@tsarnick
weirdly, I think it is existing celebrities. Sure you can make AI influencers, and they can and will compete, but even with interesting personality, the brands of at least some existing people remain valuable. People are interested in people, even with superinteresting AI around.
WizardLM: An Instruction-following LLM Using Evol-Instruct
uuh
"WizardLM-7B outperforms ChatGPT in the high-complexity instructions...
Evol-Instruct is a novel method using LLMs instead of humans to automatically mass-produce open-domain instructions"
GeneZC/MiniMA-3B
A language model distilled from an adapted version of LLaMA2-7B following "Towards the Law of Capacity Gap in Distilling Language Models".
TheBloke/alpaca-lora-65B-GGML
-4bit
-2bit
Already with positive community reception:
"This is the best model I have tried locally this far. Thank you!"
digitous/Alpacino30b
A triple model merge of (Alpaca+(CoT+Storytelling)), resulting in a comprehensive boost in Alpaca's reasoning and story writing capabilities.
LLMs/AlpacaGPT4-LoRA-7B-OpenLLaMA
Soon there will be high performance LLMs with permissive licenses. It won't take long until the fully trained openllama get's a fine tune on OpenAssistant data. Later we will have 30b redpajama on that.
Trelis/Mistral-7B-Instruct-v0.1-Summarize-16k
used patent dataset to make it. They also have a 60k version, which you can buy for 30 euro.
that's the first model I have seen on the hub which you can purchase. I don't think that's a bad idea.
BLING: "Best Little Instruction-following No-GPU-required"
- BLING is designed for enterprise automation use cases, especially in knowledge-intensive industries
- BLING is not designed for 'chat-bot' or 'consumer-oriented' applications
News:
"Preparing the 33B version and we expect to empower WizardLM with the ability to perform instruction evolution itself, aiming to evolve your specific data at a low cost.
- released 13B version of WizardLM trained with 250k evolved instructions.
TheBloke/MPT-30B-Dolphin-v2-GGML
This and the non-quantized weights might now be one of the best LLMs on HF. Replicating orca, I guess without censorship.
Altman says they brought cost down of operating gpt-3 (I guess that's gpt3.5 by now) by a factor of 40.
Tell that anybody who thinks chatgpt has 175 billion parameters.
amazon/FalconLite2
"By utilizing 4-bit GPTQ quantization and adapted RotaryEmbedding, FalconLite2 is able to process 10x longer contexts while consuming 4x less GPU memory than the original model."
LumiOpen/Poro-34B
"Poro is a 34B parameter decoder-only transformer pretrained on Finnish, English and code. It is being trained on 1 trillion tokens (300 billion as of this release)."
More and more are coming, even for languages of small countries. 🇩🇪?
dranger003/LWM-Text-Chat-128K-iMat.GGUF
"The imatrix Q4-K quant fits with 32K context on 24GB and gives me ~100 t/s inference on a 3090.
With IQ3_XXS it seems to fit ~37K context on 24GB (and it is even faster than Q4-K)."
VAGOsolutions/SauerkrautLM-7b-HerO
merge of Teknium's OpenHermes-2.5-Mistral-7B and Open-Orca's Mistral-7B-OpenOrca, fine-tuned on the Sauerkraut dataset
KnutJaegersberg/Tess-M-34B-2bit 🦾
A quip# 2 bit quantization of Tess-M by
@migtissera
based on Yi-34B-200k by
@01AI_Yi
. Weights are 10 GB smol, yet model quality is very good! Made with 8k context hessians to enhance long context inference quality.
DriveML and forester ML libraries have a very streamlined workflow to use tuned ML models for explainable ML, forester has EDA functions. One can expect good performance and insights ratio for coding time invested.
#rstats
#machinelearning
#datascience
There was this interview with
@sama
that suggested they considered open sourcing gpt3, but they thought most businesses could not handle it, as it is so big.
I could download below and set up a system in the cloud with 2 80gb vram gpus, if I wanted to.
@kimmonismus
this wave of AI is indeed disruptive. entire professions almost disappear. translators, transcribers, narrators, and the field is expanding. Like professional drawers that were replaced by PCs, software and printers. It's that kind of change and it's accelerating.
AI: What is the future of artificial intelligence? - BBC News
"I've tried to brief policymakers: it is like explaining particle physics to a chocolade chip cookie"