Add model card

README.md

@@ -72,8 +72,7 @@ official Q4_K_M GGUF and reached **35.62 tokens/second** during the complete
 
 ## Runtime V3 patch notes
 
-Runtime V3 adds parser, grammar, and template portability fixes on top of the
-V2 Vulkan runtime:
+Runtime V3 adds parser and grammar fixes on top of the V2 Vulkan runtime:
 
 - ports [llama.cpp PR #24835](https://github.com/ggml-org/llama.cpp/pull/24835)
   so generated JSON values no longer carry trailing grammar whitespace that
@@ -81,19 +80,15 @@ V2 Vulkan runtime:
 - aligns PEG JSON-array comma whitespace handling with the generated grammar;
 - adds a Laguna/Pi regression for an `edit` call containing an
   `array<object>` argument and source-code strings;
-- removes the render-only `{% generation %}` extension from the Laguna S 2.1
-  template, preserving the rendered prompt while allowing standard Jinja and
-  MiniJinja engines such as Hipfire to parse it;
 - retains the V2 RADV DeviceLost safeguards and validated 128K defaults.
 
 Focused V3 validation on Ryzen AI Max+ 395 / Radeon 8060S with Mesa RADV
 26.1.2 passed JSON-schema grammar conversion, automatic parser selection, the
-nested Pi edit-call regression, the Laguna architecture test, and a strict
-standard-Jinja syntax parse. Real-model smokes returned `Paris.`, preserved the
-nested `edit` tool call, completed an 8,061-token functional smoke with the
-expected `omega` response, and released the slot after cancellation at 82% of
-a 15,000-token request. A follow-up request returned `Paris.` with no
-`VK_ERROR_DEVICE_LOST`.
+nested Pi edit-call regression, and the Laguna architecture test. Real-model
+smokes returned `Paris.`, preserved the nested `edit` tool call, completed an
+8,061-token functional smoke with the expected `omega` response, and released
+the slot after cancellation at 82% of a 15,000-token request. A follow-up
+request returned `Paris.` with no `VK_ERROR_DEVICE_LOST`.
 
 ### V2 Vulkan stability baseline
 
@@ -346,26 +341,6 @@ full-depth prefill, multi-turn, and cache-replay gates.
 The complete production recipe is preserved in the
 [ROCmFPX Laguna Runtime V3 guide](https://github.com/ciru-ai/ROCmFPX/blob/e4db35e027d0a9fa60e3324e4cc12f63ba7e0a00/docs/recipes/laguna-s21-chadrock-rocmfp4-strixkvspine-v4.md).
 
-## Hipfire template compatibility
-
-Hipfire releases that do not implement the Hugging Face `{% generation %}`
-extension reject the older embedded template and fall back to raw-encoding the
-user text, which produces incoherent output. Runtime V3's Laguna template
-avoids that extension.
-
-For an existing `.mq4` with the older embedded template, run this from the
-ROCmFPX repository root to override it without re-quantizing the model:
-
-```bash
-HIPFIRE_LOCAL=1 \
-HIPFIRE_CHAT_TEMPLATE_FILE="$PWD/models/templates/poolside-Laguna-S-2.1.jinja" \
-  hipfire run ~/.hipfire/models/laguna-s21.mq4 --max-tokens 64 \
-  "The capital of France is"
-```
-
-If a background Hipfire daemon is already running, stop it with `hipfire stop`
-before testing without `HIPFIRE_LOCAL=1`.
-
 ## Example request
 
 After starting the compatible server: