anything-llm/server/utils/AiProviders/fireworksAi/index.js
Timothy Carambat dd7c4675d3
LLM performance metric tracking ()
* WIP performance metric tracking

* fix: patch UI trying to .toFixed() null metric
Anthropic tracking migraiton
cleanup logs

* Apipie implmentation, not tested

* Cleanup Anthropic notes, Add support for AzureOpenAI tracking

* bedrock token metric tracking

* Cohere support

* feat: improve default stream handler to track for provider who are actually OpenAI compliant in usage reporting
add deepseek support

* feat: Add FireworksAI tracking reporting
fix: improve handler when usage:null is reported (why?)

* Add token reporting for GenericOpenAI

* token reporting for koboldcpp + lmstudio

* lint

* support Groq token tracking

* HF token tracking

* token tracking for togetherai

* LiteLLM token tracking

* linting + Mitral token tracking support

* XAI token metric reporting

* native provider runner

* LocalAI token tracking

* Novita token tracking

* OpenRouter token tracking

* Apipie stream metrics

* textwebgenui token tracking

* perplexity token reporting

* ollama token reporting

* lint

* put back comment

* Rip out LC ollama wrapper and use official library

* patch images with new ollama lib

* improve ollama offline message

* fix image handling in ollama llm provider

* lint

* NVIDIA NIM token tracking

* update openai compatbility responses

* UI/UX show/hide metrics on click for user preference

* update bedrock client

---------

Co-authored-by: shatfield4 <seanhatfield5@gmail.com>
2024-12-16 14:31:17 -08:00

157 lines
4.5 KiB
JavaScript

const { NativeEmbedder } = require("../../EmbeddingEngines/native");
const {
LLMPerformanceMonitor,
} = require("../../helpers/chat/LLMPerformanceMonitor");
const {
handleDefaultStreamResponseV2,
} = require("../../helpers/chat/responses");
function fireworksAiModels() {
const { MODELS } = require("./models.js");
return MODELS || {};
}
class FireworksAiLLM {
constructor(embedder = null, modelPreference = null) {
if (!process.env.FIREWORKS_AI_LLM_API_KEY)
throw new Error("No FireworksAI API key was set.");
const { OpenAI: OpenAIApi } = require("openai");
this.openai = new OpenAIApi({
baseURL: "https://api.fireworks.ai/inference/v1",
apiKey: process.env.FIREWORKS_AI_LLM_API_KEY ?? null,
});
this.model = modelPreference || process.env.FIREWORKS_AI_LLM_MODEL_PREF;
this.limits = {
history: this.promptWindowLimit() * 0.15,
system: this.promptWindowLimit() * 0.15,
user: this.promptWindowLimit() * 0.7,
};
this.embedder = !embedder ? new NativeEmbedder() : embedder;
this.defaultTemp = 0.7;
}
#appendContext(contextTexts = []) {
if (!contextTexts || !contextTexts.length) return "";
return (
"\nContext:\n" +
contextTexts
.map((text, i) => {
return `[CONTEXT ${i}]:\n${text}\n[END CONTEXT ${i}]\n\n`;
})
.join("")
);
}
allModelInformation() {
return fireworksAiModels();
}
streamingEnabled() {
return "streamGetChatCompletion" in this;
}
static promptWindowLimit(modelName) {
const availableModels = fireworksAiModels();
return availableModels[modelName]?.maxLength || 4096;
}
// Ensure the user set a value for the token limit
// and if undefined - assume 4096 window.
promptWindowLimit() {
const availableModels = this.allModelInformation();
return availableModels[this.model]?.maxLength || 4096;
}
async isValidChatCompletionModel(model = "") {
const availableModels = this.allModelInformation();
return availableModels.hasOwnProperty(model);
}
constructPrompt({
systemPrompt = "",
contextTexts = [],
chatHistory = [],
userPrompt = "",
}) {
const prompt = {
role: "system",
content: `${systemPrompt}${this.#appendContext(contextTexts)}`,
};
return [prompt, ...chatHistory, { role: "user", content: userPrompt }];
}
async getChatCompletion(messages = null, { temperature = 0.7 }) {
if (!(await this.isValidChatCompletionModel(this.model)))
throw new Error(
`FireworksAI chat: ${this.model} is not valid for chat completion!`
);
const result = await LLMPerformanceMonitor.measureAsyncFunction(
this.openai.chat.completions.create({
model: this.model,
messages,
temperature,
})
);
if (
!result.output.hasOwnProperty("choices") ||
result.output.choices.length === 0
)
return null;
return {
textResponse: result.output.choices[0].message.content,
metrics: {
prompt_tokens: result.output.usage.prompt_tokens || 0,
completion_tokens: result.output.usage.completion_tokens || 0,
total_tokens: result.output.usage.total_tokens || 0,
outputTps: result.output.usage.completion_tokens / result.duration,
duration: result.duration,
},
};
}
async streamGetChatCompletion(messages = null, { temperature = 0.7 }) {
if (!(await this.isValidChatCompletionModel(this.model)))
throw new Error(
`FireworksAI chat: ${this.model} is not valid for chat completion!`
);
const measuredStreamRequest = await LLMPerformanceMonitor.measureStream(
this.openai.chat.completions.create({
model: this.model,
stream: true,
messages,
temperature,
}),
messages,
false
);
return measuredStreamRequest;
}
handleStream(response, stream, responseProps) {
return handleDefaultStreamResponseV2(response, stream, responseProps);
}
// Simple wrapper for dynamic embedder & normalize interface for all LLM implementations
async embedTextInput(textInput) {
return await this.embedder.embedTextInput(textInput);
}
async embedChunks(textChunks = []) {
return await this.embedder.embedChunks(textChunks);
}
async compressMessages(promptArgs = {}, rawHistory = []) {
const { messageArrayCompressor } = require("../../helpers/chat");
const messageArray = this.constructPrompt(promptArgs);
return await messageArrayCompressor(this, messageArray, rawHistory);
}
}
module.exports = {
FireworksAiLLM,
fireworksAiModels,
};