What is the browser requirement and hardware/driver requirement to run agent LLM?
I found this error when running the latest commit, the sandbox you hosted also spit out the same error, I guess probably my mistake somewhere.

❌ Error retrieving initial tasks array. Refresh and retry, clarify your goal, or revise your goal such that it is allowed by our model's policies. Error message: Cannot find adapter that matches the request
If you are facing any issues, please check out the Help page or Github

Ubuntu 22.04
Edge Dev version 115.0.1843.0 dev (64 bit)
Edge://GPU WebGPU: Hardware accelerated
GPI 3090 driver 530

I installed fork to Win environment, and there were no erors, and the Process starts ok - althoug having also error [(https://github.com//issues/7)]
When I try to use the agent offline, there is error, since program tries to connect to hugginface:
Entering new llm_chain chain...
console.js:13 Starting (...) with name webLlm with prompts: You have the following objective: plan a trip to Havaji. Create a list of zero to three tasks to be completed by you such that your goal is more closely reached or completely reached. The response list MUST be a SINGLE ARRAY OF STRINGS where each string is a TASK NAME wrapped in quotes (EXAMPLE OF RESPONSE FORMAT: "["task name", "task name"]"). You MUST be able to parse the list with Javascript’s JSON.parse() function. You must respond with the array WITHOUT ANY OTHER TEXT.

tvmjs_runtime.wasi.js:2660 GET https://huggingface.co/mlc-ai/web-lm/resolve/main/vicuna-7b-v1/ndarray-cache.json net::ERR_INTERNET_DISCONNECTED

The json can be fetced easily if the connection is open. Does the json need to be fetced, could it be local? Or is my setup just broken?

Prompt:

export const analyzeTaskPrompt = new PromptTemplate({
  template: `You have the following higher level objective "{goal}". You currently are focusing on the following task: "{task}". Based on this information, evaluate what the best action to take is strictly from the list of actions: {actions}. You should use 'search' for searching the web about current events where "arg" is a simple clear search query based on the task only. Use "reason" for all other actions. Return the response as an object of the form {{ "action": "string", "arg": "string" }} that can be used in JSON.parse() and NOTHING ELSE.`,
  inputVariables: ["goal", "actions", "task"],
});

Actual Response:

'Understood. Based on the current task, the best action to take is to search the web about current events related to "Alexander Graham Bell". Therefore, the action to take is "search" and the argument is "arg". The response will be of the form { "action": "search", "arg": "arg" }.'

Expected response

'{ "action": "search", "arg": "Alexander Graham Bell" }'

It seems Wizard LM 7B is not able to follow the exact prompted instructions. Is anyone else have the same issue?

The Edge Function "pages/api/agent/execute" must export a default function

Inspired by an earlier issue, the pace of development in this space is overwhelming, and I'm sure this project will gain traction. It wouldn't be feasible for you to manually add support for everything users are requesting then.

Hence, would it be possible to support users choosing their own models, probably locally downloaded as GGML or GPTQ as is popular with Ooga Booga? This would also allow users flexibilty (i.e. scaling down to 1.3B models or up to 30+B models) according to their devices.

First of all, thanks for exellent update!
In previous version, and also after updating to commit 927c4ae, execution for me always ended with these errors (see pic). I couldn't figure the reason, but then, after increasing the AutonomousAgent.ts timeout values substantially:
const TIMEOUT_LONG = 100000;
const TIMOUT_SHORT = 80000;
I was able to get a succesful shutdown after all tasks completed.

My setup is not optimal for this kind of things:
arch: AMD64
cpu: Intel64 Family 6 Model 158 Stepping 9,
GPU device: Quadro P4000 (1) (compute_37) (6, 1)
cuda: 11.3
cudnn: 8302

So, maybe having a option to set the timeout in settings for non-coding users might be nice feature, or just have this thing mentioned in faq might be useful for others.
I'll run some more test, and if problem still persiss, I'll let you know.

Hi there,

I'm new to AgentLLM and I'm having some trouble getting it to work. I followed the instructions and ran the agent for the first time. It downloaded everything it needed and initialized correctly. The agent also added two tasks, but then it got stuck and didn't do anything else.

I tried canceling the agent and running it again, but now I'm constantly seeing the error message in the attached screenshot. I'm not sure what's causing this error or how to fix it.

A bug report (I think something has changed on Hugging Face):

Running from https://gentllm.vercel.app

Any attempt at running creating a new goal results in the message:

Error retrieving initial tasks array. Refresh and retry, clarify your goal, or revise your goal such that it is allowed by our model's policies. > Error message: Failed to execute 'add' on 'Cache': Request failed
If you are facing any issues, please check out the Help page or Github

In the Network tools, you can see the error is being cause by https://huggingface.co/spaces/idosal/web-llm/resolve/main/wizardlm-7b/ndarray-cache.json returning a 401 "Unauthorized" status code. If you go directly to that URL in your browser, body of the page is just: "Invalid username or password.".

In my testing, WizardLM-7b model is superior to vicuna-7b in terms of following complex instructions, especially langchain style prompting. Would it be possible to add a feature where one could select the model to be loaded?

After initialization 100%, I see the GPU loaded with 5-6Gb, working some seconds and always receive the error:

❌ Error retrieving initial tasks array. Refresh and retry, clarify your goal, or revise your goal such that it is allowed by our model's policies. Error message: The operation failed for an operation-specific reason
If you are facing any issues, please check out the Help page or Github

With previous commit (tested yesterday) from time to time I received the task list, and later same error, but with today commit always error at first.

I receive the error with local docker and with online https://agentllm.vercel.app/

rt

Where is the model in local?
I want to replace by 13B

Look at it, I am still at 0% and don't know why

I am not why, while running AgentLLM, all console logs that I inserted are not shown in the terminal? I can hardly debug this project where the web search is never used (I have put env serp api key)