Compare commits

...

7 Commits

Author SHA1 Message Date
Ettore Di Giacinto
c28e8ca697 Merge branch 'master' into ci/static-check 2024-07-18 19:44:59 +02:00
Ettore Di Giacinto
031627584b ci: try to fix commit-pr workflow
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
2024-07-18 19:44:18 +02:00
Ettore Di Giacinto
24a8eebcef refactor: move federated server logic to its own service (#2914)
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2024-07-18 19:15:15 +02:00
Ettore Di Giacinto
bf9dd1de7f feat(functions): parse broken JSON when we parse the raw results, use dynamic rules for grammar keys (#2912)
* feat(functions): enhance parsing with broken JSON when we parse the raw results

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* breaking: make function name by default

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* feat(grammar): dynamically generate grammars with mutating keys

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* refactor: simplify condition

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Update docs

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2024-07-18 17:52:22 +02:00
Ettore Di Giacinto
35d55572ac fix: do not list txt files as potential models (#2910)
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2024-07-18 14:44:44 +02:00
Ettore Di Giacinto
c7357a9872 fix: short-circuit when nodes aren't detected (#2909)
Fixes:

```
panic: invalid argument to IntN

goroutine 401 [running]:
math/rand/v2.(*Rand).IntN(...)
        /home/mudler/_git/go/pkg/mod/golang.org/toolchain@v0.0.1-go1.22.4.linux-amd64/src/math/rand/v2/rand.go:190
math/rand/v2.IntN(...)
        /home/mudler/_git/go/pkg/mod/golang.org/toolchain@v0.0.1-go1.22.4.linux-amd64/src/math/rand/v2/rand.go:307
github.com/mudler/LocalAI/core/cli.Proxy.func2()
        /home/mudler/_git/LocalAI/core/cli/federated.go:104 +0x76e
created by github.com/mudler/LocalAI/core/cli.Proxy in goroutine 1
        /home/mudler/_git/LocalAI/core/cli/federated.go:91 +0x3c5
```

When no nodes are found and something is trying to hit the federated
endpoint (and no tunnels are ready yet).

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2024-07-18 14:44:31 +02:00
Ettore Di Giacinto
ecaaff8f03 ci: add static-checker
**Description**

This PR adds a static-checker pipeline as part of our workflows

**Notes for Reviewers**
N/A

**[Signed commits](../CONTRIBUTING.md#signing-off-on-commits-developer-certificate-of-origin)**
- [x] Yes, I signed my commits.

Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
2024-07-12 10:28:44 +02:00
19 changed files with 512 additions and 326 deletions

View File

@@ -8,9 +8,10 @@ jobs:
MODEL_NAME: hermes-2-theta-llama-3-8b
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Checkout code
uses: actions/checkout@v3
with:
fetch-depth: 0 # needed to checkout all branches for this Action to work
ref: "${{ github.event.pull_request.merge_commit_sha }}"
- uses: mudler/localai-github-action@v1
with:
model: 'hermes-2-theta-llama-3-8b' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"

70
.github/workflows/static-check.yml vendored Normal file
View File

@@ -0,0 +1,70 @@
name: static check
on: pull_request
jobs:
imports:
name: Imports
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: imports
token: ${{ secrets.GITHUB_TOKEN }}
errcheck:
name: Errcheck
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: errcheck
token: ${{ secrets.GITHUB_TOKEN }}
lint:
name: Lint
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: lint
token: ${{ secrets.GITHUB_TOKEN }}
shadow:
name: Shadow
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: shadow
token: ${{ secrets.GITHUB_TOKEN }}
staticcheck:
name: StaticCheck
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: staticcheck
token: ${{ secrets.GITHUB_TOKEN }}
sec:
name: Sec
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@master
- name: check
uses: danhunsaker/golang-github-actions@v1.3.0
with:
run: sec
token: ${{ secrets.GITHUB_TOKEN }}
flags: "-exclude=G104"

View File

@@ -2,20 +2,9 @@ package cli
import (
"context"
"errors"
"fmt"
"io"
"net"
"time"
"math/rand/v2"
cliContext "github.com/mudler/LocalAI/core/cli/context"
"github.com/mudler/LocalAI/core/p2p"
"github.com/mudler/edgevpn/pkg/node"
"github.com/mudler/edgevpn/pkg/protocol"
"github.com/mudler/edgevpn/pkg/types"
"github.com/rs/zerolog/log"
)
type FederatedCLI struct {
@@ -25,106 +14,7 @@ type FederatedCLI struct {
func (f *FederatedCLI) Run(ctx *cliContext.Context) error {
n, err := p2p.NewNode(f.Peer2PeerToken)
if err != nil {
return fmt.Errorf("creating a new node: %w", err)
}
err = n.Start(context.Background())
if err != nil {
return fmt.Errorf("creating a new node: %w", err)
}
fs := p2p.NewFederatedServer(f.Address, p2p.FederatedID, f.Peer2PeerToken)
if err := p2p.ServiceDiscoverer(context.Background(), n, f.Peer2PeerToken, p2p.FederatedID, nil); err != nil {
return err
}
return Proxy(context.Background(), n, f.Address, p2p.FederatedID)
}
func Proxy(ctx context.Context, node *node.Node, listenAddr, service string) error {
log.Info().Msgf("Allocating service '%s' on: %s", service, listenAddr)
// Open local port for listening
l, err := net.Listen("tcp", listenAddr)
if err != nil {
log.Error().Err(err).Msg("Error listening")
return err
}
// ll.Info("Binding local port on", srcaddr)
ledger, _ := node.Ledger()
// Announce ourselves so nodes accepts our connection
ledger.Announce(
ctx,
10*time.Second,
func() {
// Retrieve current ID for ip in the blockchain
//_, found := ledger.GetKey(protocol.UsersLedgerKey, node.Host().ID().String())
// If mismatch, update the blockchain
//if !found {
updatedMap := map[string]interface{}{}
updatedMap[node.Host().ID().String()] = &types.User{
PeerID: node.Host().ID().String(),
Timestamp: time.Now().String(),
}
ledger.Add(protocol.UsersLedgerKey, updatedMap)
// }
},
)
defer l.Close()
for {
select {
case <-ctx.Done():
return errors.New("context canceled")
default:
log.Debug().Msg("New for connection")
// Listen for an incoming connection.
conn, err := l.Accept()
if err != nil {
fmt.Println("Error accepting: ", err.Error())
continue
}
// Handle connections in a new goroutine, forwarding to the p2p service
go func() {
var tunnelAddresses []string
for _, v := range p2p.GetAvailableNodes(p2p.FederatedID) {
if v.IsOnline() {
tunnelAddresses = append(tunnelAddresses, v.TunnelAddress)
} else {
log.Info().Msgf("Node %s is offline", v.ID)
}
}
// open a TCP stream to one of the tunnels
// chosen randomly
// TODO: optimize this and track usage
tunnelAddr := tunnelAddresses[rand.IntN(len(tunnelAddresses))]
tunnelConn, err := net.Dial("tcp", tunnelAddr)
if err != nil {
log.Error().Err(err).Msg("Error connecting to tunnel")
return
}
log.Info().Msgf("Redirecting %s to %s", conn.LocalAddr().String(), tunnelConn.RemoteAddr().String())
closer := make(chan struct{}, 2)
go copyStream(closer, tunnelConn, conn)
go copyStream(closer, conn, tunnelConn)
<-closer
tunnelConn.Close()
conn.Close()
// ll.Infof("(service %s) Done handling %s", serviceID, l.Addr().String())
}()
}
}
}
func copyStream(closer chan struct{}, dst io.Writer, src io.Reader) {
defer func() { closer <- struct{}{} }() // connection is closed, send signal to stop proxy
io.Copy(dst, src)
return fs.Start(context.Background())
}

View File

@@ -119,7 +119,7 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
}
log.Info().Msg("Starting P2P server discovery...")
if err := p2p.ServiceDiscoverer(context.Background(), node, token, "", func() {
if err := p2p.ServiceDiscoverer(context.Background(), node, token, "", func(serviceID string, node p2p.NodeData) {
var tunnelAddresses []string
for _, v := range p2p.GetAvailableNodes("") {
if v.IsOnline() {

View File

@@ -225,18 +225,10 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
}
// Update input grammar
// Handle if we should return "name" instead of "functions"
if config.FunctionsConfig.FunctionName {
jsStruct := funcs.ToJSONNameStructure()
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
} else {
jsStruct := funcs.ToJSONFunctionStructure()
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
}
jsStruct := funcs.ToJSONStructure(config.FunctionsConfig.FunctionNameKey, config.FunctionsConfig.FunctionNameKey)
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
case input.JSONFunctionGrammarObject != nil:
config.Grammar = input.JSONFunctionGrammarObject.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
case input.JSONFunctionGrammarObjectName != nil:
config.Grammar = input.JSONFunctionGrammarObjectName.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
default:
// Force picking one of the functions by the request
if config.FunctionToCall() != "" {

15
core/p2p/federated.go Normal file
View File

@@ -0,0 +1,15 @@
package p2p
const FederatedID = "federated"
type FederatedServer struct {
listenAddr, service, p2ptoken string
}
func NewFederatedServer(listenAddr, service, p2pToken string) *FederatedServer {
return &FederatedServer{
listenAddr: listenAddr,
service: service,
p2ptoken: p2pToken,
}
}

View File

@@ -0,0 +1,127 @@
//go:build p2p
// +build p2p
package p2p
import (
"context"
"errors"
"fmt"
"net"
"time"
"math/rand/v2"
"github.com/mudler/edgevpn/pkg/node"
"github.com/mudler/edgevpn/pkg/protocol"
"github.com/mudler/edgevpn/pkg/types"
"github.com/rs/zerolog/log"
)
func (f *FederatedServer) Start(ctx context.Context) error {
n, err := NewNode(f.p2ptoken)
if err != nil {
return fmt.Errorf("creating a new node: %w", err)
}
err = n.Start(ctx)
if err != nil {
return fmt.Errorf("creating a new node: %w", err)
}
if err := ServiceDiscoverer(ctx, n, f.p2ptoken, f.service, func(servicesID string, tunnel NodeData) {
log.Debug().Msgf("Discovered node: %s", tunnel.ID)
}); err != nil {
return err
}
return f.proxy(ctx, n)
}
func (fs *FederatedServer) proxy(ctx context.Context, node *node.Node) error {
log.Info().Msgf("Allocating service '%s' on: %s", fs.service, fs.listenAddr)
// Open local port for listening
l, err := net.Listen("tcp", fs.listenAddr)
if err != nil {
log.Error().Err(err).Msg("Error listening")
return err
}
// ll.Info("Binding local port on", srcaddr)
ledger, _ := node.Ledger()
// Announce ourselves so nodes accepts our connection
ledger.Announce(
ctx,
10*time.Second,
func() {
// Retrieve current ID for ip in the blockchain
//_, found := ledger.GetKey(protocol.UsersLedgerKey, node.Host().ID().String())
// If mismatch, update the blockchain
//if !found {
updatedMap := map[string]interface{}{}
updatedMap[node.Host().ID().String()] = &types.User{
PeerID: node.Host().ID().String(),
Timestamp: time.Now().String(),
}
ledger.Add(protocol.UsersLedgerKey, updatedMap)
// }
},
)
defer l.Close()
for {
select {
case <-ctx.Done():
return errors.New("context canceled")
default:
log.Debug().Msg("New for connection")
// Listen for an incoming connection.
conn, err := l.Accept()
if err != nil {
fmt.Println("Error accepting: ", err.Error())
continue
}
// Handle connections in a new goroutine, forwarding to the p2p service
go func() {
var tunnelAddresses []string
for _, v := range GetAvailableNodes(fs.service) {
if v.IsOnline() {
tunnelAddresses = append(tunnelAddresses, v.TunnelAddress)
} else {
log.Info().Msgf("Node %s is offline", v.ID)
}
}
if len(tunnelAddresses) == 0 {
log.Error().Msg("No available nodes yet")
return
}
// open a TCP stream to one of the tunnels
// chosen randomly
// TODO: optimize this and track usage
tunnelAddr := tunnelAddresses[rand.IntN(len(tunnelAddresses))]
tunnelConn, err := net.Dial("tcp", tunnelAddr)
if err != nil {
log.Error().Err(err).Msg("Error connecting to tunnel")
return
}
log.Info().Msgf("Redirecting %s to %s", conn.LocalAddr().String(), tunnelConn.RemoteAddr().String())
closer := make(chan struct{}, 2)
go copyStream(closer, tunnelConn, conn)
go copyStream(closer, conn, tunnelConn)
<-closer
tunnelConn.Close()
conn.Close()
// ll.Infof("(service %s) Done handling %s", serviceID, l.Addr().String())
}()
}
}
}

View File

@@ -6,7 +6,6 @@ import (
)
const defaultServicesID = "services_localai"
const FederatedID = "federated"
type NodeData struct {
Name string

View File

@@ -137,14 +137,9 @@ func allocateLocalService(ctx context.Context, node *node.Node, listenAddr, serv
}
func copyStream(closer chan struct{}, dst io.Writer, src io.Reader) {
defer func() { closer <- struct{}{} }() // connection is closed, send signal to stop proxy
io.Copy(dst, src)
}
// This is the main of the server (which keeps the env variable updated)
// This starts a goroutine that keeps LLAMACPP_GRPC_SERVERS updated with the discovered services
func ServiceDiscoverer(ctx context.Context, n *node.Node, token, servicesID string, discoveryFunc func()) error {
func ServiceDiscoverer(ctx context.Context, n *node.Node, token, servicesID string, discoveryFunc func(serviceID string, node NodeData)) error {
if servicesID == "" {
servicesID = defaultServicesID
}
@@ -166,7 +161,7 @@ func ServiceDiscoverer(ctx context.Context, n *node.Node, token, servicesID stri
case tunnel := <-tunnels:
AddNode(servicesID, tunnel)
if discoveryFunc != nil {
discoveryFunc()
discoveryFunc(servicesID, tunnel)
}
}
}
@@ -396,3 +391,8 @@ func newNodeOpts(token string) ([]node.Option, error) {
return nodeOpts, nil
}
func copyStream(closer chan struct{}, dst io.Writer, src io.Reader) {
defer func() { closer <- struct{}{} }() // connection is closed, send signal to stop proxy
io.Copy(dst, src)
}

View File

@@ -14,7 +14,11 @@ func GenerateToken() string {
return "not implemented"
}
func ServiceDiscoverer(ctx context.Context, node *node.Node, token, servicesID string, fn func()) error {
func (f *FederatedServer) Start(ctx context.Context) error {
return fmt.Errorf("not implemented")
}
func ServiceDiscoverer(ctx context.Context, node *node.Node, token, servicesID string, fn func(string, NodeData)) error {
return fmt.Errorf("not implemented")
}

View File

@@ -179,8 +179,7 @@ type OpenAIRequest struct {
// A grammar to constrain the LLM output
Grammar string `json:"grammar" yaml:"grammar"`
JSONFunctionGrammarObject *functions.JSONFunctionStructureFunction `json:"grammar_json_functions" yaml:"grammar_json_functions"`
JSONFunctionGrammarObjectName *functions.JSONFunctionStructureName `json:"grammar_json_name" yaml:"grammar_json_name"`
JSONFunctionGrammarObject *functions.JSONFunctionStructure `json:"grammar_json_functions" yaml:"grammar_json_functions"`
Backend string `json:"backend" yaml:"backend"`

View File

@@ -152,7 +152,8 @@ function:
replace_function_results: [] # Placeholder to replace function call results with arbitrary strings or patterns.
replace_llm_results: [] # Replace language model results with arbitrary strings or patterns.
capture_llm_results: [] # Capture language model results as text result, among JSON, in function calls. For instance, if a model returns a block for "thinking" and a block for "response", this will allow you to capture the thinking block.
return_name_in_function_response: false # Some models might prefer to use "name" rather then "function" when returning JSON data. This will allow to use "name" as a key in the JSON response.
function_name_key: "name"
function_arguments_key: "arguments"
# Feature gating flags to enable experimental or optional features.
feature_flags: {}

View File

@@ -6,6 +6,11 @@ import (
"github.com/rs/zerolog/log"
)
const (
defaultFunctionNameKey = "name"
defaultFunctionArgumentsKey = "arguments"
)
type Function struct {
Name string `json:"name"`
Description string `json:"description"`
@@ -19,50 +24,18 @@ type Tool struct {
}
type Tools []Tool
// ToJSONFunctionStructure converts a list of functions to a JSON structure that can be parsed to a grammar
// This allows the LLM to return a response of the type: { "function": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }
func (f Functions) ToJSONFunctionStructure() JSONFunctionStructureFunction {
js := JSONFunctionStructureFunction{}
for _, function := range f {
// t := function.Parameters["type"]
//tt := t.(string)
properties := function.Parameters["properties"]
defs := function.Parameters["$defs"]
dat, _ := json.Marshal(properties)
dat2, _ := json.Marshal(defs)
prop := map[string]interface{}{}
defsD := map[string]interface{}{}
err := json.Unmarshal(dat, &prop)
if err != nil {
log.Error().Err(err).Msg("error unmarshalling dat")
}
err = json.Unmarshal(dat2, &defsD)
if err != nil {
log.Error().Err(err).Msg("error unmarshalling dat2")
}
if js.Defs == nil {
js.Defs = defsD
}
js.OneOf = append(js.OneOf, ItemFunction{
Type: "object",
Properties: FunctionProperties{
Function: FunctionName{Const: function.Name},
Arguments: Argument{
Type: "object",
Properties: prop,
},
},
})
}
return js
}
// ToJSONNameStructure converts a list of functions to a JSON structure that can be parsed to a grammar
// This allows the LLM to return a response of the type: { "name": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }
func (f Functions) ToJSONNameStructure() JSONFunctionStructureName {
js := JSONFunctionStructureName{}
func (f Functions) ToJSONStructure(name, args string) JSONFunctionStructure {
nameKey := defaultFunctionNameKey
argsKey := defaultFunctionArgumentsKey
if name != "" {
nameKey = name
}
if args != "" {
argsKey = args
}
js := JSONFunctionStructure{}
for _, function := range f {
// t := function.Parameters["type"]
//tt := t.(string)
@@ -85,15 +58,16 @@ func (f Functions) ToJSONNameStructure() JSONFunctionStructureName {
if js.Defs == nil {
js.Defs = defsD
}
js.OneOf = append(js.OneOf, ItemName{
Type: "object",
Properties: NameProperties{
Function: FunctionName{Const: function.Name},
Arguments: Argument{
Type: "object",
Properties: prop,
},
},
property := map[string]interface{}{}
property[nameKey] = FunctionName{Const: function.Name}
property[argsKey] = Argument{
Type: "object",
Properties: prop,
}
js.OneOf = append(js.OneOf, Item{
Type: "object",
Properties: property,
})
}
return js

View File

@@ -35,21 +35,35 @@ var _ = Describe("LocalAI grammar functions", func() {
},
}
js := functions.ToJSONFunctionStructure()
js := functions.ToJSONStructure("function", "arguments")
Expect(len(js.OneOf)).To(Equal(2))
Expect(js.OneOf[0].Properties.Function.Const).To(Equal("create_event"))
Expect(js.OneOf[0].Properties.Arguments.Properties["event_name"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(js.OneOf[0].Properties.Arguments.Properties["event_date"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(js.OneOf[1].Properties.Function.Const).To(Equal("search"))
Expect(js.OneOf[1].Properties.Arguments.Properties["query"].(map[string]interface{})["type"]).To(Equal("string"))
fnName := js.OneOf[0].Properties["function"].(FunctionName)
fnArgs := js.OneOf[0].Properties["arguments"].(Argument)
Expect(fnName.Const).To(Equal("create_event"))
Expect(fnArgs.Properties["event_name"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(fnArgs.Properties["event_date"].(map[string]interface{})["type"]).To(Equal("string"))
jsN := functions.ToJSONNameStructure()
fnName = js.OneOf[1].Properties["function"].(FunctionName)
fnArgs = js.OneOf[1].Properties["arguments"].(Argument)
Expect(fnName.Const).To(Equal("search"))
Expect(fnArgs.Properties["query"].(map[string]interface{})["type"]).To(Equal("string"))
// Test with custom keys
jsN := functions.ToJSONStructure("name", "arguments")
Expect(len(jsN.OneOf)).To(Equal(2))
Expect(jsN.OneOf[0].Properties.Function.Const).To(Equal("create_event"))
Expect(jsN.OneOf[0].Properties.Arguments.Properties["event_name"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(jsN.OneOf[0].Properties.Arguments.Properties["event_date"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(jsN.OneOf[1].Properties.Function.Const).To(Equal("search"))
Expect(jsN.OneOf[1].Properties.Arguments.Properties["query"].(map[string]interface{})["type"]).To(Equal("string"))
fnName = jsN.OneOf[0].Properties["name"].(FunctionName)
fnArgs = jsN.OneOf[0].Properties["arguments"].(Argument)
Expect(fnName.Const).To(Equal("create_event"))
Expect(fnArgs.Properties["event_name"].(map[string]interface{})["type"]).To(Equal("string"))
Expect(fnArgs.Properties["event_date"].(map[string]interface{})["type"]).To(Equal("string"))
fnName = jsN.OneOf[1].Properties["name"].(FunctionName)
fnArgs = jsN.OneOf[1].Properties["arguments"].(Argument)
Expect(fnName.Const).To(Equal("search"))
Expect(fnArgs.Properties["query"].(map[string]interface{})["type"]).To(Equal("string"))
})
})
Context("Select()", func() {

View File

@@ -331,6 +331,7 @@ func (sc *JSONSchemaConverter) resolveReference(ref string, rootSchema map[strin
return def
}
func (sc *JSONSchemaConverter) Grammar(schema map[string]interface{}, options ...func(*GrammarOption)) string {
sc.addRule("freestring", PRIMITIVE_RULES["freestring"])
sc.visit(schema, "", schema)
@@ -352,52 +353,23 @@ type FunctionName struct {
Const string `json:"const"`
}
type FunctionProperties struct {
Function FunctionName `json:"function"`
Arguments Argument `json:"arguments"`
}
type NameProperties struct {
Function FunctionName `json:"name"`
Arguments Argument `json:"arguments"`
}
type Argument struct {
Type string `json:"type"`
Properties map[string]interface{} `json:"properties"`
}
type ItemName struct {
Type string `json:"type"`
Properties NameProperties `json:"properties"`
type Item struct {
Type string `json:"type"`
Properties map[string]interface{} `json:"properties"`
}
type ItemFunction struct {
Type string `json:"type"`
Properties FunctionProperties `json:"properties"`
}
type JSONFunctionStructureName struct {
OneOf []ItemName `json:"oneOf,omitempty"`
AnyOf []ItemName `json:"anyOf,omitempty"`
type JSONFunctionStructure struct {
OneOf []Item `json:"oneOf,omitempty"`
AnyOf []Item `json:"anyOf,omitempty"`
Defs map[string]interface{} `json:"$defs,omitempty"`
}
func (j JSONFunctionStructureName) Grammar(options ...func(*GrammarOption)) string {
grammarOpts := &GrammarOption{}
grammarOpts.Apply(options...)
dat, _ := json.Marshal(j)
return NewJSONSchemaConverter(grammarOpts.PropOrder).GrammarFromBytes(dat, options...)
}
type JSONFunctionStructureFunction struct {
OneOf []ItemFunction `json:"oneOf,omitempty"`
AnyOf []ItemFunction `json:"anyOf,omitempty"`
Defs map[string]interface{} `json:"$defs,omitempty"`
}
func (j JSONFunctionStructureFunction) Grammar(options ...func(*GrammarOption)) string {
func (j JSONFunctionStructure) Grammar(options ...func(*GrammarOption)) string {
grammarOpts := &GrammarOption{}
grammarOpts.Apply(options...)

View File

@@ -9,69 +9,65 @@ import (
. "github.com/onsi/gomega"
)
var testFunctions = []ItemFunction{
func createFunction(field1 string, field2 string, name string, properties map[string]interface{}) map[string]interface{} {
property := map[string]interface{}{}
property[field1] = FunctionName{Const: name}
property[field2] = Argument{
Type: "object",
Properties: properties,
}
return property
}
var testFunctions = []Item{
{
Type: "object",
Properties: FunctionProperties{
Function: FunctionName{
Const: "create_event",
Properties: createFunction(
"function",
"arguments",
"create_event",
map[string]interface{}{
"title": map[string]string{"type": "string"},
"date": map[string]string{"type": "string"},
"time": map[string]string{"type": "string"},
},
Arguments: Argument{ // this is OpenAI's parameter
Type: "object",
Properties: map[string]interface{}{
"title": map[string]string{"type": "string"},
"date": map[string]string{"type": "string"},
"time": map[string]string{"type": "string"},
},
},
},
),
},
{
Type: "object",
Properties: FunctionProperties{
Function: FunctionName{
Const: "search",
},
Arguments: Argument{
Type: "object",
Properties: map[string]interface{}{
"query": map[string]string{"type": "string"},
},
},
},
Properties: createFunction(
"function",
"arguments",
"search",
map[string]interface{}{
"query": map[string]string{"type": "string"},
}),
},
}
var testFunctionsName = []ItemName{
var testFunctionsName = []Item{
{
Type: "object",
Properties: NameProperties{
Function: FunctionName{
Const: "create_event",
Properties: createFunction(
"name",
"arguments",
"create_event",
map[string]interface{}{
"title": map[string]string{"type": "string"},
"date": map[string]string{"type": "string"},
"time": map[string]string{"type": "string"},
},
Arguments: Argument{ // this is OpenAI's parameter
Type: "object",
Properties: map[string]interface{}{
"title": map[string]string{"type": "string"},
"date": map[string]string{"type": "string"},
"time": map[string]string{"type": "string"},
},
},
},
),
},
{
Type: "object",
Properties: NameProperties{
Function: FunctionName{
Const: "search",
},
Arguments: Argument{
Type: "object",
Properties: map[string]interface{}{
"query": map[string]string{"type": "string"},
},
},
},
Properties: createFunction(
"name",
"arguments",
"search",
map[string]interface{}{
"query": map[string]string{"type": "string"},
}),
},
}
@@ -270,7 +266,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects", func() {
structuredGrammar := JSONFunctionStructureFunction{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctions}
grammar := structuredGrammar.Grammar()
@@ -284,7 +280,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects for multiple function return", func() {
structuredGrammar := JSONFunctionStructureFunction{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctions}
grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
@@ -302,7 +298,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects for multiple function return", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
@@ -320,7 +316,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects for multiple function return with a suffix and array", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(
@@ -340,7 +336,7 @@ var _ = Describe("JSON schema grammar tests", func() {
Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))), grammar)
})
It("generates a valid grammar from JSON Objects with a suffix", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"))
@@ -357,7 +353,7 @@ var _ = Describe("JSON schema grammar tests", func() {
Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))), grammar)
})
It("generates a valid grammar from JSON Objects with a suffix and could return string", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString)
@@ -374,7 +370,7 @@ var _ = Describe("JSON schema grammar tests", func() {
Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))), grammar)
})
It("generates a valid grammar from JSON Objects with a suffix that could return text or an array of tools", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString, functions.EnableMaybeArray)
@@ -393,7 +389,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects without a suffix that could return text or an array of tools or just string", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray)
@@ -411,7 +407,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates a valid grammar from JSON Objects without a suffix that could return text or an array of tools or just string. Disables mixedstring", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray, functions.NoMixedFreeString)
@@ -429,7 +425,7 @@ var _ = Describe("JSON schema grammar tests", func() {
})
It("generates parallel tools without newlines in JSON", func() {
structuredGrammar := JSONFunctionStructureName{
structuredGrammar := JSONFunctionStructure{
OneOf: testFunctionsName}
content := `arr ::=
"[" (

View File

@@ -2,6 +2,8 @@ package functions
import (
"encoding/json"
"errors"
"io"
"regexp"
"strings"
@@ -76,7 +78,8 @@ type FunctionsConfig struct {
// FunctionName enable the LLM to return { "name": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }
// instead of { "function": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }.
// This might be useful for certain models trained with the function name as the first token.
FunctionName bool `yaml:"return_name_in_function_response"`
FunctionNameKey string `yaml:"function_name_key"`
FunctionArgumentsKey string `yaml:"function_arguments_key"`
}
type ReplaceResult struct {
@@ -145,6 +148,47 @@ func ParseTextContent(llmresult string, functionConfig FunctionsConfig) string {
return ""
}
// ParseJSON is a function that parses a JSON string that might contain multiple JSON objects
// and syntax errors in between by shifting the offset
// This for e.g. allow to parse
// { "foo": "bar" } invalid { "baz": "qux" }
// into
// [ { "foo": "bar" }, { "baz": "qux" } ]
// Credits to Michael Yang (https://github.com/mxyng) for the original implementation
// This is a slighly reworked version, improved for readability and error handling
func ParseJSON(s string) ([]map[string]any, error) {
var objs []map[string]any
offset := 0
for offset < len(s) {
var obj map[string]any
decoder := json.NewDecoder(strings.NewReader(s[offset:]))
err := decoder.Decode(&obj)
switch {
case errors.Is(err, io.EOF):
return objs, nil
case err == nil:
offset += int(decoder.InputOffset())
objs = append(objs, obj)
default: // handle the error type
var syntaxErr *json.SyntaxError
var unmarshalTypeErr *json.UnmarshalTypeError
switch {
case errors.As(err, &syntaxErr):
offset += int(syntaxErr.Offset)
case errors.As(err, &unmarshalTypeErr):
offset += int(unmarshalTypeErr.Offset)
default:
return objs, err
}
}
}
return objs, nil
}
func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncCallResults {
log.Debug().Msgf("LLM result: %s", llmresult)
@@ -157,9 +201,13 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
}
log.Debug().Msgf("LLM result(function cleanup): %s", llmresult)
functionNameKey := "function"
if functionConfig.FunctionName {
functionNameKey = "name"
functionNameKey := defaultFunctionNameKey
functionArgumentsKey := defaultFunctionArgumentsKey
if functionConfig.FunctionNameKey != "" {
functionNameKey = functionConfig.FunctionNameKey
}
if functionConfig.FunctionArgumentsKey != "" {
functionArgumentsKey = functionConfig.FunctionArgumentsKey
}
results := []FuncCallResults{}
@@ -170,19 +218,13 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
result = make([]FuncCallResults, 0)
for _, s := range results {
var ss []map[string]interface{}
var ss []map[string]any
s = utils.EscapeNewLines(s)
err := json.Unmarshal([]byte(s), &ss)
ss, err := ParseJSON(s)
//err := json.Unmarshal([]byte(s), &ss)
if err != nil {
// If the LLM result is a single object, try unmarshaling it into a single map
var singleObj map[string]interface{}
err = json.Unmarshal([]byte(s), &singleObj)
if err != nil {
log.Debug().Err(err).Str("escapedLLMResult", s).Msg("unable to unmarshal llm result in a single object or an array of JSON objects")
} else {
ss = []map[string]interface{}{singleObj}
}
log.Debug().Err(err).Str("escapedLLMResult", s).Msg("unable to unmarshal llm result in a single object or an array of JSON objects")
}
log.Debug().Msgf("Function return: %s %+v", s, ss)
@@ -195,7 +237,7 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
//return result, fmt.Errorf("unable to find function name in result")
}
// Similarly, while here arguments is a map[string]interface{}, OpenAI actually want a stringified object
args, ok := s["arguments"] // arguments needs to be a string, but we return an object from the grammar result (TODO: fix)
args, ok := s[functionArgumentsKey] // arguments needs to be a string, but we return an object from the grammar result (TODO: fix)
if !ok {
continue
//return result, fmt.Errorf("unable to find arguments in result")
@@ -253,7 +295,7 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
if functionName == "" {
return results
}
results = append(results, FuncCallResults{Name: result[functionNameKey], Arguments: result["arguments"]})
results = append(results, FuncCallResults{Name: result[functionNameKey], Arguments: result[functionArgumentsKey]})
}
}
} else {

View File

@@ -16,7 +16,7 @@ var _ = Describe("LocalAI function parse tests", func() {
Context("when using grammars and single result expected", func() {
It("should parse the function name and arguments correctly", func() {
input := `{"function": "add", "arguments": {"x": 5, "y": 3}}`
input := `{"name": "add", "arguments": {"x": 5, "y": 3}}`
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(1))
@@ -28,13 +28,22 @@ var _ = Describe("LocalAI function parse tests", func() {
Context("when not using grammars and regex is needed", func() {
It("should extract function name and arguments from the regex", func() {
input := `add({"x":5,"y":3})`
functionConfig.ResponseRegex = []string{`(?P<function>\w+)\s*\((?P<arguments>.*)\)`}
functionConfig.ResponseRegex = []string{`(?P<name>\w+)\s*\((?P<arguments>.*)\)`}
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("add"))
Expect(results[0].Arguments).To(Equal(`{"x":5,"y":3}`))
})
It("should extract function name and arguments from the regex", func() {
input := `add({"x":5,"y":3})`
functionConfig.ResponseRegex = []string{`(?P<function>\w+)\s*\((?P<arguments>.*)\)`}
functionConfig.FunctionNameKey = "function"
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("add"))
Expect(results[0].Arguments).To(Equal(`{"x":5,"y":3}`))
})
})
Context("when having invalid input", func() {
@@ -53,7 +62,7 @@ var _ = Describe("LocalAI function parse tests", func() {
Context("when parallel calls are enabled", func() {
It("should handle multiple function calls", func() {
input := `[{"function": "add", "arguments": {"x": 5, "y": 3}}, {"function": "subtract", "arguments": {"x": 10, "y": 7}}]`
input := `[{"name": "add", "arguments": {"x": 5, "y": 3}}, {"name": "subtract", "arguments": {"x": 10, "y": 7}}]`
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(2))
@@ -66,8 +75,8 @@ var _ = Describe("LocalAI function parse tests", func() {
Context("without grammars and without regex", func() {
It("should parse the function name and arguments correctly with the name key", func() {
input := `{"name": "add", "arguments": {"x": 5, "y": 3}}`
functionConfig.FunctionName = true
input := `{"function": "add", "arguments": {"x": 5, "y": 3}}`
functionConfig.FunctionNameKey = "function"
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(1))
@@ -76,7 +85,7 @@ var _ = Describe("LocalAI function parse tests", func() {
})
It("should parse the function name and arguments correctly with the function key", func() {
input := `{"function": "add", "arguments": {"x": 5, "y": 3}}`
input := `{"name": "add", "arguments": {"x": 5, "y": 3}}`
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(1))
@@ -87,7 +96,7 @@ var _ = Describe("LocalAI function parse tests", func() {
It("should parse the result by matching the JSONRegexMatch", func() {
input := `
<tool_call>
{"function": "add", "arguments": {"x": 5, "y": 3}}
{"name": "add", "arguments": {"x": 5, "y": 3}}
</tool_call>`
functionConfig.JSONRegexMatch = []string{`(?s)<tool_call>(.*?)</tool_call>`}
@@ -100,7 +109,7 @@ var _ = Describe("LocalAI function parse tests", func() {
It("should parse the result by matching the JSONRegexMatch", func() {
input := `
{"function": "add", "arguments": {"x": 5, "y": 3}}
{"name": "add", "arguments": {"x": 5, "y": 3}}
</tool_call>`
functionConfig.JSONRegexMatch = []string{`(?s)(.*?)</tool_call>`}
@@ -110,13 +119,21 @@ var _ = Describe("LocalAI function parse tests", func() {
Expect(results[0].Name).To(Equal("add"))
Expect(results[0].Arguments).To(Equal(`{"x":5,"y":3}`))
})
It("should parse the result even with invalid JSON", func() {
input := `{"name": "add", "arguments": {"x": 5, "y": 3}} invalid {"name": "add", "arguments": {"x": 5, "y": 3}}`
results := ParseFunctionCall(input, functionConfig)
Expect(results).To(HaveLen(2))
Expect(results[0].Name).To(Equal("add"))
Expect(results[0].Arguments).To(Equal(`{"x":5,"y":3}`))
})
})
Context("when using ReplaceResults to clean up input", func() {
It("should replace text before and after JSON blob", func() {
input := `
Some text before the JSON
{"function": "add", "arguments": {"x": 5, "y": 3}}
{"name": "add", "arguments": {"x": 5, "y": 3}}
Some text after the JSON
`
@@ -134,7 +151,7 @@ Some text after the JSON
It("should replace text before and after array JSON blob", func() {
input := `
Some text before the JSON
[{"function": "add", "arguments": {"x": 5, "y": 3}}, {"function": "subtract", "arguments": {"x": 10, "y": 7}}]
[{"name": "add", "arguments": {"x": 5, "y": 3}}, {"name": "subtract", "arguments": {"x": 10, "y": 7}}]
Some text after the JSON
`
functionConfig.ReplaceFunctionResults = []ReplaceResult{
@@ -153,7 +170,7 @@ Some text after the JSON
It("should convert single-quoted key-value pairs to double-quoted and escape double quotes within values", func() {
input := `
Some text before the JSON
{'function': '"add"', 'arguments': {'x': 5, 'z': '"v"', 'y': 'v"value"'}}
{'name': '"add"', 'arguments': {'x': 5, 'z': '"v"', 'y': 'v"value"'}}
Some text after the JSON
`
functionConfig.JSONRegexMatch = []string{`(?s)<tool_call>(.*?)</tool_call>`}
@@ -186,7 +203,7 @@ Some text after the JSON
It("should convert single-quoted key-value pairs to double-quoted and escape double quotes within values", func() {
input := `
Some text before the JSON
<tool_call>{'function': '"add"', 'arguments': {'x': 5, 'z': '"v"', 'y': 'v"value"'}}</tool_call>
<tool_call>{'name': '"add"', 'arguments': {'x': 5, 'z': '"v"', 'y': 'v"value"'}}</tool_call>
Some text after the JSON
`
functionConfig.JSONRegexMatch = []string{`(?s)<tool_call>(.*?)</tool_call>`}
@@ -219,8 +236,8 @@ Some text after the JSON
It("should detect multiple functions call where the JSONRegexMatch is repeated", func() {
input := `
Some text before the JSON
<tool_call>{"function": "add", "arguments": {"x": 5, "y": 3}}</tool_call>
<tool_call>{"function": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
<tool_call>{"name": "add", "arguments": {"x": 5, "y": 3}}</tool_call>
<tool_call>{"name": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
Some text after the JSON
`
functionConfig.JSONRegexMatch = []string{`(?s)<tool_call>(.*?)</tool_call>`}
@@ -240,7 +257,7 @@ Some text after the JSON
<sketchpad>
roses are red
</sketchpad>
<tool_call>{"function": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
<tool_call>{"name": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
Some text after the JSON
`
functionConfig.CaptureLLMResult = []string{`(?s)<sketchpad>(.*?)</sketchpad>`}
@@ -251,7 +268,7 @@ roses are red
It("Defaults to empty if doesn't catch any", func() {
input := `
Some text before the JSON
<tool_call>{"function": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
<tool_call>{"name": "subtract", "arguments": {"x": 10, "y": 7}}</tool_call>
Some text after the JSON
`
functionConfig.CaptureLLMResult = []string{`(?s)<sketchpad>(.*?)</sketchpad>`}
@@ -259,4 +276,74 @@ roses are red
Expect(results).To(Equal(""))
})
})
Context("ParseJSON - when given valid JSON strings", func() {
It("should parse multiple JSON objects", func() {
input := `{"key1": "value1"} {"key2": "value2"}`
expected := []map[string]any{
{"key1": "value1"},
{"key2": "value2"},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
It("should parse a single JSON object with various types", func() {
input := `{"key1": "value1", "key2": 2}`
expected := []map[string]any{
{"key1": "value1", "key2": float64(2)},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
It("should handle JSON without syntax errors gracefully", func() {
input := `{"key1": "value1"}`
expected := []map[string]any{
{"key1": "value1"},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
It("should handle JSON without syntax errors gracefully", func() {
input := `[{"key1": "value1"}]`
expected := []map[string]any{
{"key1": "value1"},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
})
Context("ParseJSON - when given invalid JSON strings", func() {
It("should return an error for completely invalid JSON", func() {
input := `invalid json`
result, err := ParseJSON(input)
Expect(err).To(HaveOccurred())
Expect(result).To(BeNil())
})
It("should skip invalid JSON parts and parse valid parts", func() {
input := `{"key1": "value1"} invalid {"key2": "value2"}`
expected := []map[string]any{
{"key1": "value1"},
{"key2": "value2"},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
PIt("should handle JSON with syntax errors gracefully", func() {
input := `{"key1": "value1", "key2": }`
expected := []map[string]any{
{"key1": "value1"},
}
result, err := ParseJSON(input)
Expect(err).NotTo(HaveOccurred())
Expect(result).To(Equal(expected))
})
})
})

View File

@@ -98,6 +98,9 @@ var knownModelsNameSuffixToSkip []string = []string{
".yaml",
".yml",
".json",
".txt",
".md",
".MD",
".DS_Store",
".",
".partial",