danielgross / llamaacademy Goto Github PK

It seems individual model training is lost to other users, which would be more beneficial to share and use each run to improve the model for all, over time improving the base model beyond what any single 'student' can do on its own.

Similar to llama_index for external data, can you have an 'awesome llama learnings' collection as easily integrate diffs one can merge for their own needs or to a single central model, making it the sum greater than all the parts?

That would be grand

Trying as suggested in README.md to run this on "1 RTX A6000" with docker image pytorch:latest f5540ef1a1398b8499546edb53dae704 from https://cloud.vast.ai/

[email protected]:~/llamaacademy$ conda env create --file=environment.yaml

Returns out of memory error

CondaError: Failed to write to /opt/conda/pkgs/nccl-2.15.5.1-h0800d71_0.conda
  errno: 28
CondaError: Failed to write to /opt/conda/pkgs/pytorch-2.0.0-cuda112py310he33e0d6_200.conda
  errno: 28
CondaError: Failed to write to /opt/conda/pkgs/libcusparse-12.0.0.76-hcb278e6_1.conda
  errno: 28
CondaError: Failed to write to /opt/conda/pkgs/cuda-sanitizer-api-12.1.105-0.tar.bz2
  errno: 28
[Errno 28] No space left on device: '/opt/conda/pkgs/libzlib-1.2.13-h166bdaf_4.tar.bz2'

Debugging memory

[email protected]:~/llamaacademy$ df -h
Filesystem      Size  Used Avail Use% Mounted on
overlay          10G  2.4G  7.7G  24% /
tmpfs            64M     0   64M   0% /dev
shm              12G     0   12G   0% /dev/shm
/dev/sdb1       100G  100G   32K 100% /etc/hosts
/dev/sda2        49G   16G   31G  34% /usr/bin/nvidia-smi
tmpfs            25G     0   25G   0% /sys/fs/cgroup
tmpfs            25G   12K   25G   1% /proc/driver/nvidia
tmpfs            25G  4.0K   25G   1% /etc/nvidia/nvidia-application-profiles-rc.d
tmpfs           4.9G  1.9M  4.9G   1% /run/nvidia-persistenced/socket
udev             25G     0   25G   0% /dev/nvidia1
tmpfs            25G     0   25G   0% /proc/asound
tmpfs            25G     0   25G   0% /proc/acpi
tmpfs            25G     0   25G   0% /proc/scsi
tmpfs            25G     0   25G   0% /sys/firmware

I found your project by a youtube channel named WorldofAI
, and it seems awesome!
I have used autogpt, and I wanted to learn it code in svelte framework. With specific prompts, I even place a plan for it to follow:

The plan to make a working login page with functionality and error handlings:

• learn how to run node js code or system code inside python using subprocess
• write a sample run.py with that knowledge of previous step
• learn svelte using the web
• learn how to create a svelte project by using this link: 'https://www.section.io/engineering-education/svelte-with-vite-typescript-and-tailwind-css/'
• create a sample svelte project
• write a py file which runs a sample echo on the terminal 'hello world'.
• extend the written py file to run pnpm install inside of the created sample svelete project
• make sure that node_module folder is created inside the project directory
• make another py file which runs pnpm dev to run the project
• check whether a new process is created and is up on the desired port
• make a simple login page with two inputs username and password and a button with text login
• make a mock server which accepts post requests on /success path
• on that path server accepts a json with username and password, and checks whether both values are equal to 'admin' and responds 200 if it's ok, and 401 if not
• interact with user with sending fake username and password to a mock server
• handling errors with those interactions
• login with username admin and password admin and show appropriate welcoming message
• re route to main page which is /
  but It is not following the path accordingly, making silly descions and not correct itself.
  Does your project is more suitable to train my ai in order to program in javascript?
  Thanks !

Running ubuntu server ,and follow 'README' todo :
/home/miniconda3/envs/llama-academy/lib/python3.10/site-packages/selenium/webdriver/remote/w │
│ ebdriver.py:378 in start_session │
│ │
│ 375 │ │ │ │ capabilities.update({"firefox_profile": browser_profile.encoded}) │
│ 376 │ │ w3c_caps = _make_w3c_caps(capabilities) │
│ 377 │ │ parameters = {"capabilities": w3c_caps} │
│ ❱ 378 │ │ response = self.execute(Command.NEW_SESSION, parameters) │
│ 379 │ │ if "sessionId" not in response: │
│ 380 │ │ │ response = response["value"] │
│ 381 │ │ self.session_id = response["sessionId"] │
│ │
│ /home/miniconda3/envs/llama-academy/lib/python3.10/site-packages/selenium/webdriver/remote/w │
│ ebdriver.py:440 in execute │
│ │
│ 437 │ │ │
│ 438 │ │ response = self.command_executor.execute(driver_command, params) │
│ 439 │ │ if response: │
│ ❱ 440 │ │ │ self.error_handler.check_response(response) │
│ 441 │ │ │ response["value"] = self._unwrap_value(response.get("value", None)) │
│ 442 │ │ │ return response │
│ 443 │ │ # If the server doesn't send a response, assume the command was │
│ │
│ /home/miniconda3/envs/llama-academy/lib/python3.10/site-packages/selenium/webdriver/remote/e │
│ rrorhandler.py:209 in check_response │
│ │
│ 206 │ │ if not value: │
│ 207 │ │ │ value = response["value"] │
│ 208 │ │ if isinstance(value, str): │
│ ❱ 209 │ │ │ raise exception_class(value) │
│ 210 │ │ if message == "" and "message" in value: │
│ 211 │ │ │ message = value["message"] │
│ 212 │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
WebDriverException: Message:

Dear LlamaAcademy developer,

Greetings! I am vansinhu, a community developer and volunteer at InternLM. Your work has been immensely beneficial to me, and I believe it can be effectively utilized in InternLM as well. Welcome to add Discord https://discord.gg/gF9ezcmtM3 . I hope to get in touch with you.

Best regards,
vansinhu

Similar to SLAPA? -https://analyticsindiamag.com/introducing-slapa-self-learning-agent-for-performing-apis/

" SLAPA is trained to learn from itself. If it retrieves wrong information, it makes API calls again until it works. The information keeps getting stored in the model so users do not need to feed it information every time. As more users start using it and feeding in specific APIs, SLAPA will keep accumulating the information, while also becoming more comprehensive with time. "

Hope to know, after instruction generation process, how much instructions is generated to finetune Vincua?

Love this repo, would love to contribute!

Feels like the most needed feature is to have a good evals on couple of test docs (Notion, Github..) to evaluate the incremental models trained.

I also noticed it's non-trival to run the whole main.py flow as all the task are executed serially and if a task break you need to run the whole 1 hour script again. I would love helping to make the docs_ingestion, data_generation etc executable in parallel so the infra can scale easier!

Also open to any other suggestion for PR you guys have

curious what you guys think @huyphan168 @danielgross

Can you add support for LLAMA-Index Loaders?
This is the website: https://llamahub.ai/
This is the Github repo: https://github.com/jerryjliu/llama_index