pseudotensor
Follow
Popular repositories
5,094 contributions in the last year
Activity overview
Contributed to
h2oai/h2ogpt,
h2oai/driverlessai-recipes,
gradio-app/gradio
and 16 other
repositories
Contribution activity
May 2023
Created 471 commits in 1 repository
Created a pull request in h2oai/h2ogpt that received 18 comments
Langchain integration
basic isolated demo with OpenAI or HF/sentence_transformers use better embedding and LLM (ours) to see if comparable to OpenAI incorporate into…
+2,327
−449
•
18
comments
Opened 38 other pull requests in 1 repository
h2oai/h2ogpt
34
merged
2
open
2
closed
- Quantized model updates, switch to recommending TheBloke
- UI help in FAQ
- Improve use of ctx vs. max_new_tokens for non-HF models, and if no docs, don't insert == since no docs, just confuses model
- Fix and test llamacpp
- Ensure can use offline
- Generalize h2oai_pipeline so works for any instruct model we have prompt_type for, so run_db_qa will stream and stop just like non-db code path
- Separate out FAISS from requirements
- Trying to upgrade gradio, but even when to remove or replace replaces…
- Add CLI mode with tests
- Refactor out run_eval out of generate.py
- Constraint output using template
- Refactor run_qa_db a bit, so can do other tasks
- Improve datasource layout
- Allow selection of subset of docs in collection for query
- Move loaders out of finetune, which is only for training, while loader used for generation too
- Mac/Windows install and GPT4All as base model for pure CPU mode support
- Control chats, save, export, import and otherwise manage
- Improve caption, include blip2 as option
- ArXiv support via URL in chatbot UI
- Add image support and show sources after upload
- Add import control check to avoid leaking optional langchain stuff into generate/gradio. Add test
- odt, pptx, epub, UI text paste, eml support (both text/html and text/plain) and refactor so glob simpler
- HTML, DOCX, and better markdown support
- Add URL support
- Control visibility of buttons, but still gradio issues mean can't spin/block button while processing in background
- Some pull requests not shown.
Reviewed 7 pull requests in 1 repository
Created an issue in gradio-app/gradio that received 15 comments
chatbot in gradio server operates through json files, so gradio client cannot actually use remotely
Describe the bug Related: h2oai/h2ogpt#115 (comment) This was my quick test to get chatbot through client working, but because it passes back files…
•
15
comments
Opened 26 other issues in 3 repositories
h2oai/h2ogpt
6
closed
12
open
- AttributeError: 'DualStreamProcessor' object has no attribute 'flush'
- Handle long contexts better
- long generations, eventually fails RuntimeError: expected scalar type Half but found Float
- AttributeError: 'NoneType' object has no attribute 'replace'
- Cannot upgrade transformers due to deepcopy
- argilla 1.7.0 requires numpy<1.24.0, but you have numpy 1.24.2 which is incompatible.
- tasklist mac/windows
- serve html by mounting gradio
- add support for qdrant
- allow soft stopping condition on human-related chars, like line breaks, up to harder condition. So output cleaner
- Explain OS/GPU constraints
- Failed to build sentencepiece
- LangChain task list
- LLaMa spits out <unk> tokens too often, even at start, from generate. And causes repeats and other issues
- chatbot doesn't show queue
- recover old streaming code for num_beams>1 not supported by transformers streamer
- consider improved memory handling for 8-bit
- bitsandbytes: RuntimeError: mat1 and mat2 shapes cannot be multiplied (1x2 and 1x5120)
gradio-app/gradio
4
closed
3
open
- HTML no longer renders in chat output
- Bad handling of \n in chatbot
- Button values wrong, borrowed from others [after others clicked]
- ValueError ... is not in list
- button clickable during action click already did, no way to block, user ends up clicking madly and breaking things.
- Extremely slow generation in chatbot in some cases
- large number of problems with tex2svg taking down entire server
huggingface/transformers
1
closed
Answered 1 discussion in 1 repository
imartinez/privateGPT
imartinez/privateGPT
-
Any way in which we can increase the source document count?
This contribution was made on May 24
7
contributions
in private repositories
May 2 – May 26