iAkashPaul

Is this applicable to public & private buckets alike? Like if I keep making requests with the wrong token?

My go to for lightweight deployments is HuggingFace's Spaces, free 2-core CPU, anything else can go to either serverless container images on runpod or baseten.

With unsloth's optimizations you can do llama-3-8b's QLoRA fine-tuning on your 8GB card(mine's a 2070S) with 900MB to spare with BS of 4.

Slip in a line asking them to reverse the string - .DefaultCellStyle

I actually want Play Store apps to be equally configurable. Apps like Truecaller refuse to work without so many permissions. Being Android you'd expect some way to provide a dummy contact book, location & other accesses that apps need.

Heck I'd want an option to freeze apps once I'm logged in & only unfreeze them when I need to use them. Would absolutely improve battery life.

Seems like using RAFT(https://techcommunity.microsoft.com/t5/ai-ai-platform-blog/r...) to generate the glossary+KGs for a given doc should give better results for cross-doc or intra doc answers instead of just relying on vDB chunks alone.

Yep, termux is a good way to do this. Llama.cpp has Android example as well, I forked it here GitHub.com/iakashpaul/portal you can try it with any supported GGUF/Q4+Q8 models

You can either modify the Android example inside llama.cpp or my fork of it at GitHub.com/iakashpaul/portal

Increase the Ctx to more than 100 & link to any Q4 GGUF of 7B

You can plug sileroVAD in the browser for this sort of interruptions, if you can make use of threads/workers in JS then you can mute/stop your output & instead have the chunks dumped to the STT websocket

It was also possible to install an OS without needing to create an account.

Expecting this to be continuously possible on a proprietary OS is a lost cause. Even the LLM world has practically mirrored this.

TGI+grammar loaded with Mistral/Mixtral works great for structured output now! No more langchain exception handling for unmatched Pydantic definitions.

It's basically guidelines to adhere in the output. Ex. Your prompt asks for a summary, the LLM won't necessarily just spit the summary, there might be phrases like here's the summary - then the actual summary or even basic JSON key/values can get messed up. So grammar allows for defining the expected output in a variety of ways, some via Pydantic class definitions, Microsoft's Guidance, Outlines along with llama.cpp's grammar are attempts at making structured output reliable.

Most of langchain is basically a specific prompt with exception handling for missing fields.

You should do the new HF TGI server, it has both grammar & tool support now. Works fabulously with Mistral Instruct & Mixtral Instruct.

Damn

CC-by-NC still seems better than the DBRX's license prohibiting usage for improving other models-

>2.3 Use Restrictions > You will not use DBRX or DBRX Derivatives or any Output to improve any other large language model (excluding DBRX or DBRX Derivatives).

I use Zoho Mail with custom domains

His robots are just automated scripts which do the bulk repetitive tasks unlike the agents that comment OP thought.

I'm using ReACT for RAG based chatbot & tools which are just different retrievers.

Hey Jake, I meant capacity per shard in this case, not exhausting the IDs. Any potential solutions for that or is that not an immediate challenge?

Even notion has a similar approach to sharding postgres but both of them could benefit from having shard IDs prefixed with YY/MM/DD(as needed) otherwise it's back to the shard navigator once they max out against org-ids for each shard's capacity

Absolutely

Reduced memory requirements, dropping higher precision IP blocks for starters

FPGAs for native FP4 will change the entire landscape

Seems great, coming from having to install code-server(hosted VSCode on remote machines) along side SSH via local VSCode. A better managed experience for both would be pretty neat!

The last bit is already under way with speculative decoding but wonder what exactly are they proposing

Despite reading it twice I couldn't come away with why they chose char/s or Hz as an appropriate measure. They also provided no benchmarks or model sizes except for a relative comparison with models 10x or 100x in size, which leads me to assume this is a small model maybe?

Cheaper phones with a better Ubuntu/Android ROMs could actually make for better SBCs when considering all sensors that come with them.

Meta wouldn't even officially sell this let alone have repair shops in India, unlike Apple's strategy of eventually rolling out worldwide

Base Mistral 7B is hardly suitable for the evaluations, even one team at Intel tried to pull a fast one with NeuralChat in the exact same way https://huggingface.co/Intel/neural-chat-7b-v3#quantitative-...