I downloaded Chat with RTX and got it to install successfully.
On first use, I could point to a YouTube video, but not to a directory on my computer.
The next day, I couldn’t use Chat with RTX at all. The DOS window that opens gave me an error message. And the browser window with the user interface never opened. I don’t know how to access/start the app right now. I’m using a 3060Ti. Can you assist please?
Thanks - Gary
I have the same issue here. It installed correctly to the default user directory.
It runs after installation.
The question is, after a computer shut down, or log-out; how do you access (restart) the program?
The server on http://127.0.0.1:17938 is NOT there after rebooting.
The only way I’ve been able get it back again is to re-install and check fresh installation.
Regards, ~Ken
Thanks Keng. I appreciate the info. And yes - that’s what I’m seeing. Can’t access the program and going to the prior url doesn’t help. I tried re-launching this morning and got a message that said “failed to load file” and it’s pointing to an older temporary file. Has been “parsing nodes” all day. It’s been going for hours and neither works nor gives an error. Taking up about 20% of my CPU but nothing on the GPU.
12 cores, 20 logical CPUs, (i7-12700K) 16GB ram used, 10-12% CPU used next to 0% on GPU.
Its looking at 2.7GB in 2,200 files as .pdf. I thought the app. was to use the GPU & its warp speed memory.
I’ve no clue yet as to if or when it finishes.
Almost the same here - i7 - 11700 w/32GB ram. No idea why it’s taking so long nor what the reward is at the end if the program won’t load/work.
I keep seeing the only solution (assuming you used the default director and have no spaces in your name) is to uninstall and start again. Could be an interesting program, but I’m not doing that every day.
I had the same. Initially, upon starting to load files, my GPU is hardly being used. Later on in the process, when embeddings are being generated (you will see the white bars growing in the CMD window), my GPY is being used up to (almost) 100%. This latter can take some time, especially with a large dataset of 2.7GB.
1 Like
Any idea how many of those white embeddings bars need to be filled?
Last night, I had 6-7 of those bars filled, and the GPU was starting to overheat.
Overall, the parsing nodes and embedding had taken around 10-11 hours, and I had no idea how much longer it was going to keep going.
In the end, I just uninstalled - has been too much time/effort with no valuable result.
1 Like
My initial test failed.
RuntimeError: Unable to generate the inference engine: maximum recursion depth exceeded while calling a Python object
Apparently, my trial was a big ask for the demo.
As noted above, I uninstalled the program. I think it could be good, and I like the privacy benefits of handling the data on my own computer. Nvdia described this as a demo. I think that’s right - it shows what can be done, but isn’t really ready for full roll-out yet. The number of identical problems just trying to get the program to work demonstrate that. Think I’m going to wait until there’s another more stable version and try again in the future.
1 Like
That being said, the ? of restarting the demo still deserves an answer.
I have the same issue as keng1. OK, the local server is not expected to restart on the same port as before. But various AI-based support centers say to look for “ChatWithRTX[dot]exe.” I cannot find that on my system or even in the registry. There did not appear to be any re-entry/re-start documentation.
My third installation seemed to work somewhat better. This time, I gathered about 36,000 files which were .pdf, .doc?, .txt, .csv, and placed them in the folder, “testdataset,” on an external SSD drive. Chat With RTX’s Folder Path was set to “E:\testdataset” Mistral 7B int4 was the AI model selected.
The dataset was refreshed, and it ran for a couple of hours.
Some errors were reported in the CMD window. Examples of the kinds of errors were:
Failed to load file E:\testdataset\Financial Statement 1xy0_001 sec March 20zz.pdf with error: PyCryptodome is required for AES algorithm. Skipping…
[02/27/2024-14:26:42] Multiple definitions in dictionary at byte 0x7d for key /Subtype
[02/27/2024-14:29:18] /Kids or /Limits found in PageLabels. Please share this PDF with pypdf: ENH: Add support for page labels by MartinThoma · Pull Request #1519 · py-pdf/pypdf · GitHub**
Failed to load file E:\testdataset\Statement (wertyuy) for 07-01-2020–09-30-2020 memberview.pdf with error: local variable ‘cm’ referenced before assignment. Skipping…
[02/27/2024-14:31:17] incorrect startxref pointer(3)
[02/27/2024-14:30:08] EOF marker not found
Nevertheless thousands were apparently ok. Examples:
[02/27/2024-14:28:57] Object 4 0 found
[02/27/2024-14:28:57] Object 2 0 found
[02/27/2024-14:28:57] Object 4 0 found
[02/27/2024-14:28:57] Object 2 0 found
[02/27/2024-14:28:57] Object 4 0 found
[02/27/2024-14:28:57] Object 2 0 found
The issue I raise here is that none of the queries I posed to Chat With RTX were answered in terms of anything except the default dataset downloaded with Chat With RTX. No queries were responded to in terms of the testdataset. It appears that the testdataset had no practical impact.