docs: name the first upstream llama.cpp release for Q2_0 (b11556)

AGENTS.md

@@ -23,15 +23,15 @@ The 27B generation is a step change over the earlier 8B/4B/1.7B demos:
 
 | `BONSAI_FAMILY` / `BONSAI_MODEL` | Weights | Notes |
 |---|---|---|
-| `bonsai2` / `27B` (**default**) | GGUF 5.9-7.2 GB + mmproj 0.63 GB; MLX 2-bit 8.6 GB | The current generation and what plain `./setup.sh` installs. Two GGUF bands, **both requiring this demo's binaries**: `PTQ1_0` (1.75 bpw, 5.9 GB, densely packed trits, smallest) and `PQ2_0` (2.13 bpw, 7.2 GB, faster prompt processing, what the scripts download). Q2_0 is supported by current official llama.cpp with Bonsai 2 integration. MLX uses native Hadamard-aware mlx-vlm 0.7.8 from `.venv-vlm` for one-shot generation and serving, not the fork in `.venv` |
+| `bonsai2` / `27B` (**default**) | GGUF 5.9-7.2 GB + mmproj 0.63 GB; MLX 2-bit 8.6 GB | The current generation and what plain `./setup.sh` installs. Two GGUF bands, **both requiring this demo's binaries**: `PTQ1_0` (1.75 bpw, 5.9 GB, densely packed trits, smallest) and `PQ2_0` (2.13 bpw, 7.2 GB, faster prompt processing, what the scripts download). Q2_0 is supported by official llama.cpp b11556 or newer. MLX uses native Hadamard-aware mlx-vlm 0.7.8 from `.venv-vlm` for one-shot generation and serving, not the fork in `.venv` |
 | `ternary` / `27B` (previous generation) | GGUF ~6.7-7.1 GB + mmproj 0.9 GB; MLX 2-bit ~7.9 GB | Superseded by `bonsai2`. Two GGUF formats since the mainline rebase (`prism-b10658+`): `PQ2_0` (group 128, 6.66 GiB, smallest/fastest where supported: CUDA, Metal, CPU, ROCm) and official `Q2_0` group 64 (`Ternary-Bonsai-27B-Q2_g64.gguf`, 7.05 GiB, adds Vulkan/SYCL); smaller sizes use `*-Q2_0_g64.gguf` naming. The scripts pick per backend. Legacy `*-Q2_0.gguf` files (no `g64`) only load on old `prism-v5` releases; new binaries refuse them with an error |
 | `bonsai` / `27B` | GGUF Q1_0 ~3.5 GB + mmproj 0.9 GB; MLX 1-bit ~4.8 GB | Smallest and fastest; fits on a modern iPhone without offloading |
 | `8B` / `4B` / `1.7B` (both families) | smaller | Text-only, no tools wiring, legacy tested flag set |
 
 ### Current runtime support
 
-Current official llama.cpp supports Bonsai 2 in Q2_0 with the merged model and
-Hadamard/sign-flip integration. PQ2_0 and PTQ1_0 still need our fork. Older engines
+Official llama.cpp supports Bonsai 2 in Q2_0 from release b11556, which merged the
+model and Hadamard/sign-flip integration. PQ2_0 and PTQ1_0 still need our fork. Older engines
 may recognize Q2_0 but miss the transforms; check the build when diagnosing gibberish.
 See [upstream support](README.md#upstream-support) instead of relying on old PR lists.
 

BACKEND-SUPPORT.md

@@ -89,8 +89,8 @@ The demo setup scripts do not automatically select or install SYCL yet.
 
 ## Model files and upstream compatibility
 
-- **Q2_0:** works with current official llama.cpp containing the merged Bonsai 2
-  model/Hadamard integration. Older bundled engines need updating; recognizing
+- **Q2_0:** works with official llama.cpp b11556 or newer, which contains the merged
+  Bonsai 2 model/Hadamard integration. Older bundled engines need updating; recognizing
   the type alone did not guarantee correct output in older builds.
 - **PQ2_0 and PTQ1_0:** remain fork-specific. PQ2_0 is this demo's download default.
 - **MTP/DFlash:** compatibility is checkpoint/runtime-specific; some paths still

MODEL-FORMATS.md

@@ -36,7 +36,7 @@ and the fork's group-128 layout lives under its own name and id, PQ2_0 (142).
 
 | Packing | Runtime | Trade-off |
 |---|---|---|
-| `Q2_0` | Current official llama.cpp with Bonsai 2 support, or our fork | Upstream format |
+| `Q2_0` | Official llama.cpp b11556 or newer, or our fork | Upstream format |
 | `PQ2_0` | PrismML fork | Faster prompt processing than PTQ1_0 on many devices |
 | `PTQ1_0` | PrismML fork | Smallest packed files |
 

README.md

@@ -46,7 +46,7 @@ model's card for its specifications:
   faster prompt processing and is what this demo downloads by default.
   See [MODEL-FORMATS.md](MODEL-FORMATS.md).
 
-**Official `Q2_0` works with current upstream llama.cpp containing Bonsai 2 support.**
+**Official `Q2_0` works with upstream llama.cpp [b11556](https://github.com/ggml-org/llama.cpp/releases/tag/b11556) or newer.**
 `PQ2_0` and `PTQ1_0` still require the [PrismML fork](https://github.com/PrismML-Eng/llama.cpp).
 This demo downloads `PQ2_0` and the matching fork binaries by default.
 All three packings contain the same model weights, with different memory/speed trade-offs.
@@ -155,7 +155,7 @@ See [community-benchmarks/](community-benchmarks/) for results on different hard
 
 ### Bonsai 2 (ternary, default)
 
-Available in GGUF and MLX 2-bit formats. Use current official llama.cpp for `Q2_0`,
+Available in GGUF and MLX 2-bit formats. Use official llama.cpp b11556 or newer for `Q2_0`,
 or our fork for `PQ2_0` / `PTQ1_0`. Setup installs the fork for its default `PQ2_0` download.
 
 | Model | GGUF | MLX (2-bit) |
@@ -201,12 +201,13 @@ Every launcher is configured through environment variables. The most common ones
 
 ## Upstream support
 
-- **llama.cpp:** Bonsai 2 works in official `Q2_0` format with current upstream
-  builds containing the merged model/Hadamard integration. Use a recent version;
-  older engines may recognize the format but miss the required transforms.
-- **PrismML fork:** still required for `PQ2_0` and `PTQ1_0`. Some MTP/DFlash
-  configurations and other experimental features may also need our fork; follow
-  the instructions for the specific checkpoint and runtime.
+- **llama.cpp:** Bonsai 2 (2B, 4B, 9B and 27B) works in official `Q2_0` format on
+  upstream llama.cpp from release [b11556](https://github.com/ggml-org/llama.cpp/releases/tag/b11556) onward, which merged the
+  model/Hadamard integration. Older engines may recognize the format but miss the
+  required transforms.
+- **PrismML fork:** still required for `PQ2_0` and `PTQ1_0`, and for the MTP and
+  DFlash2 drafters on the model cards. Other experimental features may also need
+  our fork; follow the instructions for the specific checkpoint and runtime.
 - **MLX:** the 1-bit support and Hadamard changes have merged upstream. Support
   should reach mlx-lm and other MLX engines as they integrate the model support
   and ship updated runtimes. A core MLX merge alone does not make every engine

community-benchmarks/bonsai2/README.md

@@ -61,9 +61,9 @@ Benchmark results submitted by the community running
    The default family is Bonsai 2; setup downloads `PQ2_0` and the required fork binaries.
    For `PTQ1_0`, download it from the model repository above and pass its path to
    `llama-bench -m` (`BONSAI_MODEL` selects a size, not a file path).
-   The development [Q2_0 file](https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf-dev/blob/main/Ternary-Bonsai-2-27B-Q2_0-prism-fork-required.gguf) is also welcome for benchmarking via `-m`.
-   It uses the official llama.cpp `Q2_0` format but **currently requires our fork**
-   for Bonsai 2's Hadamard transform support. Upstream support is pending our PRs.
+   The [Q2_0 file](https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/main/Ternary-Bonsai-2-27B-Q2_0.gguf) is also welcome for benchmarking via `-m`.
+   It uses the official llama.cpp `Q2_0` format and runs on our fork or on upstream
+   llama.cpp [b11556](https://github.com/ggml-org/llama.cpp/releases/tag/b11556) or newer.
 2. Copy [TEMPLATE-llama-cpp.md](TEMPLATE-llama-cpp.md) to
    `<backend>-<hardware>-<os>.md` here (lowercase, dashes). Keep tested formats in
    the same machine report, with separate commands and raw results for each.

community-benchmarks/bonsai2/TEMPLATE-llama-cpp.md

@@ -54,14 +54,13 @@ Use the same settings when comparing packings, and list any differences.
 
 (Paste the exact command and raw results, or state that this packing was not tested.)
 
-### Q2_0 (optional, development)
+### Q2_0 (optional)
 
-Use [Ternary-Bonsai-2-27B-Q2_0-prism-fork-required.gguf](https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf-dev/blob/main/Ternary-Bonsai-2-27B-Q2_0-prism-fork-required.gguf)
+Use [Ternary-Bonsai-2-27B-Q2_0.gguf](https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/blob/main/Ternary-Bonsai-2-27B-Q2_0.gguf)
 and pass its path to `-m` with the same benchmark settings.
-This is the official llama.cpp `Q2_0` format, but Bonsai 2 currently requires our
-fork's Hadamard transform support. Upstream support is pending our PRs; stock
-llama.cpp may load the file but produce incorrect output without those changes.
-Record the exact fork build and model revision or hash.
+This is the official llama.cpp `Q2_0` format. It runs on our fork and on upstream
+llama.cpp from release [b11556](https://github.com/ggml-org/llama.cpp/releases/tag/b11556) onward; older upstream builds load it but produce
+incorrect output. Record the exact build (fork or upstream) and model revision or hash.
 
 (Paste the exact command and raw results, or state that this format was not tested.)