chris-aeviator

I'm getting a lot of these errors - some pages work just fine, all the warc files I'm reading have HTML, the error itself is strange enough since `200 ok`...

hx reports a language server healthy even if it is not

6

I'm repeatingly running into issues with HX being able to load my language server on different machines. It's a hit or miss sometimes and I'm missing a way to figure...

Unable to start Consul Docker container - node-id permission denied

3

``` badger_1 | ==> failed to setup node ID: open /consul/data/node-id: permission denied ``` ``` version: '3' services: badger: image: consul:1.8.5 volumes: - ./consul:/consul command: agent -server --data-dir=/consul/data -ui -node=server-1...

When using "model_id = "bigscience/bloomz-1b1" via a huggingface pipeline, I'm getting warnings about the input_ids not being on the GPU and > "topk_cpu" not implemented for 'Half' when doing do_sample=True...

validate huge files last

2

When interacting with h2ogpt I regularly run into a situation where the base model is loaded just to realize a validation fails, e.g. local_files_only but missing a tokenizer, wrong folder...

quantizing a 4096 context length model leads to corrupted output

1

when converting and quantizing CarperAI/pythia-2.8b-deduped-4k (I've added it to Pythia Dict inside config.py with the only change being block_size=4096 ) I'm getting nonsensical output. I'd be happy to contribute a...

for anybody running into this when loading a gptq.int4 model, it can be fixed by running ` `pip install -U --index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/Triton-Nightly/pypi/simple/ triton-nightly `` as per https://github.com/openai/triton/pull/1741

implement unlimiformer into transformers

3

### Feature request https://github.com/abertsch72/unlimiformer promises to support unlimited input length on any transformer based encoder/decoder model with sub-linear cost in time. ### Motivation Context lengths are fairly limited ### Your...

Support full display of long lines

3

Nowadays machine learning (training) datasets often come as JSONL documents. When working with natural language, a typical user utterance or dialog context often comes with all information in one sentence,...

Exponential reconnect backoff

2

When fetch-event-source loses it's connection to the server, it tries to reconnect rather aggressively. Given enough clients, this continious repeating traffic pattern may pose an unneccesary risk of overloading a...