id stringlengths 16 16 | text stringlengths 351 4.11k | source dict | provenance unknown | stats unknown | license dict |
|---|---|---|---|---|---|
a96e1380af54add9 | # Android
## Build GUI binding using Android Studio
Import the `examples/llama.android` directory into Android Studio, then perform a Gradle sync and build the project.
This Android binding supports hardware acceleration up to `SME2`... | {
"path": "docs/android.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"sha256": "bc66f29afc8e081714be092363fdc64522933... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Android",
"line_start": 2,
"line_end": 22,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Android",
"Android > Build GUI binding using Android Studio"
]
} | {
"est_tokens": 516,
"chars": 2204,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
477ad045a927001d | ## Build CLI on Android using Termux
[Termux](https://termux.dev/en/) is an Android terminal emulator and Linux environment app (no root required). As of writing, Termux is available experimentally in the Google Play Store; otherwise, it may be obtained directly from the project repo or on F-Droid.
With Termux, you c... | {
"path": "docs/android.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"sha256": "bc66f29afc8e081714be092363fdc64522933... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Android > Build CLI on Android using Termux",
"line_start": 24,
"line_end": 53,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 358,
"chars": 1443,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
d5d2e67af4e71451 | ## Cross-compile CLI using Android NDK
It's possible to build `llama.cpp` for Android on your host system via CMake and the Android NDK. If you are interested in this path, ensure you already have an environment prepared to cross-compile programs for Android (i.e., install the Android SDK/NDK and set `ANDROID_NDK` to t... | {
"path": "docs/android.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"sha256": "bc66f29afc8e081714be092363fdc64522933... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Android > Cross-compile CLI using Android NDK",
"line_start": 55,
"line_end": 105,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 597,
"chars": 2423,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
74c29a7f9eba3fce | Be aware that Android will not find the library path `lib` on its own, so we must specify `LD_LIBRARY_PATH` in order to run the installed executables. Android does support `RPATH` in later API levels, so this could change in the future. Refer to the previous section for information about `context-size` (very important!... | {
"path": "docs/android.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/android.md",
"sha256": "bc66f29afc8e081714be092363fdc64522933... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Android > Cross-compile CLI using Android NDK",
"line_start": 107,
"line_end": 107,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 88,
"chars": 351,
"token_method": "chars/4 estimate (whitespace-normalised)",
"short": true
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
d41c60a4ad16e771 | # Auto-Parser Architecture
The auto-parser automatically analyzes chat templates to determine how to parse model outputs, including content, reasoning, and tool calls.
## Overview
The unified auto-parser uses a pure differential, compositional approach (inspired by the `git diff` algorithm) to analyze chat templates... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture",
"line_start": 1,
"line_end": 17,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture",
"Auto-Parser Architecture > Overview"
]
... | {
"est_tokens": 248,
"chars": 1000,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
fe0cb1204bc3f557 | ## Data Structures
All structs are defined in [common/chat-auto-parser.h](common/chat-auto-parser.h).
### Top-Level: `autoparser` (main analyzer and generator)
[common/chat-auto-parser.h:367-388](common/chat-auto-parser.h#L367-L388) — top-level analysis result aggregating `jinja_caps`, `reasoning`, `content`, and `t... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Data Structures",
"line_start": 19,
"line_end": 29,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Data Structures",
"Auto... | {
"est_tokens": 150,
"chars": 607,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
e9e9218ffcdf997e | ### `analyze_content`
[common/chat-auto-parser.h:280-295](common/chat-auto-parser.h#L280-L295) — content analysis result: `mode` enum, `start`/`end` markers, and `requires_nonnull_content` flag.
### `analyze_tools` and its sub-structs
- [common/chat-auto-parser.h:176-194](common/chat-auto-parser.h#L176-L194) — `tool... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Data Structures > `analyze_content`",
"line_start": 31,
"line_end": 41,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Data St... | {
"est_tokens": 299,
"chars": 1199,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
023f3895f4739548 | ### Enums
**`reasoning_mode`**: How the template handles reasoning/thinking blocks.
| Value | Description |
|-----------------|-----------------------------------------------------------------------------------|
| `NONE` | No rea... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Data Structures > Enums",
"line_start": 43,
"line_end": 59,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 466,
"chars": 2035,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
0867e55f016cb14d | 1. **Start+end in generation prompt** (e.g. `<think></think>\n`): the parser sees reasoning as opened and immediately closed; whitespace-only reasoning content is discarded.
2. **Only start in generation prompt** (e.g. `<think>\n`): the parser sees reasoning as already open.
3. **Start marker present but not at the end... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Data Structures > Enums",
"line_start": 61,
"line_end": 89,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 472,
"chars": 2388,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
ca10ede50f4ea908 | ## Tool Calling Formats
### JSON_NATIVE
**Structure**: The entire tool call (function name, arguments, values) is in JSON format. Optional enclosing tags around the section.
**Detection**: Function name appears inside a JSON structure (quotes preceded by `{` or `:`).
**Examples**:
Standard OpenAI-style:
```json
<... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Tool Calling Formats",
"line_start": 91,
"line_end": 122,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Tool Calling Formats"... | {
"est_tokens": 158,
"chars": 641,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
852f8936252ddbc0 | ### TAG_WITH_JSON
**Structure**: Function name is outside JSON, in tag attributes or XML-style tags. Arguments are a JSON object.
**Detection**: Function name not in JSON, but argument names appear in JSON context.
**Examples**:
Functionary v3.1:
```xml
<function=get_weather>{"location": "Paris", "unit": "celsius"... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Tool Calling Formats > TAG_WITH_JSON",
"line_start": 124,
"line_end": 178,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Tool... | {
"est_tokens": 276,
"chars": 1122,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
286137bbf0bf2d31 | ## Analysis Flow
```text
autoparser::autoparser(tmpl)
|
|-- Phase 1: analyze_reasoning(tmpl, jinja_caps.supports_tool_calls)
| |-- R1: compare_reasoning_presence() — with/without reasoning_content field
| |-- R2: compare_thinking_enabled() — enable_thinking=false vs true
| '-- R3:... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Analysis Flow",
"line_start": 180,
"line_end": 240,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 599,
"chars": 2937,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
0032c38d3466a5fe | ## Entry Point
The auto-parser is invoked in [common/chat.cpp:1280-1310](common/chat.cpp#L1280-L1310) in `common_chat_templates_apply_jinja`. A few specialized templates are handled first (Ministral/Magistral Large 3, GPT-OSS with `<|channel|>`, Functionary v3.2 with `>>>all`), then the auto-parser handles everything ... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Entry Point",
"line_start": 242,
"line_end": 265,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Entry Point",
"Auto-Parse... | {
"est_tokens": 355,
"chars": 1428,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
dff12acd2c518dfc | ### Phase 1: Reasoning Analysis
**R1 — `compare_reasoning_presence()`**: Compares assistant message with vs without a `reasoning_content` field.
- Searches `diff.right` (output with reasoning) for the reasoning content needle
- Uses PEG parsers to find surrounding markers:
- If both pre/post markers found in `diff.... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Phase 1: Reasoning Analysis",
"line_start": 267,
"line_end": 300,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architec... | {
"est_tokens": 490,
"chars": 1979,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
1925a77bb6d7b386 | ### Phase 3: Tool Call Analysis
**T1 — `analyze_tool_calls()`**: Compares no-tools vs with-tools output.
- Extracts the tool call section as `diff.right`
- Calls `analyze_tool_call_format()` which first strips reasoning markers from the haystack, then:
- Calls `in_json_haystack()` for both function name and argumen... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Phase 3: Tool Call Analysis",
"line_start": 302,
"line_end": 328,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 502,
"chars": 2031,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
1455ebf90bd9203d | - **A1 `extract_argument_name_markers()`**: Compares `arg_name_A` vs `arg_name_B` (two different argument names).
- Finds shared surrounding structure → `arguments.name_prefix`, `arguments.name_suffix`
- **A2 `extract_argument_value_markers()`**: Compares argument value `"XXXX"` vs `"YYYY"` (same arg, different value... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Phase 3: Tool Call Analysis",
"line_start": 330,
"line_end": 349,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 343,
"chars": 1385,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
8414a38a398498be | ### Workarounds
A workaround array in `common/chat-diff-analyzer.cpp` applies post-hoc patches after analysis. Each workaround is a lambda that inspects the template source and overrides analysis results. Current workarounds:
1. **Old Qwen/DeepSeek thinking templates** — source contains `content.split('</think>')` bu... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Workarounds",
"line_start": 351,
"line_end": 359,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 280,
"chars": 1120,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
2ed200de6cc43582 | ### Parser Building
Each analyzer struct (`analyze_reasoning`, `analyze_content`, `analyze_tools`) implements `build_parser(parser_build_context&)`. They share a `parser_build_context` that carries the PEG builder, inference inputs, the pre-built reasoning parser, and a pointer to the content analyzer.
#### Reasoning... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Parser Building",
"line_start": 361,
"line_end": 373,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Algor... | {
"est_tokens": 274,
"chars": 1320,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
5088df62293714ab | #### Content Parser (`analyze_content::build_parser`)
| Condition | Parser |
|----------------------------------------|---------------------------------------------------------------------------------|
| `json_schema`... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Parser Building > Content Parser (`analyze_content::build_parser`)",
"line_start": 375,
"line_end": 383,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 160,
"chars": 929,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
56eec19f9bcabf74 | #### Tool Parsers (`analyze_tools::build_parser`)
Dispatches by `format.mode`:
**`build_tool_parser_json_native()`**: Calls `p.standard_json_tools()` which internally dispatches to:
- `build_json_tools_function_is_key()` — function name is the JSON key: `{"get_weather": {...}}`
- `build_json_tools_nested_keys()` — n... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Algorithm Details > Parser Building > Tool Parsers (`analyze_tools::build_parser`)",
"line_start": 385,
"line_end": 423,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 437,
"chars": 1771,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
2188b155c7c52124 | ## Mapper
`common_chat_peg_mapper` maps PEG parse results (AST nodes) into `common_chat_msg` structures. Key design:
- **Buffered arguments**: Before `tool_name` is known, argument text goes to `args_buffer`; once the name is set, the buffer is flushed to `current_tool->arguments`
- **`args_target()`**: Returns a ref... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Mapper",
"line_start": 425,
"line_end": 433,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 212,
"chars": 850,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
ceae4104a9015594 | ## Files
| File | Purpose |
|-------------------------------------------|---------------------------------------------------------------------------------|
| `common/chat-auto-parser.h` | All anal... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Files",
"line_start": 435,
"line_end": 447,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 252,
"chars": 1417,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
a2c0945dcf73306f | ## Testing & Debugging
### Debug Tools
**Template Debugger**: `tests/test-chat-auto-parser.cpp`
- Usage: `./bin/test-chat-auto-parser path/to/template.jinja` (without a path, it runs the automated tests)
- Shows detected format, markers, generated parser, and GBNF grammar
**Template Analysis**: `tests/test-chat-ana... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Testing & Debugging",
"line_start": 449,
"line_end": 481,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Testing & Debugging",... | {
"est_tokens": 273,
"chars": 1122,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
0084e8615a7b8df4 | | Template | Format | Notes |
| -------- | ------ | ----- |
| Ministral-3-14B-Reasoning | Reasoning | `[THINK]...[/THINK]` tags (specialized handler) |
| NVIDIA-Nemotron-3-Nano-30B | TAG_WITH_TAGGED | Reasoning + tools |
| CohereForAI Command-R7B | JSON_NATIVE | `<\|START_THINKING\|>`/`<\|START_RESPONSE\|>` markers |
|... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Testing & Debugging > Tested Templates",
"line_start": 483,
"line_end": 516,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 585,
"chars": 2339,
"token_method": "chars/4 estimate (whitespace-normalised)",
"table_part": "1/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
8d49206ef380023f | | Template | Format | Notes |
| -------- | ------ | ----- |
| Spark2.5 | TAG_WITH_TAGGED | `<tool_call>name<arg_key>...<arg_value>...` format |
## Adding Support for New Templates
To support a new template format:
1. **If it follows standard patterns** — The auto-parser should detect it automatically. Run `test-chat... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Testing & Debugging > Tested Templates",
"line_start": 483,
"line_end": 525,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Auto-Parser Architecture > Te... | {
"est_tokens": 197,
"chars": 791,
"token_method": "chars/4 estimate (whitespace-normalised)",
"table_part": "2/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
61c971b1c4b37ed2 | ## Edge Cases and Quirks
1. **Generation Prompt & Reasoning Prefill**: The generation prompt is extracted by diffing `add_generation_prompt=false` vs `true` in `common_chat_templates_apply_jinja`, so it contains exactly what the template appends — avoiding false positives from prior conversation turns.
2. **Per-Call v... | {
"path": "docs/autoparser.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/autoparser.md",
"sha256": "575e05723714578b46c9721ad6e0... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Auto-Parser Architecture > Edge Cases and Quirks",
"line_start": 527,
"line_end": 535,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 416,
"chars": 1665,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
71260925ad5372d4 | BLIS Installation Manual
------------------------
BLIS is a portable software framework for high-performance BLAS-like dense linear algebra libraries. It has received awards and recognition, including the 2023 James H. Wilkinson Prize for Numerical Software and the 2020 SIAM Activity Group on Supercomputing Best Paper... | {
"path": "docs/backend/BLIS.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/BLIS.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/BLIS.md",
"sha256": "a868d79159ed096e925048... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "(root)",
"line_start": 1,
"line_end": 6,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 140,
"chars": 560,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
40f1a56c7e313ead | ### Prepare:
Compile BLIS:
```bash
git clone https://github.com/flame/blis
cd blis
./configure --enable-cblas -t openmp,pthreads auto
# will install to /usr/local/ by default.
make -j
```
Install BLIS:
```bash
sudo make install
```
We recommend using openmp since it's easier to modify the cores being used.
### ll... | {
"path": "docs/backend/BLIS.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/BLIS.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/BLIS.md",
"sha256": "a868d79159ed096e925048... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Prepare:",
"line_start": 8,
"line_end": 60,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Prepare:",
"Prepare: > llama.cpp compilation",
"Prepare: > llama.cpp compilation ... | {
"est_tokens": 269,
"chars": 1093,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
58fddf19b59eb65a | # llama.cpp for CANN
- [Background](#background)
- [News](#news)
- [OS](#os)
- [Hardware](#hardware)
- [Model Supports](#model-supports)
- [DataType Supports](#datatype-supports)
- [Docker](#docker)
- [Linux](#linux)
- [Environment variable setup](#environment-variable-setup)
- [TODO](#todo)
## Background
... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN",
"line_start": 1,
"line_end": 23,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN",
"llama.cpp for CANN > Background"
]
} | {
"est_tokens": 314,
"chars": 1274,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
b629149c14747d5b | ## News
- 2024.11
- Support F16 and F32 data type model for Ascend 310P NPU.
- 2024.8
- Support `Q4_0` and `Q8_0` data type for Ascend NPU.
- 2024.7
- Create CANN backend for Ascend NPU.
## OS
| OS | Status | Verified |
|:-------:|:-------:|:-------------------------... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > News",
"line_start": 25,
"line_end": 65,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > News",
"llama.cpp for CANN > OS",
"llama.c... | {
"est_tokens": 282,
"chars": 1422,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
640f51d04b54702a | ## Model Supports
<details>
<summary>Text-only</summary>
| Model Name | FP16 | Q4_0 | Q8_0 |
|:----------------------------|:-----:|:----:|:----:|
| Llama-2 | √ | √ | √ |
| Llama-3 | √ | √ | √ |
| Mistral-7B | √ | √ ... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Model Supports",
"line_start": 68,
"line_end": 153,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 552,
"chars": 4108,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
3c0dd0a24f44435b | ## DataType Supports
| DataType | 910B | 310P |
|:----------------------:|:-------:|:-------:|
| FP16 | Support | Support |
| Q8_0 | Support | Partial |
| Q4_0 | Support | Partial |
| BF16 | Support | |
> **310P note**... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > DataType Supports",
"line_start": 157,
"line_end": 170,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > DataType Supports",
"llama.cpp ... | {
"est_tokens": 122,
"chars": 590,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
f2ea3451d149f910 | ### Build Images
You can get a image with llama.cpp in one command.
```sh
docker build -t llama-cpp-cann -f .devops/llama-cli-cann.Dockerfile .
```
### Run container
```sh
# Find all cards.
npu-smi info
# Select the cards that you want to use, make sure these cards are not used by someone.
# Following using cards of... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Docker > Build Images",
"line_start": 172,
"line_end": 204,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Docker > Build Images",
"ll... | {
"est_tokens": 253,
"chars": 1043,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
dd4add18c15d3c71 | ## Linux
### I. Setup Environment
1. **Configure Ascend user and group**
```sh
sudo groupadd HwHiAiUser
sudo useradd -g HwHiAiUser -d /home/HwHiAiUser -m HwHiAiUser -s /bin/bash
sudo usermod -aG HwHiAiUser $USER
```
2. **Install dependencies**
**Ubuntu/Debian:**
```sh
sudo apt-get u... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Linux",
"line_start": 206,
"line_end": 260,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Linux",
"llama.cpp for CANN > Linux > I. Se... | {
"est_tokens": 352,
"chars": 1556,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
72ff96398ff3d38e | ### II. Build llama.cpp
```sh
cmake -B build -DGGML_CANN=on -DCMAKE_BUILD_TYPE=release
cmake --build build --config release
```
### III. Run the inference
1. **Retrieve and prepare model**
You can refer to the general [*Obtaining and quantizing models*](../../README.md#obtaining-and-quantizing-models) guide for... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Linux > II. Build llama.cpp",
"line_start": 262,
"line_end": 303,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Linux > II. Build llama.c... | {
"est_tokens": 282,
"chars": 1265,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
e4c41956d6e50888 | ### **GitHub contribution**:
Please add the **[CANN]** prefix/tag in issues/PRs titles to help the CANN-team check/address them without delay.
## Updates
### Basic Flash Attention Support
The basic FA kernel with aclnnops has been added in aclnn_ops.cpp.
Currently, the FA only supports the cases with FP16 KV tensors ... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Linux > **GitHub contribution**:",
"line_start": 305,
"line_end": 316,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Linux > **GitHub con... | {
"est_tokens": 204,
"chars": 818,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
889a268d13c984de | ## Environment variable setup
### GGML_CANN_MEM_POOL
Specifies the memory pool management strategy, Default is vmm.
- vmm: Utilizes a virtual memory manager pool. If hardware support for VMM is unavailable, falls back to the legacy (leg) memory pool.
- prio: Employs a priority queue-based memory pool management.
-... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Environment variable setup",
"line_start": 318,
"line_end": 332,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Environment variable setup... | {
"est_tokens": 129,
"chars": 523,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
7bcbac75844cd2ed | ### GGML_CANN_WEIGHT_NZ
Converting the matmul weight format from ND to NZ to improve performance. Enabled by default.
### GGML_CANN_ACL_GRAPH
Operators are executed using ACL graph execution, rather than in op-by-op (eager) mode. Enabled by default. This option is only effective if `USE_ACL_GRAPH` was enabled at com... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Environment variable setup > GGML_CANN_WEIGHT_NZ",
"line_start": 334,
"line_end": 345,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for CANN > Envi... | {
"est_tokens": 121,
"chars": 489,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
58028d96d253361e | ### GGML_CANN_GRAPH_CACHE_CAPACITY
Maximum number of compiled CANN graphs kept in the LRU cache, default is 12. When the number of cached graphs exceeds this capacity, the least recently used graph will be evicted.
### GGML_CANN_PREFILL_USE_GRAPH
Enable ACL graph execution during the prefill stage, default is false.... | {
"path": "docs/backend/CANN.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CANN.md",
"sha256": "c794f8a9fb448358beeed5... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for CANN > Environment variable setup > GGML_CANN_GRAPH_CACHE_CAPACITY",
"line_start": 347,
"line_end": 357,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for ... | {
"est_tokens": 141,
"chars": 569,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
257d9599eb5f8b0b | # Setting Up CUDA on Fedora
In this guide we setup [Nvidia CUDA](https://docs.nvidia.com/cuda/) in a toolbox container. This guide is applicable for:
- [Fedora Workstation](https://fedoraproject.org/workstation/)
- [Atomic Desktops for Fedora](https://fedoraproject.org/atomic-desktops/)
- [Fedora Spins](https://fedor... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora",
"line_start": 1,
"line_end": 24,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA on Fedora",
"Setting Up CUDA on Fedora > Table of Co... | {
"est_tokens": 309,
"chars": 1240,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
119e1d6cfbb365a8 | ## Prerequisites
- **Toolbox Installed on the Host System** `Fedora Silverblue` and `Fedora Workstation` both have toolbox by default, other distributions may need to install the [toolbox package](https://containertoolbx.org/install/).
- **NVIDIA Drivers and Graphics Card installed on Host System (recommended)** To ru... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Prerequisites",
"line_start": 26,
"line_end": 30,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 137,
"chars": 549,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
cbfb42feee604035 | ### Using the Fedora 41 CUDA Repository
The latest release is 41.
- [Fedora 41 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora41/x86_64/)
**Note:** We recommend using a toolbox environment to prevent system conflicts.
## Creating a Fedora Toolbox Environment
This guide focuses on F... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Prerequisites > Using the Fedora 41 CUDA Repository",
"line_start": 32,
"line_end": 58,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA o... | {
"est_tokens": 236,
"chars": 978,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
226b355e04380870 | ## Installing Essential Development Tools
1. **Synchronize the DNF Package Manager:**
```bash
sudo dnf distro-sync
```
2. **Install **Vim** the default text editor (Optional):**
```bash
sudo dnf install vim-default-editor --allowerasing
```
The `--allowerasing` flag will allow the removal of t... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Installing Essential Development Tools",
"line_start": 60,
"line_end": 82,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 142,
"chars": 610,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
63626e8b20d695d2 | ## Adding the CUDA Repository
Add the NVIDIA CUDA repository to your DNF configuration:
```bash
sudo dnf config-manager addrepo --from-repofile=https://developer.download.nvidia.com/compute/cuda/repos/fedora41/x86_64/cuda-fedora41.repo
```
After adding the repository, synchronize the package manager again:
```bash
... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Adding the CUDA Repository",
"line_start": 84,
"line_end": 104,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA on Fedora > Adding the CU... | {
"est_tokens": 148,
"chars": 600,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
e8e584d015a023a2 | ### If *`libcuda.so.1`* is missing:
```
ls: cannot access '/usr/lib64/libcuda.so.1': No such file or directory
```
**Explanation:**
The host dose not supply the CUDA drivers, **install them now:**
#### Install the Nvidia Driver Libraries on Guest:
```bash
sudo dnf install nvidia-driver-cuda nvidia-driver-libs nvidi... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Installing Nvidia Driver Libraries > If *`libcuda.so.1`* is missing:",
"line_start": 106,
"line_end": 127,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
... | {
"est_tokens": 150,
"chars": 605,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
4ea2310906b7e885 | #### Update the Toolbox RPM Database to include the Host-Supplied Libraries:
Note: we do not actually install the libraries, we just update the DB so that the guest system knows they are supplied by the host.
##### 1. Download `nvidia-` parts that are supplied by the host RPM's (with dependencies)
```bash
sudo dnf d... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Installing Nvidia Driver Libraries > If *`libcuda.so.1`* exists: > Update the Toolbox RPM Database to include the Host-Supplied Libraries:",
"line_start": 129,
"line_end": 169,
... | {
"est_tokens": 387,
"chars": 1562,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
d8f7779d47e21b4f | ## Installing the CUDA Meta-Package
Now that the driver libraries are installed, proceed to install CUDA:
```bash
sudo dnf install cuda
```
This installs the CUDA toolkit and associated packages.
## Configuring the Environment
To use CUDA, add its binary directory to your system's `PATH`.
1. **Create a Profile Sc... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Installing the CUDA Meta-Package",
"line_start": 171,
"line_end": 208,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA on Fedora > Instal... | {
"est_tokens": 265,
"chars": 1113,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
94fdc652ea2e8943 | ## Verifying the Installation
To confirm that CUDA is correctly installed and configured, check the version of the NVIDIA CUDA Compiler (`nvcc`):
```bash
nvcc --version
```
You should see output similar to:
```
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Fri_Feb_21_20:2... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Verifying the Installation",
"line_start": 210,
"line_end": 228,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 129,
"chars": 522,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
cdd3e4cfef1b2d7e | ## Conclusion
You have successfully set up CUDA on Fedora within a toolbox environment using the Fedora 41 CUDA repository. By manually updating the RPM db and configuring the environment, you can develop CUDA applications without affecting your host system.
## Troubleshooting
- **Installation Failures:**
- If yo... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Conclusion",
"line_start": 230,
"line_end": 253,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA on Fedora > Conclusion",
"Setting Up... | {
"est_tokens": 266,
"chars": 1090,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
edbfdc67dbd9de73 | ## Additional Notes
- **Updating CUDA in the Future:**
- Keep an eye on the official NVIDIA repositories for updates to your Fedora version.
- When an updated repository becomes available, adjust your `dnf` configuration accordingly.
- **Building `llama.cpp`:**
- With CUDA installed, you can follow these [bui... | {
"path": "docs/backend/CUDA-FEDORA.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/CUDA-FEDORA.md",
"sha256": "6... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "Setting Up CUDA on Fedora > Additional Notes",
"line_start": 255,
"line_end": 283,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"Setting Up CUDA on Fedora > Additional Notes",
... | {
"est_tokens": 411,
"chars": 1668,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
2fc59dfe741cf573 | # llama.cpp for ET
- [Background](#background)
- [Limitations](#limitations)
- [Build](#build)
- [Develop](#develop)
- [Roadmap](#roadmap)
## Background
**ET** is a llama.cpp backend targeting the fully open source manycore
RISC-V accelerator platform [ET-SOC](https://github.com/aifoundry-org/et-man).
## Limitation... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET",
"line_start": 1,
"line_end": 41,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for ET",
"llama.cpp for ET > Background",
"llama.cpp for ET > L... | {
"est_tokens": 400,
"chars": 1620,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
c6d107c5ac7f18cd | ## Build
### I. Prerequisites
1. **Install custom RISC-V toolchain** - Follow instructions at:
[https://github.com/aifoundry-org/riscv-gnu-toolchain/tree/et/aifoundry](https://github.com/aifoundry-org/riscv-gnu-toolchain/tree/et/aifoundry)
2. **Install ET platform** - Follow instructions at:
[https://github.co... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET > Build",
"line_start": 44,
"line_end": 61,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for ET > Build",
"llama.cpp for ET > Build > I. Prerequisi... | {
"est_tokens": 156,
"chars": 634,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
37dbbf0a8ff5a849 | ### II. Build llama.cpp
Check out llama.cpp with ET backend (this should checkout `et` branch):
```sh
git clone https://github.com/aifoundry-org/llama.cpp
cd llama.cpp
```
Build:
```sh
cmake -B build -DGGML_ET=ON
cmake --build build --config Release
# Optionally:
# cmake --install build
```
Build targeting sysemu ... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET > Build > II. Build llama.cpp",
"line_start": 63,
"line_end": 106,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for ET > Build > II. Build llama.cpp",
... | {
"est_tokens": 224,
"chars": 918,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
6bc41a84a4403c24 | ## Develop
Compute kernels are developed within `ggml/src/ggml-et/et-kernels` folder.
Build is performed using custom RISC-V GNU toolchain and is managed by cmake.
At the moment kernels are build as baremetal elf files, without
standard lib or any other dependencies. All the yummy parts are written
in inline assembler... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET > Develop",
"line_start": 108,
"line_end": 147,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 525,
"chars": 2106,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
844ae683ec2261c5 | ```
ET_PERF|op=MUL_MAT|kernel=mul_mat_f32_Q8_0xf32|duration_us=3112|tensor=Qcur-0|shape=[4096,2,1,1]|start_us=48437862009|end_us=48437865121|flops=67100672
ET_PERF|op=ROPE|kernel=rope_f32|duration_us=9266|tensor=Qcur-0|shape=[128,32,2,1]|start_us=48437865128|end_us=48437874394|mode=0x0|n_dims=128|freq_base=500000.00|fr... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET > Develop",
"line_start": 149,
"line_end": 159,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 187,
"chars": 749,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
72228df85d037c1c | ### Uberkernel
The in-kernel implementation of device dispatch/kernel fusion. The ET SDK has a non-trivial op-to-op gap. `Uberkernel` (name taken from the original Esperanto AI's compiler)
dispatches multiple already existing kernel implementations with device side synchronization. Due to the processor's design, there... | {
"path": "docs/backend/ET.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/ET.md",
"sha256": "83be648e5da5609c443d73ee02ae... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for ET > Develop > Uberkernel",
"line_start": 161,
"line_end": 177,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for ET > Develop > Uberkernel",
"llama.cp... | {
"est_tokens": 266,
"chars": 1068,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
c0ebc2189fc27724 | # llama.cpp for OpenCL
- [llama.cpp for OpenCL](#llamacpp-for-opencl)
- [Background](#background)
- [Llama.cpp + OpenCL](#llamacpp--opencl)
- [OS](#os)
- [Hardware](#hardware)
- [Adreno GPU](#adreno-gpu)
- [DataType Supports](#datatype-supports)
- [Model Preparation](#model-preparation)
- [Binary K... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL",
"line_start": 1,
"line_end": 23,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 195,
"chars": 837,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
330f3812e654a22f | ## Background
OpenCL (Open Computing Language) is an open, royalty-free standard for cross-platform, parallel programming of diverse accelerators found in supercomputers, cloud servers, personal computers, mobile devices and embedded platforms. OpenCL specifies a programming language (based on C99) for programming the... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Background",
"line_start": 25,
"line_end": 27,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 136,
"chars": 544,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
f4b328fb55105e06 | ### Llama.cpp + OpenCL
The llama.cpp OpenCL backend is designed to enable llama.cpp on **Qualcomm Adreno GPU** firstly via OpenCL. Thanks to the portabilty of OpenCL, the OpenCL backend can also run on certain Intel GPUs such as those that do not have [SYCL](/docs/backend/SYCL.md) support although the performance is n... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Background > Llama.cpp + OpenCL",
"line_start": 29,
"line_end": 39,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Background > Llama.... | {
"est_tokens": 155,
"chars": 694,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
0df0c2deb7e2641b | ## Hardware
### Adreno GPU
**Verified devices**
| Adreno GPU | Status |
|:-------------------------------------:|:-------:|
| Adreno 750 (Snapdragon 8 Gen 3) | Support |
| Adreno 810 (Snapdragon 7s Gen 3) | Support |
| Adreno 830 (Snapdragon 8 Elite) | Support |
| Adreno ... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Hardware",
"line_start": 41,
"line_end": 57,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Hardware",
"llama.cpp for OpenCL > Har... | {
"est_tokens": 152,
"chars": 662,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
43cd42c9c893e137 | ## DataType Supports
| DataType | Status |
|:----------------------:|:--------------------------:|
| Q1_0 | Support |
| Q4_0 | Support |
| Q4_1 | Support |
| Q5_0 ... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > DataType Supports",
"line_start": 59,
"line_end": 88,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > DataType Supports",
"llama.cp... | {
"est_tokens": 292,
"chars": 1610,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
ac13957051d7b7be | ## CMake Options
The OpenCL backend has the following CMake options that control the behavior of the backend.
| CMake options | Default value | Description |
|:------------------------------------:|:--------------:|:------------------------------------------|
| `G... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > CMake Options",
"line_start": 91,
"line_end": 99,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 129,
"chars": 621,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
0090857c22bd8739 | ## Program Binary Cache
Compiled `cl_program` binaries are cached on disk, so subsequent runs skip the expensive
compile-from-source step when nothing relevant has changed (kernel source, compile options,
device, driver, or platform version).
The cache is controlled with the `GGML_OPENCL_KERNEL_CACHE_DIR` environment... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Program Binary Cache",
"line_start": 101,
"line_end": 117,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 251,
"chars": 1152,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
bd02aa871dddd7a5 | ## Android
Ubuntu 22.04 is used for targeting Android. Make sure the following tools are accessible from command line,
* Git
* CMake 3.29
* Ninja
* Python3
### I. Setup Environment
1. **Install NDK**
```sh
cd ~
wget https://dl.google.com/android/repository/commandlinetools-linux-8512546_latest.zip && \
unzip comma... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Android",
"line_start": 119,
"line_end": 167,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Android",
"llama.cpp for OpenCL > And... | {
"est_tokens": 375,
"chars": 1521,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
6ee10f9632aee451 | ### II. Build llama.cpp
```sh
cd ~/dev/llm
git clone https://github.com/ggml-org/llama.cpp && \
cd llama.cpp && \
mkdir build-android && cd build-android
cmake .. -G Ninja \
-DCMAKE_TOOLCHAIN_FILE=$HOME/android-sdk/ndk/26.3.11579264/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=arm64-v8a \
-DANDROID_PLAT... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Android > II. Build llama.cpp",
"line_start": 169,
"line_end": 206,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Android > II. Build... | {
"est_tokens": 307,
"chars": 1247,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
c13dfdfc339f73e6 | ### I. Setup Environment
1. **Install OpenCL Headers and Library**
```powershell
mkdir -p ~/dev/llm
cd ~/dev/llm
git clone https://github.com/KhronosGroup/OpenCL-Headers && cd OpenCL-Headers
mkdir build && cd build
cmake .. -G Ninja `
-DBUILD_TESTING=OFF `
-DOPENCL_HEADERS_BUILD_TESTING=OFF `
-DOPENCL_HEADERS_... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Windows 11 Arm64 > I. Setup Environment",
"line_start": 208,
"line_end": 233,
"ingested_at": "2026-09-27T15:46:14+00:00",
"near_duplicates": [
{
"id": "1bef372c2aa97a72",... | {
"est_tokens": 177,
"chars": 727,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
1be76a2e0bac27b6 | ### II. Build llama.cpp
```powershell
mkdir -p ~/dev/llm
cd ~/dev/llm
git clone https://github.com/ggml-org/llama.cpp && cd llama.cpp
mkdir build && cd build
cmake .. -G Ninja `
-DCMAKE_TOOLCHAIN_FILE="$HOME/dev/llm/llama.cpp/cmake/arm64-windows-llvm.cmake" `
-DCMAKE_BUILD_TYPE=Release `
-DCMAKE_PREFIX_PATH="... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Windows 11 Arm64 > II. Build llama.cpp",
"line_start": 235,
"line_end": 258,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Windows 11... | {
"est_tokens": 199,
"chars": 812,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
1bef372c2aa97a72 | ### I. Setup Environment
1. **Install OpenCL Headers and Library**
```bash
mkdir -p ~/dev/llm
cd ~/dev/llm
git clone https://github.com/KhronosGroup/OpenCL-Headers && cd OpenCL-Headers
mkdir build && cd build
cmake .. -G Ninja \
-DBUILD_TESTING=OFF \
-DOPENCL_HEADERS_BUILD_TESTING=OFF \
-DOPENCL_HEADERS_BUILD_... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Linux > I. Setup Environment",
"line_start": 260,
"line_end": 285,
"ingested_at": "2026-09-27T15:46:14+00:00",
"near_duplicate_of": {
"id": "c13dfdfc339f73e6",
"path": "doc... | {
"est_tokens": 176,
"chars": 721,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
51fc788e65eb0d13 | ### II. Build llama.cpp
```bash
mkdir -p ~/dev/llm
cd ~/dev/llm
git clone https://github.com/ggml-org/llama.cpp && cd llama.cpp
mkdir build && cd build
cmake .. -G Ninja \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_PREFIX_PATH="$HOME/dev/llm/opencl" \
-DBUILD_SHARED_LIBS=OFF \
-DGGML_OPENCL=ON
ninja
```
## Known ... | {
"path": "docs/backend/OPENCL.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENCL.md",
"sha256": "385272f5f7a40eee... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "llama.cpp for OpenCL > Linux > II. Build llama.cpp",
"line_start": 287,
"line_end": 313,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"llama.cpp for OpenCL > Linux > II. Build lla... | {
"est_tokens": 161,
"chars": 660,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
a963a797a083a278 | # OpenVINO Backend for llama.cpp
> [!NOTE]
> Performance and memory optimizations, accuracy validation, broader quantization coverage, broader operator and model support are work in progress.
[OpenVINO](https://docs.openvino.ai/) is an open-source toolkit for optimizing and deploying high-performance AI inference, sp... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp",
"line_start": 1,
"line_end": 15,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 454,
"chars": 1821,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
75b655828bc5a325 | ## Contents
- [Supported Devices](#supported-devices)
- [Supported Model Precisions](#supported-model-precisions)
- [Supported Llama.cpp Tools](#supported-llamacpp-tools)
- [Validated Models](#validated-models)
- [Build Instructions](#build-instructions)
- [0. Prerequisites](#0-prerequisites)
- [1. Install OpenVIN... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Contents",
"line_start": 17,
"line_end": 34,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 222,
"chars": 908,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
973876ae214a8c24 | ## Supported Devices
OpenVINO backend supports the following hardware:
- Intel CPUs
- Intel GPUs (integrated and discrete)
- Intel NPUs
Although OpenVINO supports a wide range of [Intel hardware](https://docs.openvino.ai/2026/about-openvino/release-notes-openvino/system-requirements.html), the llama.cpp OpenVINO bac... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Supported Devices",
"line_start": 36,
"line_end": 72,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"OpenVINO Backend for llama.cpp > Supported Dev... | {
"est_tokens": 384,
"chars": 1545,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
518116923f35df32 | ## Supported Llama.cpp Tools
The OpenVINO backend integrates with the standard llama.cpp tools listed below.
However, all the tools coverage across all devices is not uniform and exhaustive validation is work in progress.
- llama-bench
- llama-cli
- llama-completion
- llama-embedding
- llama-perplexity
- llama-run
- ... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Supported Llama.cpp Tools",
"line_start": 74,
"line_end": 102,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"OpenVINO Backend for llama.cpp > Supp... | {
"est_tokens": 380,
"chars": 1533,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
e4d8c8f1ce90564b | | Model | CPU (SL / SF) | GPU (SL / SF) | NPU (SL) |
| :--- | :---: | :---: | :---: |
| [bartowski/Llama-3.2-1B-Instruct-Q4_K_M](https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF) | ✓ / ✓ | ✓ / ✓ | ✓ |
| [bartowski/Llama-3.2-3B-Instruct-Q4_K_M](https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF) | ✓ ... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Validated Models",
"line_start": 104,
"line_end": 127,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 589,
"chars": 2369,
"token_method": "chars/4 estimate (whitespace-normalised)",
"table_part": "1/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
aa632a4f3ce793ed | | Model | CPU (SL / SF) | GPU (SL / SF) | NPU (SL) |
| :--- | :---: | :---: | :---: |
| [bartowski/microsoft_Phi-4-mini-instruct-Q4_K_M](https://huggingface.co/bartowski/microsoft_Phi-4-mini-instruct-GGUF) | ✓ / ✓ | ✓ / ✓ | ✓ |
| | | | |
| [bartowski/Mistral-7B-Instruct-v0.3-Q4_K_M](https://huggingface.co/bartowski... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Validated Models",
"line_start": 104,
"line_end": 148,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 548,
"chars": 2212,
"token_method": "chars/4 estimate (whitespace-normalised)",
"table_part": "2/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
240e3aedba602cc8 | ## Build Instructions
### 0. Prerequisites
- Linux or Windows system with Intel hardware (CPU, GPU, or NPU)
- **For Intel GPU or NPU Usage**: Install the appropriate hardware drivers for your Intel GPU or NPU. For detailed instructions, see: [Additional Configurations for Hardware Acceleration](https://docs.openvino.... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions",
"line_start": 150,
"line_end": 189,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"OpenVINO Backend for llama.cpp > Build Inst... | {
"est_tokens": 336,
"chars": 1460,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
a470b35a4b3959b7 | ### 1. Install OpenVINO Runtime
- Follow the guide to install OpenVINO Runtime from an archive file: [Linux](https://docs.openvino.ai/2026/get-started/install-openvino/install-openvino-archive-linux.html) | [Windows](https://docs.openvino.ai/2026/get-started/install-openvino/install-openvino-archive-windows.html)
- V... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 1. Install OpenVINO Runtime",
"line_start": 191,
"line_end": 238,
"ingested_at": "2026-09-27T15:46:14+00:00",
"merged_headings": [
"OpenVINO Back... | {
"est_tokens": 508,
"chars": 2050,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
960c34293c005e9d | ```bash
#!/usr/bin/env bash
# ============================================
# llama.cpp OpenVINO Build Script (Ninja)
# ============================================
set -euo pipefail
OPENVINO_VERSION_MAJOR="2026.4"
OPENVINO_VERSION_FULL="2026.4.0.22959.99c81491cc3"
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && ... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Ubuntu Build Script",
"line_start": 240,
"line_end": 291,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 519,
"chars": 2123,
"token_method": "chars/4 estimate (whitespace-normalised)",
"code_part": "1/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
2da91e8fa801af39 | ```bash
sudo mkdir -p "${OPENVINO_INSTALL_DIR}"
sudo tar -xzf "${OPENVINO_TGZ}" -C "${OPENVINO_INSTALL_DIR}" --strip-components=1
rm -f "${OPENVINO_TGZ}"
fi
# Refresh symlink: /opt/intel/openvino -> /opt/intel/openvino_${OPENVINO_VERSION_MAJOR}
sudo ln -sfn "${OPENVINO_INSTALL_DIR}" "${OPENVINO_LINK_DIR}"
... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Ubuntu Build Script",
"line_start": 292,
"line_end": 347,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 494,
"chars": 2034,
"token_method": "chars/4 estimate (whitespace-normalised)",
"code_part": "2/2"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
9eb69ae2ed6400a6 | #### Windows Build Script
For Windows users, the following `.bat` script automates the prerequisite installs (Git, Ninja, CMake, Visual Studio 2022 Build Tools, vcpkg + OpenCL), the OpenVINO Runtime download/extract, and the Ninja-based llama.cpp build.
Save the following as `build-llamacpp-ov.bat` next to where you w... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Windows Build Script",
"line_start": 349,
"line_end": 365,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 149,
"chars": 599,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
bdc13fbdb6812660 | ```bat
@echo off
setlocal enabledelayedexpansion
REM ============================================
REM llama.cpp OpenVINO Build Script (Ninja)
REM ============================================
set "OPENVINO_VERSION_MAJOR=2026.4"
set "OPENVINO_VERSION_FULL=2026.4.0.22959.99c81491cc3"
set "SCRIPT_DIR=%~dp0"
set "VCPKG_D... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Windows Build Script",
"line_start": 367,
"line_end": 429,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 682,
"chars": 2807,
"token_method": "chars/4 estimate (whitespace-normalised)",
"code_part": "1/3",
"oversize_reason": "unsplittable_block"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
a4f960222567caf8 | ```bat
if not exist "llama.cpp\CMakeLists.txt" (
echo Cloning llama.cpp...
git clone https://github.com/ggml-org/llama.cpp
)
cd /d "llama.cpp"
set "SCRIPT_DIR=%CD%"
REM ============================================
REM Setup OpenVINO: download & extract to C:\Intel\openvino_%OPENVINO_VERSION_MAJOR%,
REM then p... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Windows Build Script",
"line_start": 430,
"line_end": 491,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 565,
"chars": 2449,
"token_method": "chars/4 estimate (whitespace-normalised)",
"code_part": "2/3"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
8ff915058f458909 | ```bat
set "OPENVINO_ROOT=%OPENVINO_LINK_DIR%"
echo OpenVINO Ready: %OPENVINO_ROOT% -^> %OPENVINO_INSTALL_DIR%
echo ============================================
echo Setting up compiler environment...
echo ============================================
REM Locate Visual Studio Build Tools vcvars64.bat
set "VSWHERE=%Pro... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Windows Build Script",
"line_start": 492,
"line_end": 552,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 504,
"chars": 2099,
"token_method": "chars/4 estimate (whitespace-normalised)",
"code_part": "3/3"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
09e975fa8a341953 | > [!NOTE]
> The script pins OpenVINO `2026.4` via the `OPENVINO_VERSION_MAJOR` / `OPENVINO_VERSION_FULL` variables at the top — edit them to track a different release. From any new shell, source the matching `setupvars` script via the junction — `call "C:\Intel\openvino\setupvars.bat"` from `cmd`, or `& "C:\Intel\openv... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 2. Build llama.cpp with OpenVINO Backend > Windows Build Script",
"line_start": 554,
"line_end": 557,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 136,
"chars": 544,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
ecb82d5d5e77f466 | ### 3. Download Sample Model
Download sample model for testing.
```bash
# Linux
mkdir -p ~/models/
wget https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_K_M.gguf \
-O ~/models/Llama-3.2-1B-Instruct-Q4_K_M.gguf
# Windows PowerShell
mkdir C:\models
Invoke-WebReques... | {
"path": "docs/backend/OPENVINO.md",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"html_url": "https://github.com/ggml-org/llama.cpp/blob/c9064dded732d81f90b34e6b33d4fbd77cbfa058/docs/backend/OPENVINO.md",
"sha256": "b90161fbd4... | {
"repo": "ggml-org/llama.cpp",
"commit": "c9064dded732d81f90b34e6b33d4fbd77cbfa058",
"heading_path": "OpenVINO Backend for llama.cpp > Build Instructions > 3. Download Sample Model",
"line_start": 560,
"line_end": 577,
"ingested_at": "2026-09-27T15:46:14+00:00"
} | {
"est_tokens": 171,
"chars": 692,
"token_method": "chars/4 estimate (whitespace-normalised)"
} | {
"spdx": "MIT",
"url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/c9064dded732d81f90b34e6b33d4fbd77cbfa058/LICENSE"
} |
End of preview. Expand in Data Studio
Forge Dataset Prep — free sample packs
Three downloadable corpora made from public technical documentation. Each pack is chunked JSONL with source metadata, a QA report and a licence notice, so you can inspect the format and trace a record back to its source. Together the three packs hold 1,195 records from 132 source files.
The same packs are described at https://forgedgoods.org/dataset-prep/.
The licence is per pack, not one licence for the repository. The Ollama and llama.cpp packs are MIT; the Python PEPs pack is public domain / CC0-1.0. See Licences.
Packs
| Pack (folder) | Source | Pinned commit | Files ingested | Records | Licence basis |
|---|---|---|---|---|---|
Ollama documentation (ollama/) |
ollama/ollama, docs/ |
16b4376aeadbec58a18b9817d49c37b1b64e33d0 |
61 .md / .mdx files |
264 | MIT — Copyright (c) Ollama |
llama.cpp documentation (llamacpp/) |
ggml-org/llama.cpp, docs/ |
c9064dded732d81f90b34e6b33d4fbd77cbfa058 |
51 .md files |
399 | MIT — Copyright (c) 2023-2026 The ggml authors |
Python Enhancement Proposals (peps/) |
python/peps, peps/ |
ff16962a22fdc5e2095e0cbc5c243ea76e34fb52 |
20 selected .rst PEPs |
532 | Public domain / CC0-1.0, per each file's Copyright section |
All three packs were built by one deterministic pipeline (182d7ecd0cda63e5) using 0 LLM calls, and use the record schema forge.rag_chunk.v1.1.
Repository layout
README.md this dataset card
CHECKSUMS.sha256 SHA-256 of the three ZIPs below
LICENSE empty file (0 bytes); the licences are per pack, see Licences
forge-dsprep-sample1-ollama.zip original release ZIP, unchanged
forge-dsprep-sample2-llamacpp.zip original release ZIP, unchanged
forge-dsprep-sample3-peps.zip original release ZIP, unchanged
ollama/ extracted contents of forge-dsprep-sample1-ollama.zip
llamacpp/ extracted contents of forge-dsprep-sample2-llamacpp.zip
peps/ extracted contents of forge-dsprep-sample3-peps.zip
The three ZIPs sit at the repository root. Apart from the .gitattributes file that Hugging Face keeps in every dataset repository, the repository holds 27 files: README.md, CHECKSUMS.sha256, the empty LICENSE, the three ZIPs, and seven files in each of the three pack folders (3 + 3 + 21).
Each pack folder contains the same seven files:
chunks.jsonl— one JSON record per line, with text, source path and URL, file hash, heading path, line range and licence.schema.json— the JSON Schema the records are validated against (forge.rag_chunk.v1.1).qa_report.json— coverage, source-line and source-URL checks, duplicates and size statistics.sample.md— the first 20 records in a readable form.README.md— chunking policy and limits.manifest.json— SHA-256 values for the five other distribution files (it does not list itself orSOURCE-AND-LICENSE.md).SOURCE-AND-LICENSE.md— source repository, pinned commit and licence basis; the MIT packs include the complete MIT notice.
The extracted files are byte-for-byte the files inside the ZIPs.
Record format
One JSON object per line in chunks.jsonl. Required fields:
id— 16 hex characters.text— the chunk text, which is upstream documentation text.source—path,url,sha256of the source file, and optionallyhtml_url.provenance—repo,commit,heading_path,line_start,line_end(the line range in the original file) andingested_at. Optional:merged_headings,near_duplicates,near_duplicate_of.stats—est_tokens,chars,token_method. Optional flags:short,code_part,table_part,sentence_split,oversize_reason.license—spdxandurl.
Optional fields appear only on the records they apply to, so records do not all have the same set of keys.
Provenance and how to trace a record
Every record stores the repository, the pinned commit, the source path, a pinned raw URL, the SHA-256 of the source file, the heading path and the exact line range. To trace a record: open sample.md, pick a record, and compare it with the source file at the pinned commit using the path and line range stored on the record.
Chunking policy
From each pack's README.md: chunks are 120–600 estimated tokens; short leftovers are flagged stats.short; code fences are never split mid-line, and long fences are split into re-fenced parts flagged stats.code_part. Exact duplicates are removed (normalized hash). Near-duplicates (MinHash/LSH on character 5-grams, confirmed Jaccard >= 0.8) are kept and flagged (provenance.near_duplicate_of); all pairs are listed in qa_report.json under near_pairs.
Licences
The text inside every record is upstream documentation, reused under the licence stated for its pack. The pipeline adds structure and metadata only (record ids, heading paths, line ranges, size estimates, near-duplicate flags and the QA report). Forge asserts no licence of its own over the text.
- Ollama (
ollama/) — MIT. Copyright (c) Ollama. - llama.cpp (
llamacpp/) — MIT. Copyright (c) 2023-2026 The ggml authors. - Python PEPs (
peps/) — public domain / CC0-1.0, read from each PEP's own Copyright section, not MIT. 14 files say the document "has been placed in the public domain"; 6 files say it is "placed in the public domain or under the CC0-1.0-Universal license, whichever is more permissive". Thelicense.spdxfield on every PEP record readsCC0-1.0 OR public-domain (per-file Copyright section).
The MIT packs contain text copied from MIT-licensed repositories, so the notice must accompany copies. The complete notice for each is in that pack's SOURCE-AND-LICENSE.md. Each SOURCE-AND-LICENSE.md is a plain statement of source and licence, not legal advice.
Verify the download
From the root of this repository:
shasum -a 256 -c CHECKSUMS.sha256
Each line should report OK. Inside a pack folder, manifest.json lists SHA-256 values for the five other distribution files.
Limits
- Token counts are estimates: characters divided by four after whitespace normalisation.
- Near-duplicates are flagged, not removed.
- The samples cover
.md,.mdxand.rstpublic documentation only. From llama.cppdocs/, 10docs/ops/*.csvop-support tables and 7 images were excluded as not documentation text. - The QA checks are automated. No human review is claimed, and the packs make no claim about downstream model performance.
- Loading the three
configsin this card's metadata with the Hugging Facedatasetslibrary (version 5.0.1) was tested once, offline, from a local copy of this folder:ollamaloaded 264 rows,llamacpp399 andpeps532, each with the columnsid,text,source,provenance,statsandlicense. Only loading and row counts were checked, not the field contents. Other library versions, and loading from the Hub, were not tested.
- Downloads last month
- 88