-
Notifications
You must be signed in to change notification settings - Fork 1.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
New wrapper for Zephyr models + little fix in memory.py #183
Conversation
VectorStoreIndex is imported but non-existent VectorIndex is used.
With inner thoughts.
Added a wrapper for Zephyr models (and any model using |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Awesome PR @v-kamelowy , I'm going to go ahead and merge this in for now in case people want to use is ASAP, and might touch it up in a separate PR later (if needed) after I get a chance to test it.
@@ -657,7 +657,7 @@ def __init__(self, archival_memory_database: Optional[str] = None, top_k: Option | |||
storage_context = StorageContext.from_defaults(persist_dir=directory) | |||
self.index = load_index_from_storage(storage_context) | |||
else: | |||
self.index = VectorIndex() | |||
self.index = VectorStoreIndex() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@sarahwooders FYI
"Failed to decode JSON from LLM output:" |
Thanks for the report @vi1world, if possible can you post some more details?
|
It can happen if you use quantization lower than Q6_K. |
I'm using zephyr-7b-beta.Q8_0 on LM Studio - 0.2.7
It works quite well when I use --model airoboros-l2-70b-2.1 in the parameter |
Interesting, I've been testing it for a good 30 minutes and it has been working fine, no errors. I think that It can be a problem with the Beta version of Zephyr, I've tested it on Alpha. Maybe I've made a typo somewhere, will see later, thanks for report. |
Just tested with Zephyr alpha. Doesn't work with zephyr-7B-alpha in the parameter. |
I am going to paste everything I've got and see if anyone can recommend a solution. Created a new clone from the repo early this afternoon, about 3 hours ago. This is my MemGPT screens (MemGPT) C:\Users\aaron\App_Development\MemGPT>set OPENAI_API_BASE=http://localhost:1234 (MemGPT) C:\Users\aaron\App_Development\MemGPT>set BACKEND_TYPE=lmstudio (MemGPT) C:\Users\aaron\App_Development\MemGPT>python main.py --no_verify During handling of the above exception, another exception occurred: ╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮ During handling of the above exception, another exception occurred: ╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮ (MemGPT) C:\Users\aaron\App_Development\MemGPT> |
@Rivelyn you can use http://gist.github.com/ and then drop a link here... |
@Bec-k There was an update to LM Studio yesterday, plus @vivi gave me some some instructions on using the model parameter when loading MemGPT. I am not getting the Failed to decode JSON error now, there seems to be another issues with message retries. So I am working my way through that first, I need to deal with one problem at a time and stop posting until I hit a wall that I can't pass...lol So for now I am not saying the Failed to decode JSON is corrected, I am just not getting that anymore, well for the moment. |
@Bec-k So I thought I had things working, best progress so far with LM Studio and Zephyr. I was able to get MemGPT to finally load in the CLI and start with the 'Welcome' message. I was able to get a few messages back and forth with MemGPT and updated it's memory for my name and some general information. Then it broke with the Exception: Failed to decode JSON from LLM output message. What exactly would you like me past into either here, or gist? I have not modified any of the MemGPT code and everything I am doing is local on LM Studio, with a fresh install of it about an hour or so ago. I reported a bug to LM Studio from their update yesterday. I was told to re-install today and that corrected the LM Studio bug. |
@Rivelyn I just tested out zephyr beta 7B q8_0 and q6_k. Make takeaway right now is dolphin 2.1 (q6_k) with the airoboros wrapper is much more stable than Zephyr. For example even with zephyr beta q8 I'm getting consistent mistakes in the functions calls + weird prefixes. E.g. instead of
|
@cpacker Thank you, yea right now Dolphin seems to be working well. Zephyr has potential but I don't think it's there yet. Matthew Berman on YouTube ran it through some testing, some results were passes and others fails, but he pushes creative thought on the raw LLM's and I don't think there are any LLM's with fantastic creative thought yet... soon.. but not yet. |
* VectorIndex -> VectorStoreIndex VectorStoreIndex is imported but non-existent VectorIndex is used. * New wrapper for Zephyr family of models. With inner thoughts. * Update chat_completion_proxy.py for Zephyr Wrapper
VectorStoreIndex is imported, but non-existent VectorIndex is used.