phoenix-oss/llama-stack
1
0
Fork 0
forked from phoenix-oss/llama-stack-mirror
Code Pull requests Releases Packages 2 Activity Actions

LocalInferenceImpl update for LS 0.1 (#911)

Browse source 
# What does this PR do?

To work with the updated iOSCalendarAssistantWithLocalInf
[here](https://github.com/meta-llama/llama-stack-apps/compare/ios_local).

In short, provide a summary of what this PR does and why. Usually, the
relevant context should be present in a linked issue.

- [ ] Addresses issue (#issue)


## Test Plan

Please describe:
 - tests you ran to verify your changes with result summaries.
 - provide instructions so it can be reproduced.


## Sources

Please link relevant resources if necessary.


## Before submitting

- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Ran pre-commit to handle lint / formatting issues.
- [ ] Read the [contributor
guideline](https://github.com/meta-llama/llama-stack/blob/main/CONTRIBUTING.md),
      Pull Request section?
- [ ] Updated relevant documentation.
- [ ] Wrote necessary unit or integration tests.
This commit is contained in:
Jeff Tang 2025-02-02 09:49:40 -08:00 • committed by GitHub
parent 7fdbd5b642
commit 587753da2f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 53 additions and 37 deletions
Download patch file Download diff file Expand all files Collapse all files

46
llama_stack/providers/inline/ios/inference/LocalInferenceImpl/LocalInference.swift
View file

@ -40,7 +40,7 @@ public class LocalInference: Inference {
public func chatCompletion(request: Components.Schemas.ChatCompletionRequest) -> AsyncStream<Components.Schemas.ChatCompletionResponseStreamChunk> { public func chatCompletion(request: Components.Schemas.ChatCompletionRequest) -> AsyncStream<Components.Schemas.ChatCompletionResponseStreamChunk> {
return AsyncStream { continuation in return AsyncStream { continuation in
runnerQueue.async { let workItem = DispatchWorkItem {
do { do {
var tokens: [String] = [] var tokens: [String] = []
@ -69,9 +69,10 @@ public class LocalInference: Inference {
continuation.yield( continuation.yield(
Components.Schemas.ChatCompletionResponseStreamChunk( Components.Schemas.ChatCompletionResponseStreamChunk(
event: Components.Schemas.ChatCompletionResponseEvent( event: Components.Schemas.ChatCompletionResponseEvent(
delta: .ToolCallDelta(Components.Schemas.ToolCallDelta( delta: .tool_call(Components.Schemas.ToolCallDelta(
content: .case1(""), parse_status: Components.Schemas.ToolCallParseStatus.started,
parse_status: Components.Schemas.ToolCallParseStatus.started tool_call: .case1(""),
_type: Components.Schemas.ToolCallDelta._typePayload.tool_call
) )
), ),
event_type: .progress event_type: .progress
@ -95,14 +96,18 @@ public class LocalInference: Inference {
text = token text = token
} }
var delta: Components.Schemas.ChatCompletionResponseEvent.deltaPayload var delta: Components.Schemas.ContentDelta
if ipython { if ipython {
delta = .ToolCallDelta(Components.Schemas.ToolCallDelta( delta = .tool_call(Components.Schemas.ToolCallDelta(
content: .case1(text), parse_status: .in_progress,
parse_status: .in_progress tool_call: .case1(text),
_type: .tool_call
)) ))
} else { } else {
delta = .case1(text) delta = .text(Components.Schemas.TextDelta(
text: text,
_type: Components.Schemas.TextDelta._typePayload.text)
)
} }
if stopReason == nil { if stopReason == nil {
@ -129,7 +134,12 @@ public class LocalInference: Inference {
continuation.yield( continuation.yield(
Components.Schemas.ChatCompletionResponseStreamChunk( Components.Schemas.ChatCompletionResponseStreamChunk(
event: Components.Schemas.ChatCompletionResponseEvent( event: Components.Schemas.ChatCompletionResponseEvent(
delta: .ToolCallDelta(Components.Schemas.ToolCallDelta(content: .case1(""), parse_status: .failure)), delta: .tool_call(Components.Schemas.ToolCallDelta(
parse_status: Components.Schemas.ToolCallParseStatus.failed,
tool_call: .case1(""),
_type: Components.Schemas.ToolCallDelta._typePayload.tool_call
)
),
event_type: .progress event_type: .progress
) )
// TODO: stopReason // TODO: stopReason
@ -141,10 +151,12 @@ public class LocalInference: Inference {
continuation.yield( continuation.yield(
Components.Schemas.ChatCompletionResponseStreamChunk( Components.Schemas.ChatCompletionResponseStreamChunk(
event: Components.Schemas.ChatCompletionResponseEvent( event: Components.Schemas.ChatCompletionResponseEvent(
delta: .ToolCallDelta(Components.Schemas.ToolCallDelta( delta: .tool_call(Components.Schemas.ToolCallDelta(
content: .ToolCall(toolCall), parse_status: Components.Schemas.ToolCallParseStatus.succeeded,
parse_status: .success tool_call: Components.Schemas.ToolCallDelta.tool_callPayload.ToolCall(toolCall),
)), _type: Components.Schemas.ToolCallDelta._typePayload.tool_call
)
),
event_type: .progress event_type: .progress
) )
// TODO: stopReason // TODO: stopReason
@ -155,7 +167,10 @@ public class LocalInference: Inference {
continuation.yield( continuation.yield(
Components.Schemas.ChatCompletionResponseStreamChunk( Components.Schemas.ChatCompletionResponseStreamChunk(
event: Components.Schemas.ChatCompletionResponseEvent( event: Components.Schemas.ChatCompletionResponseEvent(
delta: .case1(""), delta: .text(Components.Schemas.TextDelta(
text: "",
_type: Components.Schemas.TextDelta._typePayload.text)
),
event_type: .complete event_type: .complete
) )
// TODO: stopReason // TODO: stopReason
@ -166,6 +181,7 @@ public class LocalInference: Inference {
print("Inference error: " + error.localizedDescription) print("Inference error: " + error.localizedDescription)
} }
} }
runnerQueue.async(execute: workItem)
} }
} }
} }

44
llama_stack/providers/inline/ios/inference/LocalInferenceImpl/Parsing.swift
View file

@ -6,7 +6,7 @@ func encodeHeader(role: String) -> String {
return "<|start_header_id|>\(role)<|end_header_id|>\n\n" return "<|start_header_id|>\(role)<|end_header_id|>\n\n"
} }
func encodeDialogPrompt(messages: [Components.Schemas.ChatCompletionRequest.messagesPayloadPayload]) -> String { func encodeDialogPrompt(messages: [Components.Schemas.Message]) -> String {
var prompt = "" var prompt = ""
prompt.append("<|begin_of_text|>") prompt.append("<|begin_of_text|>")
@ -20,24 +20,24 @@ func encodeDialogPrompt(messages: [Components.Schemas.ChatCompletionRequest.mess
return prompt return prompt
} }
func getRole(message: Components.Schemas.ChatCompletionRequest.messagesPayloadPayload) -> String { func getRole(message: Components.Schemas.Message) -> String {
switch (message) { switch (message) {
case .UserMessage(let m): case .user(let m):
return m.role.rawValue return m.role.rawValue
case .SystemMessage(let m): case .system(let m):
return m.role.rawValue return m.role.rawValue
case .ToolResponseMessage(let m): case .tool(let m):
return m.role.rawValue return m.role.rawValue
case .CompletionMessage(let m): case .assistant(let m):
return m.role.rawValue return m.role.rawValue
} }
} }
func encodeMessage(message: Components.Schemas.ChatCompletionRequest.messagesPayloadPayload) -> String { func encodeMessage(message: Components.Schemas.Message) -> String {
var prompt = encodeHeader(role: getRole(message: message)) var prompt = encodeHeader(role: getRole(message: message))
switch (message) { switch (message) {
case .CompletionMessage(let m): case .assistant(let m):
if (m.tool_calls.count > 0) { if (m.tool_calls.count > 0) {
prompt += "<|python_tag|>" prompt += "<|python_tag|>"
} }
@ -64,37 +64,37 @@ func encodeMessage(message: Components.Schemas.ChatCompletionRequest.messagesPay
} }
switch (message) { switch (message) {
case .UserMessage(let m): case .user(let m):
prompt += _processContent(m.content) prompt += _processContent(m.content)
case .SystemMessage(let m): case .system(let m):
prompt += _processContent(m.content) prompt += _processContent(m.content)
case .ToolResponseMessage(let m): case .tool(let m):
prompt += _processContent(m.content) prompt += _processContent(m.content)
case .CompletionMessage(let m): case .assistant(let m):
prompt += _processContent(m.content) prompt += _processContent(m.content)
} }
var eom = false var eom = false
switch (message) { switch (message) {
case .UserMessage(let m): case .user(let m):
switch (m.content) { switch (m.content) {
case .case1(let c): case .case1(let c):
prompt += _processContent(c) prompt += _processContent(c)
case .ImageMedia(let c): case .InterleavedContentItem(let c):
prompt += _processContent(c) prompt += _processContent(c)
case .case3(let c): case .case3(let c):
prompt += _processContent(c) prompt += _processContent(c)
} }
case .CompletionMessage(let m): case .assistant(let m):
// TODO: Support encoding past tool call history // TODO: Support encoding past tool call history
// for t in m.tool_calls { // for t in m.tool_calls {
// _processContent(t.) // _processContent(t.)
//} //}
eom = m.stop_reason == Components.Schemas.StopReason.end_of_message eom = m.stop_reason == Components.Schemas.StopReason.end_of_message
case .SystemMessage(_): case .system(_):
break break
case .ToolResponseMessage(_): case .tool(_):
break break
} }
@ -107,12 +107,12 @@ func encodeMessage(message: Components.Schemas.ChatCompletionRequest.messagesPay
return prompt return prompt
} }
func prepareMessages(request: Components.Schemas.ChatCompletionRequest) throws -> [Components.Schemas.ChatCompletionRequest.messagesPayloadPayload] { func prepareMessages(request: Components.Schemas.ChatCompletionRequest) throws -> [Components.Schemas.Message] {
var existingMessages = request.messages var existingMessages = request.messages
var existingSystemMessage: Components.Schemas.ChatCompletionRequest.messagesPayloadPayload? var existingSystemMessage: Components.Schemas.Message?
// TODO: Existing system message // TODO: Existing system message
var messages: [Components.Schemas.ChatCompletionRequest.messagesPayloadPayload] = [] var messages: [Components.Schemas.Message] = []
let defaultGen = SystemDefaultGenerator() let defaultGen = SystemDefaultGenerator()
let defaultTemplate = defaultGen.gen() let defaultTemplate = defaultGen.gen()
@ -123,7 +123,7 @@ func prepareMessages(request: Components.Schemas.ChatCompletionRequest) throws -
sysContent += try defaultTemplate.render() sysContent += try defaultTemplate.render()
messages.append(.SystemMessage(Components.Schemas.SystemMessage( messages.append(.system(Components.Schemas.SystemMessage(
content: .case1(sysContent), content: .case1(sysContent),
role: .system)) role: .system))
) )
@ -133,7 +133,7 @@ func prepareMessages(request: Components.Schemas.ChatCompletionRequest) throws -
let toolGen = FunctionTagCustomToolGenerator() let toolGen = FunctionTagCustomToolGenerator()
let toolTemplate = try toolGen.gen(customTools: request.tools!) let toolTemplate = try toolGen.gen(customTools: request.tools!)
let tools = try toolTemplate.render() let tools = try toolTemplate.render()
messages.append(.UserMessage(Components.Schemas.UserMessage( messages.append(.user(Components.Schemas.UserMessage(
content: .case1(tools), content: .case1(tools),
role: .user) role: .user)
)) ))