Merge cc7569dbdb into 2634ed207a

Without this I get:

> * Getting build dependencies for wheel...
> * Building wheel...
> Successfully built gguf-0.17.1-py3-none-any.whl
> Finished creating a wheel...
> Finished executing pypaBuildPhase
> Running phase: pythonRuntimeDepsCheckHook
> Executing pythonRuntimeDepsCheck
> Checking runtime dependencies for gguf-0.17.1-py3-none-any.whl
>   - requests not installed
For full logs, run:
    nix log /nix/store/x0c4a251l68bvdgang9d8v2fsmqay8a4-python3.12-gguf-0.0.0.drv

I changed a bit the style to make it more terse ~> more elegant in my
opinion.

.devops/nix/package-gguf-py.nix

@@ -3,6 +3,7 @@
   llamaVersion,
   numpy,
   tqdm,
+  requests,
   sentencepiece,
   pyyaml,
   poetry-core,
@@ -20,6 +21,7 @@ buildPythonPackage {
     tqdm
     sentencepiece
     pyyaml
+    requests
   ];
   src = lib.cleanSource ../../gguf-py;
   pythonImportsCheck = [

.devops/nix/scope.nix

@@ -7,13 +7,6 @@
 
 let
   pythonPackages = python3.pkgs;
-  buildPythonPackage = pythonPackages.buildPythonPackage;
-  numpy = pythonPackages.numpy;
-  tqdm = pythonPackages.tqdm;
-  sentencepiece = pythonPackages.sentencepiece;
-  pyyaml = pythonPackages.pyyaml;
-  poetry-core = pythonPackages.poetry-core;
-  pytestCheckHook = pythonPackages.pytestCheckHook;
 in
 
 # We're using `makeScope` instead of just writing out an attrset
@@ -23,17 +16,18 @@ in
 lib.makeScope newScope (self: {
   inherit llamaVersion;
   gguf-py = self.callPackage ./package-gguf-py.nix {
-    inherit
-      buildPythonPackage
+    inherit (pythonPackages)
       numpy
       tqdm
       sentencepiece
-      poetry-core
       pyyaml
       pytestCheckHook
+      requests
+      buildPythonPackage
+      poetry-core
       ;
   };
-  python-scripts = self.callPackage ./python-scripts.nix { inherit buildPythonPackage poetry-core; };
+  python-scripts = self.callPackage ./python-scripts.nix { inherit (pythonPackages) buildPythonPackage poetry-core; };
   llama-cpp = self.callPackage ./package.nix { };
   docker = self.callPackage ./docker.nix { };
   docker-min = self.callPackage ./docker.nix { interactive = false; };

docs/backend/SYCL.md

@@ -119,7 +119,7 @@ On older Intel GPUs, you may try [OpenCL](/docs/backend/OPENCL.md) although the
 *Notes:*
 
 - **Memory**
-  - The device memory is a limitation when running a large model. The loaded model size, *`llm_load_tensors: buffer_size`*, is displayed in the log when running `./bin/llama-cli`.
+  - The device memory is a limitation when running a large model. The loaded model size, *`llm_load_tensors: buffer_size`*, is displayed in the log when running `./bin/llama-completion`.
   - Please make sure the GPU shared memory from the host is large enough to account for the model's size. For e.g. the *llama-2-7b.Q4_0* requires at least 8.0GB for integrated GPU and 4.0GB for discrete GPU.
 
 - **Execution Unit (EU)**
@@ -423,16 +423,12 @@ Choose one of following methods to run.
 - Use device 0:
 
 ```sh
-./examples/sycl/run-llama2.sh 0
-# OR
-./examples/sycl/run-llama3.sh 0
+./examples/sycl/test.sh -mg 0
 ```
 - Use multiple devices:
 
 ```sh
-./examples/sycl/run-llama2.sh
-# OR
-./examples/sycl/run-llama3.sh
+./examples/sycl/test.sh
 ```
 
 2. Command line
@@ -455,13 +451,13 @@ Examples:
 - Use device 0:
 
 ```sh
-ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -no-cnv -m models/llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 99 -sm none -mg 0
+ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -no-cnv -m models/llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 99 -sm none -mg 0 --mmap
 ```
 
 - Use multiple devices:
 
 ```sh
-ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -no-cnv -m models/llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 99 -sm layer
+ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -no-cnv -m models/llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 99 -sm layer --mmap
 ```
 
 *Notes:*
@@ -577,13 +573,13 @@ Or, use CMake presets to build:
 
 ```sh
 cmake --preset x64-windows-sycl-release
-cmake --build build-x64-windows-sycl-release -j --target llama-cli
+cmake --build build-x64-windows-sycl-release -j --target llama-completion
 
 cmake -DGGML_SYCL_F16=ON --preset x64-windows-sycl-release
-cmake --build build-x64-windows-sycl-release -j --target llama-cli
+cmake --build build-x64-windows-sycl-release -j --target llama-completion
 
 cmake --preset x64-windows-sycl-debug
-cmake --build build-x64-windows-sycl-debug -j --target llama-cli
+cmake --build build-x64-windows-sycl-debug -j --target llama-completion
 ```
 
 #### 3. Visual Studio
@@ -608,7 +604,7 @@ You can use Visual Studio to open the `llama.cpp` folder directly as a CMake pro
 - For a minimal experimental setup, you can build only the inference executable using:
 
     ```Powershell
-    cmake --build build --config Release -j --target llama-cli
+    cmake --build build --config Release -j --target llama-completion
     ```
 
 ##### - Generating a Visual Studio Solution
@@ -714,13 +710,7 @@ Choose one of following methods to run.
 1. Script
 
 ```
-examples\sycl\win-run-llama-2.bat
-```
-
-or
-
-```
-examples\sycl\win-run-llama-3.bat
+examples\sycl\win-test.bat
 ```
 
 2. Command line
@@ -744,13 +734,13 @@ Examples:
 - Use device 0:
 
 ```
-build\bin\llama-cli.exe -no-cnv -m models\llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 99 -sm none -mg 0
+build\bin\llama-completion.exe -no-cnv -m models\llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 99 -sm none -mg 0 --mmap
 ```
 
 - Use multiple devices:
 
 ```
-build\bin\llama-cli.exe -no-cnv -m models\llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 99 -sm layer
+build\bin\llama-completion.exe -no-cnv -m models\llama-2-7b.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 99 -sm layer --mmap
 ```
 
 

examples/sycl/run-llama2.sh

@@ -18,13 +18,14 @@ CONTEXT=4096
 #support malloc device memory more than 4GB.
 export UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
 
+LOAD_MODE='--mmap'
 if [ $# -gt 0 ]; then
     GGML_SYCL_DEVICE=$1
     echo "use $GGML_SYCL_DEVICE as main GPU"
     #use signle GPU only
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT} -mg $GGML_SYCL_DEVICE -sm none
+    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT} -mg $GGML_SYCL_DEVICE -sm none ${LOAD_MODE}
 
 else
     #use multiple GPUs with same max compute units
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT}
+    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT} ${LOAD_MODE}
 fi

examples/sycl/run-llama3.sh

@@ -1,31 +0,0 @@
-#!/usr/bin/env bash
-
-#  MIT license
-#  Copyright (C) 2025 Intel Corporation
-#  SPDX-License-Identifier: MIT
-
-# If you want more control, DPC++ Allows selecting a specific device through the
-# following environment variable
-export ONEAPI_DEVICE_SELECTOR="level_zero:0"
-source /opt/intel/oneapi/setvars.sh
-
-#export GGML_SYCL_DEBUG=1
-
-#ZES_ENABLE_SYSMAN=1, Support to get free memory of GPU by sycl::aspect::ext_intel_free_memory. Recommended to use when --split-mode = layer.
-
-INPUT_PROMPT="Building a website can be done in 10 simple steps:\nStep 1:"
-MODEL_FILE=models/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf
-NGL=99 # Layers offloaded to the GPU. If the device runs out of memory, reduce this value according to the model you are using.
-CONTEXT=4096
-
-#support malloc device memory more than 4GB.
-export UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
-
-if [ $# -gt 0 ]; then
-    GGML_SYCL_DEVICE=$1
-    echo "Using $GGML_SYCL_DEVICE as the main GPU"
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT} -mg $GGML_SYCL_DEVICE -sm none
-else
-    #use multiple GPUs with same max compute units
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-completion -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONTEXT}
-fi

examples/sycl/test.sh

@@ -0,0 +1,130 @@
+#!/bin/bash
+
+#  MIT license
+#  Copyright (C) 2024 Intel Corporation
+#  SPDX-License-Identifier: MIT
+
+Help() {
+  cat << EOF
+Usage: $(basename "$0") [OPTIONS]
+
+This script processes files with specified options.
+
+Options:
+  -h, --help    Display this help message and exit.
+  -c, --context <value>    Set context length. Bigger need more memory.
+  -p, --promote <value>    Prompt to start generation with.
+  -m, --model   <value>    Full model file path.
+  -mg,--main-gpu <value>   Set main GPU ID (0 - n) for single GPU mode.
+  -sm,--split-mode <value> How to split the model across multiple GPUs, one of:
+                            - none: use one GPU only
+                            - layer (default): split layers and KV across GPUs
+                            - row: split rows across GPUs
+  -ngl,--n-gpu-layers <value>  Max. number of layers to store in VRAM (default: -1)
+  -lv,--log-verbosity <value>  Set the verbosity threshold. Messages with a higher verbosity will be
+                               ignored. Values:
+                                - 0: generic output
+                                - 1: error
+                                - 2: warning
+                                - 3: info
+                                - 4: debug
+
+
+EOF
+}
+
+BIN_FILE=./build/bin/llama-completion
+SEED=0
+GPUS_SETTING=""
+
+INPUT_PROMPT="Building a website can be done in 10 simple steps:\nStep 1:"
+MODEL_FILE=models/llama-2-7b.Q4_0.gguf
+NGL=99
+CONTEXT=4096
+GGML_SYCL_DEVICE=-1
+SPLIT_MODE=layer
+LOG_VERBOSE=3
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        -c|--context)
+            CONTEXT=$2
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -p|--promote)
+            # Option that is a simple flag (boolean)
+            INPUT_PROMPT="$2"
+            # Shift once to consume the option flag
+            shift
+            shift
+            ;;
+        -m|--model)
+            MODEL_FILE="$2"
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -mg|--main-gpu)
+            GGML_SYCL_DEVICE=$2
+            SPLIT_MODE=none
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -sm|--split-mode)
+            SPLIT_MODE=$2
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -ngl|--n-gpu-layers)
+            NGL=$2
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -lv|--log-verbosity)
+            LOG_VERBOSE=$2
+            # Shift twice to consume both the option flag and its value
+            shift
+            shift
+            ;;
+        -h|--help)
+            Help
+            exit 0
+            ;;
+        *)
+            # Handle unknown options or stop processing options
+            echo "Invalid option: $1"
+            # Optional: exit script or shift to treat remaining as positional args
+            exit 1
+            ;;
+    esac
+done
+
+
+
+source /opt/intel/oneapi/setvars.sh
+
+#export GGML_SYCL_DEBUG=1
+
+#ZES_ENABLE_SYSMAN=1, Support to get free memory of GPU by sycl::aspect::ext_intel_free_memory. Recommended to use when --split-mode = layer.
+
+#support malloc device memory more than 4GB.
+export UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
+echo "UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=${UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS}"
+
+if [ $GGML_SYCL_DEVICE -ne -1 ]; then
+    echo "Use $GGML_SYCL_DEVICE as main GPU"
+    #use signle GPU only
+    GPUS_SETTING="-mg $GGML_SYCL_DEVICE -sm ${SPLIT_MODE}"
+    export ONEAPI_DEVICE_SELECTOR="level_zero:${$GGML_SYCL_DEVICE}"
+    echo "ONEAPI_DEVICE_SELECTOR=${ONEAPI_DEVICE_SELECTOR}"
+else
+   echo "Use all Intel GPUs, including iGPU & dGPU"
+ fi
+
+echo "run cmd: ZES_ENABLE_SYSMAN=1 ${BIN_FILE} -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s ${SEED} -c ${CONTEXT} ${GPUS_SETTING} -lv ${LOG_VERBOSE}  --mmap "
+ZES_ENABLE_SYSMAN=1 ${BIN_FILE} -m ${MODEL_FILE} -no-cnv -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s ${SEED} -c ${CONTEXT} ${GPUS_SETTING} -lv ${LOG_VERBOSE} --mmap
+

examples/sycl/win-run-llama2.bat

@@ -7,5 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"
 
 :: support malloc device memory more than 4GB.
 set UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
-
-.\build\bin\llama-completion.exe -m models\llama-2-7b.Q4_0.gguf -no-cnv -p %INPUT2% -n 400 -e -ngl 99 -s 0
+set LOAD_MODE="--mmap"
+.\build\bin\llama-completion.exe -m models\llama-2-7b.Q4_0.gguf -no-cnv -p %INPUT2% -n 400 -e -ngl 99 -s 0 %LOAD_MODE%

examples/sycl/win-test.bat

@@ -7,5 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"
 
 :: support malloc device memory more than 4GB.
 set UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
-
-.\build\bin\llama-completion.exe -m models\Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf -no-cnv -p %INPUT2% -n 400 -s 0 -e -ngl 99
+set LOAD_MODE="--mmap"
+.\build\bin\llama-completion.exe -m models\llama-2-7b.Q4_0.gguf -no-cnv -p %INPUT2% -n 400 -e -ngl 99 -s 0 %LOAD_MODE%

tools/server/server-common.cpp

@@ -869,10 +869,15 @@ json oaicompat_chat_params_parse(
         if (response_type == "json_object") {
             json_schema = json_value(response_format, "schema", json::object());
         } else if (response_type == "json_schema") {
+            // https://platform.openai.com/docs/api-reference/chat/create#chat-create-response_format
+            // OpenAI expects: response_format.json_schema.schema
             auto schema_wrapper = json_value(response_format, "json_schema", json::object());
-            json_schema = json_value(schema_wrapper, "schema", json::object());
+            if (!schema_wrapper.contains("schema")) {
+                throw std::invalid_argument("response_format type \"json_schema\" requires \"json_schema.schema\" to be set");
+            }
+            json_schema = schema_wrapper.at("schema");
         } else if (!response_type.empty() && response_type != "text") {
-            throw std::invalid_argument("response_format type must be one of \"text\" or \"json_object\", but got: " + response_type);
+            throw std::invalid_argument("response_format type must be one of \"text\", \"json_object\", or \"json_schema\", but got: " + response_type);
         }
     }
 

tools/server/tests/unit/test_chat_completion.py

@@ -176,12 +176,18 @@ def test_apply_chat_template():
     ({"type": "json_object", "schema": {"const": "42"}}, 6, "\"42\""),
     ({"type": "json_object", "schema": {"items": [{"type": "integer"}]}}, 10, "[ -3000 ]"),
     ({"type": "json_schema", "json_schema": {"schema": {"const": "foooooo"}}}, 10, "\"foooooo\""),
+    # json_schema with name field (OpenAI-style)
+    ({"type": "json_schema", "json_schema": {"name": "test", "schema": {"const": "bar"}, "strict": True}}, 6, "\"bar\""),
     ({"type": "json_object"}, 10, "(\\{|John)+"),
     ({"type": "sound"}, 0, None),
     # invalid response format (expected to fail)
     ({"type": "json_object", "schema": 123}, 0, None),
     ({"type": "json_object", "schema": {"type": 123}}, 0, None),
     ({"type": "json_object", "schema": {"type": "hiccup"}}, 0, None),
+    # json_schema missing required json_schema.schema field (should fail)
+    ({"type": "json_schema", "json_schema": {"name": "test"}}, 0, None),
+    ({"type": "json_schema", "json_schema": {}}, 0, None),
+    ({"type": "json_schema"}, 0, None),
 ])
 def test_completion_with_response_format(response_format: dict, n_predicted: int, re_content: str | None):
     global server
@@ -203,6 +209,31 @@ def test_completion_with_response_format(response_format: dict, n_predicted: int
         assert "error" in res.body
 
 
+@pytest.mark.parametrize("response_format,expected_error_message", [
+    # json_schema type requires json_schema.schema to be set
+    ({"type": "json_schema", "json_schema": {"name": "test"}}, "json_schema.schema"),
+    ({"type": "json_schema", "json_schema": {}}, "json_schema.schema"),
+    ({"type": "json_schema"}, "json_schema.schema"),
+    # invalid response_format type should mention valid options
+    ({"type": "invalid_type"}, "json_schema"),
+])
+def test_response_format_error_messages(response_format: dict, expected_error_message: str):
+    """Test that invalid response_format configurations return helpful error messages."""
+    global server
+    server.start()
+    res = server.make_request("POST", "/chat/completions", data={
+        "max_tokens": 10,
+        "messages": [
+            {"role": "user", "content": "test"},
+        ],
+        "response_format": response_format,
+    })
+    assert res.status_code == 400
+    assert "error" in res.body
+    assert expected_error_message in res.body["error"]["message"], \
+        f"Expected '{expected_error_message}' in error message, got: {res.body['error']['message']}"
+
+
 @pytest.mark.parametrize("jinja,json_schema,n_predicted,re_content", [
     (False, {"const": "42"}, 6, "\"42\""),
     (True, {"const": "42"}, 6, "\"42\""),