nyo16 · nyo16 · Jun 22, 2026 · Jun 22, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,22 @@
 # Changelog
 
+## v0.8.28
+
+### Changed
+
+- **llama.cpp submodule** — Updated from 845282461 to dec5ca557 (24 commits, tag b9763). No NIF changes were required. `ggml/include/ggml.h`, `ggml/include/ggml-backend.h`, `common/chat.h`, `common/json-schema-to-grammar.h`, `common/sampling.h`, `common/speculative.h`, and `common/common.h` are all unchanged. The only touched header the binding compiles against is `include/llama.h`, and the sole functional change there is one new accessor — `llama_model_n_layer_nextn` (the number of next-token / MTP prediction layers) — added alongside the existing `llama_model_n_layer`; the rest of the diff is whitespace realignment. The binding does not call the new function. Notably, this range refactors the grammar generators that the binding's `json_schema_to_grammar_nif` links against (the `common/peg` AC parser and the JSON-schema-to-GBNF spacing rules below) even though `common/json-schema-to-grammar.h` itself is unchanged, so both the JSON-schema and raw-GBNF smoke paths were exercised against the freshly built NIF. The full test suite passes (158 tests + 4 skipped), all 7 end-to-end smoke tests pass (generation, streaming, chat templates, JSON-schema grammar, raw GBNF, and embeddings — the embedding paths exercised with a Qwen3-Embedding model), formatting is clean, and Dialyzer reports 0 errors.
+  - **model/quantization**: use `LLM_KV` for `quantization_version` & `file_type` (#24802).
+  - **common/grammar**: implement an AC parser for stricter grammar generation (#24869); refactor the until→GBNF grammar generation (#24839); align `json-schema-to-grammar` spacing rules with the parsers (#24835) — these underlie the binding's structured-output / GBNF path.
+  - **sampling**: remove the unconditional softmax+sort in the top-n-sigma sampler (#22645).
+  - **jinja** (template engine used by chat templates): implement the `call` statement (#24847).
+  - **MTP/speculative**: support Step3.5/3.7 flash mtp3 (#24340).
+  - **mtmd**: fix `mtmd_get_memory_usage` (#24867); add a load-progress callback (#24865).
+  - **server**: add an `id` to tool-call responses API (#24882); move model downloading to a dedicated process in the router (#24834); refactor/generalize the input-file schema (#24299); fix an `edit_file` crash on append at end of file (`line_start` -1) (#24893); report progress for loading spec models and add a "stages" list (#24870); refactor batch construction (#24843); add a "verbose" field to the schema (#24864); real-time model load-progress tracking via `/models/sse` (#24828).
+  - **SYCL**: support bf16 on the `bin_bcast` op and unary ops (#24838).
+  - **hexagon**: use a padded stride for ssm-conv weights (#24470).
+  - **webui**: prioritize favorite models in model selection (#24766); show model status and load progress via the `/models/sse` feed (#24878).
+  - **common/cli/build/docs**: stabilize the randomly-failing `test-args-parser` (#24826); add the `libandroid-spawn` dependency for Termux builds (#21812); whitespace clean-up (#24862).
+
 ## v0.8.27
 
 ### Changed

diff --git a/mix.exs b/mix.exs
@@ -37,7 +37,7 @@ end
 defmodule LlamaCppEx.MixProject do
   use Mix.Project
 
-  @version "0.8.27"
+  @version "0.8.28"
   @source_url "https://github.com/nyo16/llama_cpp_ex"
 
   def project do

diff --git a/vendor/llama.cpp b/vendor/llama.cpp
+13 −7		common/arg.cpp
+5 −1		common/arg.h
+5 −4		common/chat-auto-parser-generator.cpp
+90 −47		common/jinja/runtime.cpp
+1 −0		common/jinja/runtime.h
+23 −23		common/json-schema-to-grammar.cpp
+194 −81		common/peg-parser.cpp
+16 −3		common/peg-parser.h
+102 −35		common/speculative.cpp
+1 −1		docs/android.md
+21 −21		examples/json_schema_to_grammar.py
+10 −9		ggml/src/ggml-hexagon/htp/ssm-conv.c
+5 −0		ggml/src/ggml-sycl/binbcast.cpp
+155 −53		ggml/src/ggml-sycl/element_wise.cpp
+9 −8		include/llama.h
+8 −0		src/llama-context.cpp
+1 −0		src/llama-context.h
+2 −0		src/llama-cparams.h
+5 −0		src/llama-ext.h
+9 −2		src/llama-graph.h
+4 −0		src/llama-model.cpp
+2 −2		src/llama-quant.cpp
+0 −2		src/llama-sampler.cpp
+28 −29		src/models/step35.cpp
+148 −1		tests/peg-parser/test-gbnf-generation.cpp
+11 −1		tests/test-arg-parser.cpp
+2 −2		tests/test-chat.cpp
+26 −0		tests/test-jinja.cpp
+155 −155		tests/test-json-schema-to-grammar.cpp
+2 −2		tests/test-sampling.cpp
+59 −24		tools/mtmd/clip.cpp
+2 −0		tools/mtmd/clip.h
+8 −1		tools/mtmd/mtmd.cpp
+8 −0		tools/mtmd/mtmd.h
+3 −3		tools/server/README-dev.md
+42 −7		tools/server/README.md
+37 −25		tools/server/server-common.cpp
+637 −320		tools/server/server-context.cpp
+3 −1		tools/server/server-context.h
+245 −131		tools/server/server-models.cpp
+22 −6		tools/server/server-models.h
+3 −0		tools/server/server-schema.cpp
+6 −3		tools/server/server-task.cpp
+10 −5		tools/server/server-tools.cpp
+12 −1		tools/server/server.cpp
+20 −0		tools/server/tests/unit/test_chat_completion.py
+28 −7		tools/server/tests/unit/test_router.py
+10 −0		tools/ui/src/app.d.ts
+12 −3		tools/ui/src/lib/components/app/chat/ChatMessages/ChatMessage/ChatMessageAssistant/ChatMessageAssistant.svelte
+18 −1		tools/ui/src/lib/components/app/models/ModelsSelectorOption.svelte
+2 −1		tools/ui/src/lib/constants/api-endpoints.ts
+2 −0		tools/ui/src/lib/constants/index.ts
+14 −0		tools/ui/src/lib/constants/model-loading.ts
+16 −0		tools/ui/src/lib/constants/sse.ts
+1 −1		tools/ui/src/lib/enums/index.ts
+14 −0		tools/ui/src/lib/enums/server.enums.ts
+9 −7		tools/ui/src/lib/services/chat.service.ts
+245 −39		tools/ui/src/lib/stores/models.svelte.ts
+47 −1		tools/ui/src/lib/types/api.d.ts
+10 −1		tools/ui/src/lib/types/index.ts
+12 −1		tools/ui/src/lib/types/models.d.ts
+3 −0		tools/ui/src/lib/utils/index.ts
+43 −0		tools/ui/src/lib/utils/progress.ts
+14 −0		tools/ui/src/routes/+layout.svelte