[Feature] Support for Intel GPUs (HD Graphics, Iris Xe, Arc)

[zwilch]

System Info

32GB RAM Intel HD 520, Win10 Intel Graphics Version 31.0.101.2111

Information

• [X] The official example notebooks/scripts
• [ ] My own modified scripts

Reproduction

Select GPU Intel HD Graphics 520

Expected behavior

All answhere are unreadable like

Ich The „ I … ... I „ In S _ E

\ ** #smith " # Englishimster glob Qsmith

[tazz4843]

I can reproduce this on an Intel Arc A770 dGPU, without fail. Arch Linux, latest updates.

[goldyfruit]

Same here Intel ARC A770:

Python 3.11.7 (main, Dec 18 2023, 00:00:00) [GCC 13.2.1 20231205 (Red Hat 13.2.1-6)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from gpt4all import GPT4All
>>> model = GPT4All("mistral-7b-openorca.Q4_0.gguf", device='intel')
llama_new_context_with_model: max tensor size =   102.55 MB
llama.cpp: using Vulkan on Intel(R) Arc(tm) A770 Graphics (DG2)
>>> 
>>> output = model.generate("The capital of Canada is ", max_tokens=10)
>>> print(output)
5 city K   
 S ,  “

[koech-v]

OneAPI might solve this issue, i have also seen this thread.

[zwilch]

Version 2.6.1 Bug is still there

[zwilch]

Version 2.6.1

Garbage output on Intel GPUs (HD 520 Graphics )

[zwilch]

Gargage output on Intel GPUs (HD 520 Graphics) gpt4all V.2.7.0

[TheWonderfulTartiflette]

I have another problem with Arc with GPT4All, that is that it just doesn't recognize my GPU at all, leaving me the option just for CPU

[cebtenzzre]

I have another problem with Arc with GPT4All, that is that it just doesn't recognize my GPU at all, leaving me the option just for CPU

This is because we recently started hiding these GPUs in the UI, such that GPT4All doesn't use them by default given that they are known not to be compatible.

[simplybel9000]

I have another problem with Arc with GPT4All, that is that it just doesn't recognize my GPU at all, leaving me the option just for CPU

This is because we recently started hiding these GPUs in the UI, such that GPT4All doesn't use them by default given that they are known not to be compatible.

is it possible to use them and if so how can we use them? its the only gpu that i have and cant test others

[enzy]

Intel Arc A770 with the latest llama.cpp (and SYCL enabled) works for me (on Linux). Can we enable these discrete graphics?

[x0rgo]

I have another problem with Arc with GPT4All, that is that it just doesn't recognize my GPU at all, leaving me the option just for CPU

This is because we recently started hiding these GPUs in the UI, such that GPT4All doesn't use them by default given that they are known not to be compatible.

Not to be compatible as in Garbage output or simply not recognized by the library??