Ruhollah Majdoddin

Add a "stop generation" button to chat.

1

I am trying the online chat, sometimes it takes too long because of queuing. Sometimes it even seems it has hanged. That button would be very useful then.

editable prompt(L0) in chat

1

I see no practical reason not to make the first user message editable. One can superficially give a `Hi` prompt.

Install (with Docker) fails on Cloud GPUs

2

Instances of some Cloud GPUs, like colab and cloud.vast.ai, are indeed Docker images, and it is not possible to run a docker image there (see [here](https://github.com/googlecolab/colabtools/issues/299)) Practically Open-Assistant can't run...

adding IPEX, and autocast for CPU

1

duplicate of https://github.com/karpathy/nanoGPT/pull/27#issuecomment-1482609289, bucause it won't be noticed in the close RP . @karpathy @lantiga why still no autocast for CPU? (see also https://github.com/karpathy/nanoGPT/pull/27) For example, better to change https://github.com/karpathy/nanoGPT/blob/a82b33b525ca9855d705656387698e13eb8e8d4b/sample.py#L32...

Bug - model trained on Xs from two sample texts

5

Each sample of openwebtext consists of several paragraphs extracted from a single webpage. nanoGPT is trained to predict a token of a sample, given its previous tokens. For train split...

added frequency_penalty to those toml templates that missed it. Other…

6

## **User description** …wise ChatGPT makes infinte stream (?). For example in 'generate ai tests', for problem 80 - optimal insertion ___ ## **Type** Enhancement ___ ## **Description** - This...

implementin an Engine to serve the trained model by inferencing

2

see #346

Adds support for Romanian by adding Romanian phoneme-based ASR model https://huggingface.co/gigant/romanian-wav2vec2 Based on its HF Model Card, this model has better evaluations than https://huggingface.co/anton-l/wav2vec2-large-xlsr-53-romanian

On virtualized instances from Vast.ai (without CLI): On a A100 40GB it consumes all the vRAM for a 10s audio, run time ~3s, On a A100 80GB it consumes some...

is_flash_attn_2_available() returns False

1

I installed the repo without CLI on virtualized instances from Vast.ai with A100 40GB and 80 GB. is_flash_attn_2_available() is False. Does it mean flash-attn is not used by inference. does...