MoE path broke on
-# 0.19.1/0.22.0). Pin protects against a :latest regression. This digest = vLLM
-# 0.23.0, the version validated to load this checkpoint. Bump deliberately.
-MISTRAL_IMAGE=vllm/vllm-openai@sha256:6d8429e38e3747723ca07ee1b17972e09bb9c51c4032b266f24fb1cc3b22ed8f
+# PINNED to v0.22.0 — the last release BEFORE the Mistral multimodal regression
+# (#44911, MistralCommonImageProcessor.fetch_images, landed ~0.22.1; 0.23.0 is
+# affected). v0.22.0 loads the NVFP4 (compressed-tensors) AND serves VISION —
+# verified: half-blue/half-red image read correctly ("left blue, right red").
+# This gives a working vision tower as the abliteration/tuning baseline. Do NOT
+# bump to 0.23.0 (breaks vision). reasoning_effort works (none/high only) but
+# reasoning_content-splitting is unreliable on this version — vision is the
+# priority. Revisit when vLLM patches the Mistral mm path on a newer release.
+MISTRAL_IMAGE=vllm/vllm-openai:v0.22.0
 
 MISTRAL_CONTAINER_NAME=vllm-mistral4
 MISTRAL_MODEL=mistralai/Mistral-Small-4-119B-2603-NVFP4

stacks/mistral-small-4/compose.yaml

@@ -74,15 +74,14 @@ services:
       - mistral
       - --max-num-seqs
       - ${MISTRAL_MAX_NUM_SEQS}
-      # TEXT-ONLY (2026-06-15): vLLM 0.23.0's Mistral multimodal processor crashes
-      # at startup dummy-image profiling — `MistralCommonImageProcessor has no
-      # attribute fetch_images` (vLLM↔mistral_common incompat; same class hit
-      # Mistral-3.1/Devstral/Magistral). Setting image/video limit to 0 skips the
-      # vision profiling so the model loads text-only — which is all the creative-
-      # writing use needs. REMOVE this flag to restore vision once vLLM patches the
-      # Mistral mm path (track: the model is natively multimodal).
-      - --limit-mm-per-prompt
-      - '{"image":0,"video":0}'
+      # VISION ENABLED. vLLM is pinned to v0.22.0 in .env — the last release BEFORE
+      # the Mistral multimodal regression (#44911, `MistralCommonImageProcessor has
+      # no attribute fetch_images`, landed ~0.22.1+; 0.23.0 is affected). v0.22.0
+      # still has Mistral-Small-4 arch + compressed-tensors NVFP4 support (the
+      # #44081 ModelOpt-NVFP4 bug on 0.22.0 is a DIFFERENT quant path, doesn't touch
+      # this compressed-tensors checkpoint). Gives a verified working vision tower
+      # as the abliteration/tuning baseline. (qwen36 stays on 0.23.0 — separate
+      # container; it NEEDS 0.23.0 for its ModelOpt NVFP4.)
       - --dtype
       - auto
       - --enable-prefix-caching