feat(functions): relax mixedgrammars (#2365)

* feat(functions): relax mixedgrammars

Extend even more the functionalities and when mixed mode is enabled,
tolerate also both strings and JSON in the result - in this case we make
sure that the JSON can be correctly parsed.

This also updates the examples and the gallery model to configure the
grammar.

The changeset also breaks current function/grammar configuration as it
reserves now a stanza in the YAML config.

For example:

```yaml
function:
  grammar:
    # This allows the grammar to also return messages
    mixed_mode: true
    # Suffix to add to the grammar
    # prefix: '<tool_call>\n'
    # Force parallel calls in the grammar
    # parallel_calls: true
```

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

* refactor, add a way to disable mixed json and freestring

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

* Fix linting issues

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

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2024-05-22 00:14:16 +02:00 committed by GitHub
parent 1542c58466
commit 491e1d752b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
10 changed files with 469 additions and 149 deletions

View File

@ -3,7 +3,61 @@ mmap: true
parameters:
model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
stopwords:
- "<|im_end|>"
- "<dummy32000>"
- "</tool_call>"
- "<|eot_id|>"
- "<|end_of_text|>"
function:
# disable injecting the "answer" tool
disable_no_action: true
grammar:
# This allows the grammar to also return messages
mixed_mode: true
# Suffix to add to the grammar
#prefix: '<tool_call>\n'
# Force parallel calls in the grammar
# parallel_calls: true
return_name_in_function_response: true
# Without grammar uncomment the lines below
# Warning: this is relying only on the capability of the
# LLM model to generate the correct function call.
json_regex_match:
- "(?s)<tool_call>(.*?)</tool_call>"
- "(?s)<tool_call>(.*?)"
replace_llm_results:
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
replace_function_results:
# Replace everything that is not JSON array or object
#
- key: '(?s)^[^{\[]*'
value: ""
- key: '(?s)[^}\]]*$'
value: ""
- key: "'([^']*?)'"
value: "_DQUOTE_${1}_DQUOTE_"
- key: '\\"'
value: "__TEMP_QUOTE__"
- key: "\'"
value: "'"
- key: "_DQUOTE_"
value: '"'
- key: "__TEMP_QUOTE__"
value: '"'
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
template:
chat: |
{{.Input -}}
<|im_start|>assistant
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
{{- if .FunctionCall }}
@ -22,38 +76,25 @@ template:
{{- else if eq .RoleName "tool" }}
</tool_response>
{{- end }}<|im_end|>
# https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B-GGUF#prompt-format-for-function-calling
function: |
completion: |
{{.Input}}
function: |-
<|im_start|>system
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
You are a function calling AI model.
Here are the available tools:
<tools>
{{range .Functions}}
{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
{{end}}
</tools>
Use the following pydantic model json schema for each tool call you will make:
{'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
You should call the tools provided to you sequentially
Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
<scratchpad>
{step-by-step reasoning and plan in bullet points}
</scratchpad>
For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
<tool_call>
{'arguments': <args-dict>, 'name': <function-name>}
{"arguments": <args-dict>, "name": <function-name>}
</tool_call><|im_end|>
{{.Input -}}
<|im_start|>assistant
<tool_call>
chat: |
{{.Input -}}
<|im_start|>assistant
completion: |
{{.Input}}
context_size: 4096
f16: true
stopwords:
- <|im_end|>
- <dummy32000>
- "\n</tool_call>"
- "\n\n\n"
usage: |
curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
"model": "gpt-4",
"messages": [{"role": "user", "content": "How are you doing?", "temperature": 0.1}]
}'

View File

@ -3,7 +3,61 @@ mmap: true
parameters:
model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
stopwords:
- "<|im_end|>"
- "<dummy32000>"
- "</tool_call>"
- "<|eot_id|>"
- "<|end_of_text|>"
function:
# disable injecting the "answer" tool
disable_no_action: true
grammar:
# This allows the grammar to also return messages
mixed_mode: true
# Suffix to add to the grammar
#prefix: '<tool_call>\n'
# Force parallel calls in the grammar
# parallel_calls: true
return_name_in_function_response: true
# Without grammar uncomment the lines below
# Warning: this is relying only on the capability of the
# LLM model to generate the correct function call.
json_regex_match:
- "(?s)<tool_call>(.*?)</tool_call>"
- "(?s)<tool_call>(.*?)"
replace_llm_results:
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
replace_function_results:
# Replace everything that is not JSON array or object
#
- key: '(?s)^[^{\[]*'
value: ""
- key: '(?s)[^}\]]*$'
value: ""
- key: "'([^']*?)'"
value: "_DQUOTE_${1}_DQUOTE_"
- key: '\\"'
value: "__TEMP_QUOTE__"
- key: "\'"
value: "'"
- key: "_DQUOTE_"
value: '"'
- key: "__TEMP_QUOTE__"
value: '"'
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
template:
chat: |
{{.Input -}}
<|im_start|>assistant
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
{{- if .FunctionCall }}
@ -22,38 +76,25 @@ template:
{{- else if eq .RoleName "tool" }}
</tool_response>
{{- end }}<|im_end|>
# https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B-GGUF#prompt-format-for-function-calling
function: |
completion: |
{{.Input}}
function: |-
<|im_start|>system
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
You are a function calling AI model.
Here are the available tools:
<tools>
{{range .Functions}}
{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
{{end}}
</tools>
Use the following pydantic model json schema for each tool call you will make:
{'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
You should call the tools provided to you sequentially
Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
<scratchpad>
{step-by-step reasoning and plan in bullet points}
</scratchpad>
For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
<tool_call>
{'arguments': <args-dict>, 'name': <function-name>}
{"arguments": <args-dict>, "name": <function-name>}
</tool_call><|im_end|>
{{.Input -}}
<|im_start|>assistant
<tool_call>
chat: |
{{.Input -}}
<|im_start|>assistant
completion: |
{{.Input}}
context_size: 4096
f16: true
stopwords:
- <|im_end|>
- <dummy32000>
- "\n</tool_call>"
- "\n\n\n"
usage: |
curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
"model": "gpt-4",
"messages": [{"role": "user", "content": "How are you doing?", "temperature": 0.1}]
}'
<|im_start|>assistant

View File

@ -4,7 +4,61 @@ f16: false
parameters:
model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
stopwords:
- "<|im_end|>"
- "<dummy32000>"
- "</tool_call>"
- "<|eot_id|>"
- "<|end_of_text|>"
function:
# disable injecting the "answer" tool
disable_no_action: true
grammar:
# This allows the grammar to also return messages
mixed_mode: true
# Suffix to add to the grammar
#prefix: '<tool_call>\n'
# Force parallel calls in the grammar
# parallel_calls: true
return_name_in_function_response: true
# Without grammar uncomment the lines below
# Warning: this is relying only on the capability of the
# LLM model to generate the correct function call.
json_regex_match:
- "(?s)<tool_call>(.*?)</tool_call>"
- "(?s)<tool_call>(.*?)"
replace_llm_results:
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
replace_function_results:
# Replace everything that is not JSON array or object
#
- key: '(?s)^[^{\[]*'
value: ""
- key: '(?s)[^}\]]*$'
value: ""
- key: "'([^']*?)'"
value: "_DQUOTE_${1}_DQUOTE_"
- key: '\\"'
value: "__TEMP_QUOTE__"
- key: "\'"
value: "'"
- key: "_DQUOTE_"
value: '"'
- key: "__TEMP_QUOTE__"
value: '"'
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
template:
chat: |
{{.Input -}}
<|im_start|>assistant
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
{{- if .FunctionCall }}
@ -23,37 +77,25 @@ template:
{{- else if eq .RoleName "tool" }}
</tool_response>
{{- end }}<|im_end|>
# https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B-GGUF#prompt-format-for-function-calling
function: |
completion: |
{{.Input}}
function: |-
<|im_start|>system
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
You are a function calling AI model.
Here are the available tools:
<tools>
{{range .Functions}}
{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
{{end}}
</tools>
Use the following pydantic model json schema for each tool call you will make:
{'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
You should call the tools provided to you sequentially
Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
<scratchpad>
{step-by-step reasoning and plan in bullet points}
</scratchpad>
For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
<tool_call>
{'arguments': <args-dict>, 'name': <function-name>}
{"arguments": <args-dict>, "name": <function-name>}
</tool_call><|im_end|>
{{.Input -}}
<|im_start|>assistant
<tool_call>
chat: |
{{.Input -}}
<|im_start|>assistant
completion: |
{{.Input}}
context_size: 4096
stopwords:
- <|im_end|>
- "\n</tool_call>"
- <dummy32000>
- "\n\n\n"
usage: |
curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
"model": "gpt-4",
"messages": [{"role": "user", "content": "How are you doing?", "temperature": 0.1}]
}'

View File

@ -67,6 +67,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
return true
})
result = functions.CleanupLLMResult(result, config.FunctionsConfig)
results := functions.ParseFunctionCall(result, config.FunctionsConfig)
noActionToRun := len(results) > 0 && results[0].Name == noAction || len(results) == 0
@ -192,7 +193,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
}
switch {
case !config.FunctionsConfig.NoGrammar && shouldUseFn:
case !config.FunctionsConfig.GrammarConfig.NoGrammar && shouldUseFn:
noActionGrammar := functions.Function{
Name: noActionName,
Description: noActionDescription,
@ -219,15 +220,15 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
// Handle if we should return "name" instead of "functions"
if config.FunctionsConfig.FunctionName {
jsStruct := funcs.ToJSONNameStructure()
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarPrefix, "", config.FunctionsConfig.ParallelCalls, config.FunctionsConfig.GrammarMessage)
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
} else {
jsStruct := funcs.ToJSONFunctionStructure()
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarPrefix, "", config.FunctionsConfig.ParallelCalls, config.FunctionsConfig.GrammarMessage)
config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
}
case input.JSONFunctionGrammarObject != nil:
config.Grammar = input.JSONFunctionGrammarObject.Grammar(config.FunctionsConfig.GrammarPrefix, "", config.FunctionsConfig.ParallelCalls, config.FunctionsConfig.GrammarMessage)
config.Grammar = input.JSONFunctionGrammarObject.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
case input.JSONFunctionGrammarObjectName != nil:
config.Grammar = input.JSONFunctionGrammarObjectName.Grammar(config.FunctionsConfig.GrammarPrefix, "", config.FunctionsConfig.ParallelCalls, config.FunctionsConfig.GrammarMessage)
config.Grammar = input.JSONFunctionGrammarObjectName.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
default:
// Force picking one of the functions by the request
if config.FunctionToCall() != "" {
@ -470,6 +471,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
return
}
s = functions.CleanupLLMResult(s, config.FunctionsConfig)
results := functions.ParseFunctionCall(s, config.FunctionsConfig)
noActionsToRun := len(results) > 0 && results[0].Name == noActionName || len(results) == 0

View File

@ -3,7 +3,55 @@ name: "hermes-2-pro-mistral"
config_file: |
mmap: true
stopwords:
- "<|im_end|>"
- "<dummy32000>"
- "</tool_call>"
- "<|eot_id|>"
- "<|end_of_text|>"
function:
# disable injecting the "answer" tool
disable_no_action: true
grammar:
# This allows the grammar to also return messages
mixed_mode: true
return_name_in_function_response: true
# Without grammar uncomment the lines below
# Warning: this is relying only on the capability of the
# LLM model to generate the correct function call.
json_regex_match:
- "(?s)<tool_call>(.*?)</tool_call>"
- "(?s)<tool_call>(.*?)"
replace_llm_results:
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
replace_function_results:
# Replace everything that is not JSON array or object
- key: '(?s)^[^{\[]*'
value: ""
- key: '(?s)[^}\]]*$'
value: ""
- key: "'([^']*?)'"
value: "_DQUOTE_${1}_DQUOTE_"
- key: '\\"'
value: "__TEMP_QUOTE__"
- key: "\'"
value: "'"
- key: "_DQUOTE_"
value: '"'
- key: "__TEMP_QUOTE__"
value: '"'
# Drop the scratchpad content from responses
- key: "(?s)<scratchpad>.*</scratchpad>"
value: ""
template:
chat: |
{{.Input -}}
<|im_start|>assistant
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
{{- if .FunctionCall }}
@ -22,31 +70,25 @@ config_file: |
{{- else if eq .RoleName "tool" }}
</tool_response>
{{- end }}<|im_end|>
# https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B-GGUF#prompt-format-for-function-calling
function: |
completion: |
{{.Input}}
function: |-
<|im_start|>system
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
You are a function calling AI model.
Here are the available tools:
<tools>
{{range .Functions}}
{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
{{end}}
</tools>
Use the following pydantic model json schema for each tool call you will make:
{'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
You should call the tools provided to you sequentially
Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
<scratchpad>
{step-by-step reasoning and plan in bullet points}
</scratchpad>
For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
<tool_call>
{'arguments': <args-dict>, 'name': <function-name>}
{"arguments": <args-dict>, "name": <function-name>}
</tool_call><|im_end|>
{{.Input -}}
<|im_start|>assistant
<tool_call>
chat: |
{{.Input -}}
<|im_start|>assistant
completion: |
{{.Input}}
context_size: 4096
f16: true
stopwords:
- <|im_end|>
- '</tool_call>'

View File

@ -117,7 +117,16 @@ const array = `arr ::=
(",\n" realvalue)*
)? "]"`
func (sc *JSONSchemaConverter) finalizeGrammar(suffix string, maybeArray, maybeString bool) string {
func (sc *JSONSchemaConverter) finalizeGrammar(options ...func(*GrammarOption)) string {
grammarOpts := &GrammarOption{}
grammarOpts.Apply(options...)
suffix := grammarOpts.Suffix
maybeArray := grammarOpts.MaybeArray
maybeString := grammarOpts.MaybeString
noMixedFreeString := grammarOpts.NoMixedFreeString
var lines []string
swapRoot := maybeArray || maybeString || suffix != ""
@ -140,6 +149,11 @@ func (sc *JSONSchemaConverter) finalizeGrammar(suffix string, maybeArray, maybeS
newRoot = "arr | realvalue"
}
freestringRule := "mixedstring"
if noMixedFreeString {
freestringRule = "freestring"
}
if suffix != "" {
// quote newlines in suffix
suffix = utils.EscapeNewLines(suffix)
@ -150,7 +164,7 @@ func (sc *JSONSchemaConverter) finalizeGrammar(suffix string, maybeArray, maybeS
if maybeString {
//newRoot = "( (\"" + suffix + "\" " + newRoot + ") | freestring ) "
newRoot = "( \"" + suffix + "\" " + newRoot + " | freestring ) "
newRoot = "( \"" + suffix + "\" " + newRoot + " | " + freestringRule + " ) "
} else {
newRoot = "\"" + suffix + "\" " + "" + newRoot + ""
}
@ -159,12 +173,18 @@ func (sc *JSONSchemaConverter) finalizeGrammar(suffix string, maybeArray, maybeS
// newRoot = "(" + newRoot + ")"
}
newRoot = "freestring | " + newRoot
newRoot = freestringRule + " | " + newRoot
}
lines = append(lines, fmt.Sprintf("%s ::= %s", "root", newRoot))
lines = append(lines, array)
if maybeArray {
lines = append(lines, `mixedstring ::= freestring | freestring arr | freestring realvalue | realvalue | arr`)
} else {
lines = append(lines, `mixedstring ::= freestring | freestring realvalue | realvalue`)
}
return strings.Join(lines, "\n")
}
@ -289,16 +309,16 @@ func (sc *JSONSchemaConverter) resolveReference(ref string, rootSchema map[strin
return def
}
func (sc *JSONSchemaConverter) Grammar(suffix string, schema map[string]interface{}, maybeArray, maybeString bool) string {
func (sc *JSONSchemaConverter) Grammar(schema map[string]interface{}, options ...func(*GrammarOption)) string {
sc.addRule("freestring", PRIMITIVE_RULES["freestring"])
sc.visit(schema, "", schema)
return sc.finalizeGrammar(suffix, maybeArray, maybeString)
return sc.finalizeGrammar(options...)
}
func (sc *JSONSchemaConverter) GrammarFromBytes(suffix string, b []byte, maybeArray, maybeString bool) string {
func (sc *JSONSchemaConverter) GrammarFromBytes(b []byte, options ...func(*GrammarOption)) string {
var schema map[string]interface{}
_ = json.Unmarshal(b, &schema)
return sc.Grammar(suffix, schema, maybeArray, maybeString)
return sc.Grammar(schema, options...)
}
func jsonString(v interface{}) string {
@ -341,9 +361,12 @@ type JSONFunctionStructureName struct {
Defs map[string]interface{} `json:"$defs,omitempty"`
}
func (j JSONFunctionStructureName) Grammar(suffix string, propOrder string, maybeArray, maybeString bool) string {
func (j JSONFunctionStructureName) Grammar(options ...func(*GrammarOption)) string {
grammarOpts := &GrammarOption{}
grammarOpts.Apply(options...)
dat, _ := json.Marshal(j)
return NewJSONSchemaConverter(propOrder).GrammarFromBytes(suffix, dat, maybeArray, maybeString)
return NewJSONSchemaConverter(grammarOpts.PropOrder).GrammarFromBytes(dat, options...)
}
type JSONFunctionStructureFunction struct {
@ -352,7 +375,10 @@ type JSONFunctionStructureFunction struct {
Defs map[string]interface{} `json:"$defs,omitempty"`
}
func (j JSONFunctionStructureFunction) Grammar(suffix string, propOrder string, maybeArray, maybeString bool) string {
func (j JSONFunctionStructureFunction) Grammar(options ...func(*GrammarOption)) string {
grammarOpts := &GrammarOption{}
grammarOpts.Apply(options...)
dat, _ := json.Marshal(j)
return NewJSONSchemaConverter(propOrder).GrammarFromBytes(suffix, dat, maybeArray, maybeString)
return NewJSONSchemaConverter(grammarOpts.PropOrder).GrammarFromBytes(dat, options...)
}

View File

@ -3,6 +3,7 @@ package functions_test
import (
"strings"
"github.com/go-skynet/LocalAI/pkg/functions"
. "github.com/go-skynet/LocalAI/pkg/functions"
. "github.com/onsi/ginkgo/v2"
. "github.com/onsi/gomega"
@ -248,7 +249,7 @@ root-1-name ::= "\"search\""`
var _ = Describe("JSON schema grammar tests", func() {
Context("JSON", func() {
It("generates a valid grammar from JSON schema", func() {
grammar := NewJSONSchemaConverter("").GrammarFromBytes("", []byte(testInput1), false, false)
grammar := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput1))
results := strings.Split(inputResult1, "\n")
for _, r := range results {
if r != "" {
@ -258,7 +259,7 @@ var _ = Describe("JSON schema grammar tests", func() {
Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))))
})
It("generates a valid grammar from JSON schema", func() {
grammar := NewJSONSchemaConverter("").GrammarFromBytes("", []byte(testInput2), false, false)
grammar := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput2))
results := strings.Split(inputResult3, "\n")
for _, r := range results {
if r != "" {
@ -272,7 +273,7 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureFunction{
OneOf: testFunctions}
grammar := structuredGrammar.Grammar("", "", false, false)
grammar := structuredGrammar.Grammar()
results := strings.Split(inputResult1, "\n")
for _, r := range results {
if r != "" {
@ -286,8 +287,12 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureFunction{
OneOf: testFunctions}
grammar := structuredGrammar.Grammar("", "", true, false)
results := strings.Split(inputResult2, "\n")
grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
results := strings.Split(
strings.Join([]string{
inputResult2,
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -300,8 +305,12 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("", "", true, false)
results := strings.Split(inputResult4, "\n")
grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
results := strings.Split(
strings.Join([]string{
inputResult4,
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -314,8 +323,15 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("suffix", "", true, false)
results := strings.Split(rootResult(`"suffix" arr | realvalue`), "\n")
grammar := structuredGrammar.Grammar(
functions.SetPrefix("suffix"),
functions.EnableMaybeArray,
)
results := strings.Split(
strings.Join([]string{
rootResult(`"suffix" arr | realvalue`),
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -327,8 +343,12 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("suffix", "", false, false)
results := strings.Split(rootResult(`"suffix" realvalue`), "\n")
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"))
results := strings.Split(
strings.Join([]string{
rootResult(`"suffix" realvalue`),
"mixedstring ::= freestring | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -340,8 +360,12 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("suffix", "", false, true)
results := strings.Split(rootResult(`( "suffix" realvalue | freestring )`), "\n")
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString)
results := strings.Split(
strings.Join([]string{
rootResult(`( "suffix" realvalue | mixedstring )`),
"mixedstring ::= freestring | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -353,8 +377,13 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("suffix", "", true, true)
results := strings.Split(rootResult(`( "suffix" (arr | realvalue) | freestring )`), "\n")
grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString, functions.EnableMaybeArray)
results := strings.Split(
strings.Join([]string{
rootResult(`( "suffix" (arr | realvalue) | mixedstring )`),
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
@ -367,8 +396,30 @@ var _ = Describe("JSON schema grammar tests", func() {
structuredGrammar := JSONFunctionStructureName{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar("", "", true, true)
results := strings.Split(rootResult(`freestring | arr | realvalue`), "\n")
grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray)
results := strings.Split(
strings.Join([]string{
rootResult(`mixedstring | arr | realvalue`),
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))
}
}
Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))), grammar)
})
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{
OneOf: testFunctionsName}
grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray, functions.NoMixedFreeString)
results := strings.Split(
strings.Join([]string{
rootResult(`freestring | arr | realvalue`),
"mixedstring ::= freestring | freestring arr | freestring realvalue"}, "\n"),
"\n")
for _, r := range results {
if r != "" {
Expect(grammar).To(ContainSubstring(r))

33
pkg/functions/options.go Normal file
View File

@ -0,0 +1,33 @@
package functions
type GrammarOption struct {
PropOrder string
Suffix string
MaybeArray bool
MaybeString bool
NoMixedFreeString bool
}
func (o *GrammarOption) Apply(options ...func(*GrammarOption)) {
for _, l := range options {
l(o)
}
}
var EnableMaybeArray = func(o *GrammarOption) {
o.MaybeArray = true
}
var EnableMaybeString = func(o *GrammarOption) {
o.MaybeString = true
}
var NoMixedFreeString func(*GrammarOption) = func(o *GrammarOption) {
o.NoMixedFreeString = true
}
func SetPrefix(suffix string) func(*GrammarOption) {
return func(o *GrammarOption) {
o.Suffix = suffix
}
}

View File

@ -8,6 +8,26 @@ import (
"github.com/rs/zerolog/log"
)
type GrammarConfig struct {
// ParallelCalls enables the LLM to return multiple function calls in the same response
ParallelCalls bool `yaml:"parallel_calls"`
// MixedMode enables the LLM to return strings and not only JSON objects
// This is useful for models to not constraing returning only JSON and also messages back to the user
MixedMode bool `yaml:"mixed_mode"`
// NoMixedFreeString disables the mixed mode for free strings
// In this way if the LLM selects a free string, it won't be mixed necessarly with JSON objects
NoMixedFreeString bool `yaml:"no_mixed_free_string"`
// NoGrammar disables the grammar parsing and parses the responses directly from the LLM
NoGrammar bool `yaml:"disable"`
// Prefix is the suffix to append to the grammar when being generated
// This is useful when models prepend a tag before returning JSON
Prefix string `yaml:"prefix"`
}
// FunctionsConfig is the configuration for the tool/function call.
// It includes setting to map the function name and arguments from the response
// and, for instance, also if processing the requests with BNF grammars.
@ -16,34 +36,26 @@ type FunctionsConfig struct {
// By default we inject a tool that does nothing and is used to return an answer from the LLM
DisableNoAction bool `yaml:"disable_no_action"`
// Grammar is the configuration for the grammar
GrammarConfig GrammarConfig `yaml:"grammar"`
// NoActionFunctionName is the name of the function that does nothing. It defaults to "answer"
NoActionFunctionName string `yaml:"no_action_function_name"`
// NoActionDescriptionName is the name of the function that returns the description of the no action function
NoActionDescriptionName string `yaml:"no_action_description_name"`
// ParallelCalls enables the LLM to return multiple function calls in the same response
ParallelCalls bool `yaml:"parallel_calls"`
// GrammarMessage enables the LLM to return strings and not only JSON objects
// This is useful for models to not constraing returning only JSON and also messages back to the user
GrammarMessage bool `yaml:"grammar_message"`
// NoGrammar disables the grammar parsing and parses the responses directly from the LLM
NoGrammar bool `yaml:"no_grammar"`
// ResponseRegex is a named regex to extract the function name and arguments from the response
ResponseRegex string `yaml:"response_regex"`
// JSONRegexMatch is a regex to extract the JSON object from the response
JSONRegexMatch []string `yaml:"json_regex_match"`
// GrammarPrefix is the suffix to append to the grammar when being generated
// This is useful when models prepend a tag before returning JSON
GrammarPrefix string `yaml:"grammar_prefix"`
// ReplaceFunctionResults allow to replace strings in the results before parsing them
ReplaceFunctionResults []ReplaceResult `yaml:"replace_function_results"`
// ReplaceResults allow to replace strings in the results before parsing them
ReplaceResults []ReplaceResult `yaml:"replace_results"`
// ReplaceLLMResult allow to replace strings in the results before parsing them
ReplaceLLMResult []ReplaceResult `yaml:"replace_llm_results"`
// FunctionName enable the LLM to return { "name": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }
// instead of { "function": "function_name", "arguments": { "arg1": "value1", "arg2": "value2" } }.
@ -61,18 +73,49 @@ type FuncCallResults struct {
Arguments string
}
func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncCallResults {
func (g GrammarConfig) Options() []func(o *GrammarOption) {
opts := []func(o *GrammarOption){}
if g.MixedMode {
opts = append(opts, EnableMaybeString)
}
if g.ParallelCalls {
opts = append(opts, EnableMaybeArray)
}
if g.Prefix != "" {
opts = append(opts, SetPrefix(g.Prefix))
}
if g.NoMixedFreeString {
opts = append(opts, NoMixedFreeString)
}
return opts
}
func CleanupLLMResult(llmresult string, functionConfig FunctionsConfig) string {
log.Debug().Msgf("LLM result: %s", llmresult)
for _, item := range functionConfig.ReplaceResults {
for _, item := range functionConfig.ReplaceLLMResult {
k, v := item.Key, item.Value
log.Debug().Msgf("Replacing %s with %s", k, v)
re := regexp.MustCompile(k)
llmresult = re.ReplaceAllString(llmresult, v)
}
log.Debug().Msgf("LLM result(processed): %s", llmresult)
return llmresult
}
func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncCallResults {
log.Debug().Msgf("LLM result: %s", llmresult)
for _, item := range functionConfig.ReplaceFunctionResults {
k, v := item.Key, item.Value
log.Debug().Msgf("Replacing %s with %s", k, v)
re := regexp.MustCompile(k)
llmresult = re.ReplaceAllString(llmresult, v)
}
log.Debug().Msgf("LLM result(function cleanup): %s", llmresult)
functionNameKey := "function"
if functionConfig.FunctionName {
functionNameKey = "name"
@ -127,7 +170,6 @@ func ParseFunctionCall(llmresult string, functionConfig FunctionsConfig) []FuncC
// the response is a string that we have to parse
result := make(map[string]string)
if len(functionConfig.JSONRegexMatch) != 0 {
for _, r := range functionConfig.JSONRegexMatch {
// We use a regex to extract the JSON object from the response

View File

@ -120,7 +120,7 @@ Some text before the JSON
Some text after the JSON
`
functionConfig.ReplaceResults = []ReplaceResult{
functionConfig.ReplaceFunctionResults = []ReplaceResult{
{Key: `(?s)^[^{\[]*`, Value: ""},
{Key: `(?s)[^}\]]*$`, Value: ""},
}
@ -137,7 +137,7 @@ Some text before the JSON
[{"function": "add", "arguments": {"x": 5, "y": 3}}, {"function": "subtract", "arguments": {"x": 10, "y": 7}}]
Some text after the JSON
`
functionConfig.ReplaceResults = []ReplaceResult{
functionConfig.ReplaceFunctionResults = []ReplaceResult{
{Key: `(?s)^[^{\[]*`, Value: ""},
{Key: `(?s)[^}\]]*$`, Value: ""},
}
@ -163,7 +163,7 @@ Some text after the JSON
// Regex to match non-JSON characters after the JSON structure
//reAfter := regexp.MustCompile(`(?s)(?<=\}|\]).*$`)
functionConfig.ReplaceResults = []ReplaceResult{
functionConfig.ReplaceFunctionResults = []ReplaceResult{
{Key: `(?s)^[^{\[]*`, Value: ""},
{Key: `(?s)[^}\]]*$`, Value: ""},
// Regex pattern to match single quotes around keys and values
@ -196,7 +196,7 @@ Some text after the JSON
// Regex to match non-JSON characters after the JSON structure
//reAfter := regexp.MustCompile(`(?s)(?<=\}|\]).*$`)
functionConfig.ReplaceResults = []ReplaceResult{
functionConfig.ReplaceFunctionResults = []ReplaceResult{
{Key: `(?s)^[^{\[]*`, Value: ""},
{Key: `(?s)[^}\]]*$`, Value: ""},
// Regex pattern to match single quotes around keys and values