mincomk/risuai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Enhance LLM model flags to include requiresAlternateRole and update request processing to handle multimodal and thoughts data

Browse Source
This commit is contained in:
Kwaroran
2024-12-20 04:31:23 +09:00
parent 280fd31316
commit cc4bbdc8e2
2 changed files with 33 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
src/ts/model/modellist.ts
View File

@@ -778,7 +778,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-1.5-pro-exp-0827', id: 'gemini-1.5-pro-exp-0827',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -787,7 +787,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-exp-1121', id: 'gemini-exp-1121',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud, tokenizer: LLMTokenizer.GoogleCloud,
}, },
@@ -796,7 +796,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-exp-1206', id: 'gemini-exp-1206',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -805,7 +805,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-2.0-flash-exp', id: 'gemini-2.0-flash-exp',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasAudioInput, LLMFlags.hasVideoInput, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasAudioInput, LLMFlags.hasVideoInput, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'], parameters: ['temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'],
tokenizer: LLMTokenizer.GoogleCloud, tokenizer: LLMTokenizer.GoogleCloud,
recommended: true recommended: true
@@ -815,7 +815,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-2.0-flash-thinking-exp-1219', id: 'gemini-2.0-flash-thinking-exp-1219',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasAudioInput, LLMFlags.hasVideoInput, LLMFlags.hasStreaming, LLMFlags.geminiThinking], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported, LLMFlags.hasAudioInput, LLMFlags.hasVideoInput, LLMFlags.hasStreaming, LLMFlags.geminiThinking, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'], parameters: ['temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'],
tokenizer: LLMTokenizer.GoogleCloud, tokenizer: LLMTokenizer.GoogleCloud,
recommended: true recommended: true
@@ -825,7 +825,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-1.5-pro-latest', id: 'gemini-1.5-pro-latest',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
recommended: true, recommended: true,
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
@@ -835,7 +835,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-1.5-flash', id: 'gemini-1.5-flash',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
recommended: true, recommended: true,
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
@@ -846,7 +846,7 @@ export const LLMModels: LLMModel[] = [
internalID: 'gemini-exp-1121', internalID: 'gemini-exp-1121',
provider: LLMProvider.VertexAI, provider: LLMProvider.VertexAI,
format: LLMFormat.VertexAIGemini, format: LLMFormat.VertexAIGemini,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.Gemma tokenizer: LLMTokenizer.Gemma
}, },
@@ -856,7 +856,7 @@ export const LLMModels: LLMModel[] = [
internalID: 'gemini-1.5-pro-latest', internalID: 'gemini-1.5-pro-latest',
provider: LLMProvider.VertexAI, provider: LLMProvider.VertexAI,
format: LLMFormat.VertexAIGemini, format: LLMFormat.VertexAIGemini,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.Gemma tokenizer: LLMTokenizer.Gemma
}, },
@@ -866,7 +866,7 @@ export const LLMModels: LLMModel[] = [
internalID: 'gemini-1.5-flash', internalID: 'gemini-1.5-flash',
provider: LLMProvider.VertexAI, provider: LLMProvider.VertexAI,
format: LLMFormat.VertexAIGemini, format: LLMFormat.VertexAIGemini,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.Gemma tokenizer: LLMTokenizer.Gemma
}, },
@@ -875,7 +875,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-exp-1114', id: 'gemini-exp-1114',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -884,7 +884,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-1.5-pro-002', id: 'gemini-1.5-pro-002',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -893,7 +893,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-1.5-flash-002', id: 'gemini-1.5-flash-002',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -902,7 +902,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-pro', id: 'gemini-pro',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -911,7 +911,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-pro-vision', id: 'gemini-pro-vision',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -920,7 +920,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-ultra', id: 'gemini-ultra',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },
@@ -929,7 +929,7 @@ export const LLMModels: LLMModel[] = [
id: 'gemini-ultra-vision', id: 'gemini-ultra-vision',
provider: LLMProvider.GoogleCloud, provider: LLMProvider.GoogleCloud,
format: LLMFormat.GoogleCloud, format: LLMFormat.GoogleCloud,
flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming], flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.hasStreaming, LLMFlags.requiresAlternateRole],
parameters: ['temperature', 'top_k', 'top_p'], parameters: ['temperature', 'top_k', 'top_p'],
tokenizer: LLMTokenizer.GoogleCloud tokenizer: LLMTokenizer.GoogleCloud
}, },

16
src/ts/process/request.ts
View File

@@ -282,7 +282,23 @@ function reformater(formated:OpenAIChat[],modelInfo:LLMModel){
} }
if(newFormated[newFormated.length-1].role === m.role){ if(newFormated[newFormated.length-1].role === m.role){
newFormated[newFormated.length-1].content += '\n' + m.content newFormated[newFormated.length-1].content += '\n' + m.content
if(m.multimodals){
if(!newFormated[newFormated.length-1].multimodals){
newFormated[newFormated.length-1].multimodals = []
}
newFormated[newFormated.length-1].multimodals.push(...m.multimodals)
}
if(m.thoughts){
if(!newFormated[newFormated.length-1].thoughts){
newFormated[newFormated.length-1].thoughts = []
}
newFormated[newFormated.length-1].thoughts.push(...m.thoughts)
}
continue continue
} }
else{ else{