ollama/model/parsers/functiongemma.go
Daniel Hiltgen 9db4bdbad6
runner: Remove CGO engines, use llama-server exclusively for GGML models (#16031)
* broad lint fixes to sidestep CI scope glitch

* runner: Remove CGO engines, use llama-server exclusively for GGML models

Remove the vendored GGML and llama.cpp backend, CGO runner, Go model
implementations, and sample.  llama-server (built from upstream llama.cpp via
FetchContent) is now the sole inference engine for GGUF-based models.
(Safetensor based models continue to run on the new MLX engine.)  This allows
us to more rapidly pick up new capabilities and fixes from llama.cpp as they
come out.

On windows this now requires recent AMD driver versions to support ROCm v7 as
llama.cpp currently does not support building against v6.

* llama/compat: load Ollama-format GGUFs in llama-server

Squashed from upstream/jmorganca/llama-compat on 2026-04-29.
Source tip: 0c33775d37.

Original source commits:
- 25223160d llama/compat: add in-memory shim so llama-server can load Ollama-format GGUFs
- 7449b539a llm,server: route Ollama-format gemma3 blobs through llama/compat
- 436f2e2b1 llama/compat: make patch-apply idempotent
- 8c2c9d4c8 llama/compat: extend gemma3 handler to cover 1B and 270M blobs
- 021389f7b llama/compat: shrink clip.cpp injection from 18 lines to 1
- 61b367ec2 llama/compat: shrink patch to pure call-site hooks (34 -> 20 lines)
- 36049361c llama/compat: simplify shim (gemma3-tested)
- 8fa664865 llama/compat: add qwen35moe text handler
- db0c74530 llama/compat: add qwen35moe vision (clip) support
- 2a388da77 llama/compat: split shared infra into a util TU
- 9a69a17dc llama/compat: document non-public API dependencies
- d0f38a915 llama/compat: add gpt-oss and lfm2 handlers
- 086071822 llama/compat: add mistral3 text handler (vision TODO)
- 63bde9ff7 llama/compat: add mistral3 vision (clip) support
- 3a57b89d5 llama/compat: apply LLaMA RoPE permute to mistral3 vision Q/K
- 99cb87439 llama/compat: add qwen35, gemma4, deepseek-ocr handlers
- 2c7850dba llama/compat: add nemotron_h_moe handler (latent FFN + MTP skip)
- 9e3b54225 llama/compat: add llama4 text + clip handlers
- 034fee349 llama/compat: add gemma4 clip handler (gemma4v projector)
- 9945c5a93 server: remove dhiltgen/* compat redirect table
- 5d4539101 llama/compat: rewrite gemma4 tokenizer model to BPE
- 7e0765327 llama/compat: add glm-ocr text handler + text-loader load-op hook
- f1bd1a25a llama/compat: add glm-ocr clip handler (glm4v projector)
- 4b5cf3420 llama/compat: collapse text-loader hook back to one new patch line
- eb4ecf4fc llama/compat: extend gemma4 clip handler to gemma4a (audio)
- a23a5e76f llama/compat: fix gemma4a per-block norm tensor mapping
- cd2dcaff4 llama/compat: add embeddinggemma handler
- 1ce8a6b26 llama/compat: add qwen3-vl + qwen2.5-vl handlers
- fd98ffa1e llama/compat: add gemma3n + glm4moelite handlers
- cc7bdf0bc llama/compat: handle null buft in maybe_load_tensor
- 0c33775d3 llama/compat: disable mmap when load_op transforms text-side tensors

* refine implementation

* ci: fix windows MLX build

* ci: fix windows llama-server build

* ci: fix windows rocm build

* ci: windows mlx tuning

Shorten long-tail on build, and get OllamaSetup.exe back under 2g limit

* ci: fix windows dependencies

* win: fix dependency gathering

* disable openmp

* win: arm64 cross-compile build

also DRY out CI steps

* scheduler improvements

* ci: improvements from #15982

* win: favor ninja for faster developer builds

* win: fix build

* win: fix arm64 cross-compile

* win: avoid spaces in compiler path

* misc discovery fixes, and bos handling

* lint fixes

* win: fix arm cross-compile build/CI bugs

* llama.cpp update

* win: handle multiple CRT dirs

* vulkan: add windows iGPU detection

* fix creation bugs for patched models, other refactoring work

* tune batch size for better performance

* ci and lint fixes

* fix repeat_last_n bug

* build: revamp build for better developer UX

* amd, sampler, qwen3next fixes

* version bump

* fix mlx build

* revamp GPU discovery

Scanning the output of llama-server is turning out to be too error prone across
llama.cpp updates, so this switches to a thin dynamic library load against the
bundled GGML libraries so more details can be gathered from the API.

* version bump

* missing file

* ci: fix cache miss on rocm build

* refine vulkan dep handling

* fix ps reporting bug on full GPU load

* improve cmake wiring for customized local builds

* version bump

* docker build arg cleanup

* improve windows exit error logs

* fix community gemma4 support and ci flakes

* fix mlx unit test

* tighten up ps logic to avoid double counting fit log lines

* version bump

* fix ps view for full gpu layer offload

* add MTP wiring for llama-server and create with GGUFs

* pick best template by capabilities

* version bump

* ci: harden apt repos

* remove unused cpu core discovery

* adjust batch default logic to reduce OOMs

* support larger tool calls

* fix audio support, template show

* qwen35 mtp patch support

* flesh out dtypes

* rocm deps

* version bump

* lint fix

* block broken gfx1150 on windows

* fix qwen3.5 moe mtp tensors in patch

* mmproj oom fallback and vulkan on by default

* qwen MTP compat fix

* version bump

* ci: fix WoA cross-compile

* ci: workaround ui tool in cross-compile

* version bump

* win: enable OpenMP for CPU builds

* build: improve developer UX

* ci: windows path workaround for CPU build

* win: fix WoA dependencies

* win: fix large offset reads for mmproj patched loads

* version bump

* fix vulkan dup detection

* add OLLAMA_IGPU_ENABLE and largely disable iGPUs by default

* opt-in MTP, win large offset, integraton fixes

* fix unit test scheduler interaction hang

* fix multi-gpu filtering

* version bump

* review comments

* fix thinking level

* fix linux rocm ordering and granite 3.3 template

* version bump

* ci fix - non-shallow MLX checkout

* bypass linux sysfs unit test on windows

---------

Co-authored-by: jmorganca <jmorganca@gmail.com>
2026-05-29 13:35:47 -07:00

337 lines
8 KiB
Go

package parsers
import (
"fmt"
"regexp"
"strings"
"github.com/ollama/ollama/api"
)
type FunctionGemmaParserState int
const (
FunctionGemmaCollectingContent FunctionGemmaParserState = iota
FunctionGemmaCollectingToolCalls
)
const (
functionGemmaFunctionCallOpen = "<start_function_call>"
functionGemmaFunctionCallClose = "<end_function_call>"
)
// This format uses <start_function_call>call:name{args}<end_function_call> for tool calls.
type FunctionGemmaParser struct {
state FunctionGemmaParserState
buffer strings.Builder
tools []api.Tool
callIndex int
}
func (p *FunctionGemmaParser) HasToolSupport() bool { return true }
func (p *FunctionGemmaParser) HasThinkingSupport() bool { return false }
func (p *FunctionGemmaParser) PreservedTokens() []string {
return []string{
functionGemmaFunctionCallOpen,
functionGemmaFunctionCallClose,
}
}
func (p *FunctionGemmaParser) Init(tools []api.Tool, lastMessage *api.Message, thinkValue *api.ThinkValue) []api.Tool {
p.tools = tools
p.state = FunctionGemmaCollectingContent
p.callIndex = 0
return tools
}
type functionGemmaEvent interface {
isFunctionGemmaEvent()
}
type FunctionGemmaEventContent struct {
content string
}
type functionGemmaEventToolCall struct {
toolCall api.ToolCall
}
func (FunctionGemmaEventContent) isFunctionGemmaEvent() {}
func (functionGemmaEventToolCall) isFunctionGemmaEvent() {}
func (p *FunctionGemmaParser) Add(s string, done bool) (content string, thinking string, calls []api.ToolCall, err error) {
p.buffer.WriteString(s)
events := p.parseEvents()
var toolCalls []api.ToolCall
var contentSb strings.Builder
for _, event := range events {
switch event := event.(type) {
case functionGemmaEventToolCall:
toolCalls = append(toolCalls, event.toolCall)
case FunctionGemmaEventContent:
contentSb.WriteString(event.content)
}
}
for i := range toolCalls {
toolCalls[i].Function.Index = p.callIndex
p.callIndex++
}
return contentSb.String(), "", toolCalls, nil
}
func (p *FunctionGemmaParser) parseEvents() []functionGemmaEvent {
var all []functionGemmaEvent
keepLooping := true
for keepLooping {
var events []functionGemmaEvent
events, keepLooping = p.eat()
if len(events) > 0 {
all = append(all, events...)
}
}
return all
}
// emitWithPartialCheck extracts unambiguous content before a potential partial tag
func (p *FunctionGemmaParser) emitWithPartialCheck(bufStr, tag string) (unambiguous, ambiguous string) {
if overlapLen := overlap(bufStr, tag); overlapLen > 0 {
beforePartialTag := bufStr[:len(bufStr)-overlapLen]
return beforePartialTag, bufStr[len(beforePartialTag):]
}
return bufStr, ""
}
func (p *FunctionGemmaParser) eat() ([]functionGemmaEvent, bool) {
bufStr := p.buffer.String()
if bufStr == "" {
return nil, false
}
switch p.state {
case FunctionGemmaCollectingContent:
if strings.Contains(bufStr, functionGemmaFunctionCallOpen) {
split := strings.SplitN(bufStr, functionGemmaFunctionCallOpen, 2)
content := split[0]
p.buffer.Reset()
p.buffer.WriteString(split[1])
p.state = FunctionGemmaCollectingToolCalls
if content != "" {
return []functionGemmaEvent{FunctionGemmaEventContent{content: content}}, true
}
return nil, true
}
unambig, ambig := p.emitWithPartialCheck(bufStr, functionGemmaFunctionCallOpen)
p.buffer.Reset()
p.buffer.WriteString(ambig)
if unambig != "" {
return []functionGemmaEvent{FunctionGemmaEventContent{content: unambig}}, false
}
return nil, false
case FunctionGemmaCollectingToolCalls:
if strings.Contains(bufStr, functionGemmaFunctionCallClose) {
split := strings.SplitN(bufStr, functionGemmaFunctionCallClose, 2)
remaining := split[1]
p.buffer.Reset()
p.buffer.WriteString(remaining)
var events []functionGemmaEvent
if tc, err := p.parseToolCall(split[0]); err == nil {
events = append(events, functionGemmaEventToolCall{toolCall: tc})
}
if !strings.Contains(remaining, functionGemmaFunctionCallOpen) {
p.state = FunctionGemmaCollectingContent
}
return events, true
}
return nil, false
}
return nil, false
}
// Matches call:function_name{args}
var functionGemmaCallRegex = regexp.MustCompile(`call:([^{]+)\{(.*)\}`)
func (p *FunctionGemmaParser) parseToolCall(content string) (api.ToolCall, error) {
toolCall := api.ToolCall{}
// Extract function name and arguments
match := functionGemmaCallRegex.FindStringSubmatch(content)
if len(match) < 3 {
return toolCall, nil
}
toolCall.Function.Name = match[1]
argsStr := match[2]
// Parse arguments
toolCall.Function.Arguments = p.parseArguments(argsStr)
return toolCall, nil
}
// parseArguments parses the key:value,key:value format
func (p *FunctionGemmaParser) parseArguments(argsStr string) api.ToolCallFunctionArguments {
args := api.NewToolCallFunctionArguments()
if argsStr == "" {
return args
}
// Split by comma, but handle nested structures
parts := p.splitArguments(argsStr)
for _, part := range parts {
// Find the first colon to split key:value
colonIdx := strings.Index(part, ":")
if colonIdx == -1 {
continue
}
key := part[:colonIdx]
value := part[colonIdx+1:]
// Parse the value
args.Set(key, p.parseValue(value))
}
return args
}
// splitArguments splits arguments by comma, respecting nested structures
func (p *FunctionGemmaParser) splitArguments(argsStr string) []string {
var parts []string
var current strings.Builder
depth := 0
inEscape := false
for i := 0; i < len(argsStr); i++ {
ch := argsStr[i]
// Check for <escape> tags
if i+8 <= len(argsStr) && argsStr[i:i+8] == "<escape>" {
inEscape = !inEscape
current.WriteString("<escape>")
i += 7 // Skip the rest of <escape>
continue
}
if !inEscape {
switch ch {
case '{', '[':
depth++
current.WriteByte(ch)
case '}', ']':
depth--
current.WriteByte(ch)
case ',':
if depth == 0 {
if current.Len() > 0 {
parts = append(parts, current.String())
current.Reset()
}
continue
}
current.WriteByte(ch)
default:
current.WriteByte(ch)
}
} else {
current.WriteByte(ch)
}
}
if current.Len() > 0 {
parts = append(parts, current.String())
}
return parts
}
// parseValue parses a single value from the FunctionGemma format
func (p *FunctionGemmaParser) parseValue(value string) any {
// Check for escaped string
if strings.HasPrefix(value, "<escape>") && strings.HasSuffix(value, "<escape>") {
// Remove the escape tags
return value[8 : len(value)-8]
}
// Check for boolean
if value == "true" {
return true
}
if value == "false" {
return false
}
// Check for number
if num, ok := parseNumber(value); ok {
return num
}
// Check for array
if strings.HasPrefix(value, "[") && strings.HasSuffix(value, "]") {
return p.parseArray(value[1 : len(value)-1])
}
// Check for object
if strings.HasPrefix(value, "{") && strings.HasSuffix(value, "}") {
return p.parseObject(value[1 : len(value)-1])
}
// Default to string
return value
}
// parseArray parses an array value
func (p *FunctionGemmaParser) parseArray(content string) []any {
var result []any
parts := p.splitArguments(content)
for _, part := range parts {
result = append(result, p.parseValue(part))
}
return result
}
// parseObject parses an object value
func (p *FunctionGemmaParser) parseObject(content string) map[string]any {
result := make(map[string]any)
parts := p.splitArguments(content)
for _, part := range parts {
colonIdx := strings.Index(part, ":")
if colonIdx == -1 {
continue
}
key := part[:colonIdx]
value := part[colonIdx+1:]
result[key] = p.parseValue(value)
}
return result
}
// parseNumber tries to parse a string as a number
func parseNumber(s string) (any, bool) {
// Try integer first
var intVal int64
if _, err := fmt.Sscanf(s, "%d", &intVal); err == nil {
// Check if the entire string was consumed
if fmt.Sprintf("%d", intVal) == s {
return intVal, true
}
}
// Try float
var floatVal float64
if _, err := fmt.Sscanf(s, "%f", &floatVal); err == nil {
return floatVal, true
}
return nil, false
}