doing something
@doing_something
712 subscribers
911 photos
36 videos
36 files
1.94K links
@smlkw doing something, just my notes to keep updates
Download Telegram
About
Blog
Apps
Platform
Join
doing something
712 subscribers
doing something
https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models

#llm
Liquid AI
Introducing LFM2: The Fastest On-Device Foundation Models on the Market | Blog
Today, we release LFM2, a new class of Liquid Foundation Models (LFMs) that sets a new standard in quality, speed, and memory efficiency for on-device deployment. Built on a hybrid architecture, LFM2 delivers 200% faster decode and prefill performance than…
391 views
doing something
https://github.com/beowolx/rensa

#rust #nlp
GitHub
GitHub - beowolx/rensa: High-performance MinHash implementation in Rust with Python bindings for efficient similarity estimation…
High-performance MinHash implementation in Rust with Python bindings for efficient similarity estimation and deduplication of large datasets - beowolx/rensa
410 views
doing something
https://github.com/mlabonne/llm-course

#llm
GitHub
GitHub - mlabonne/llm-course: Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks.
Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks. - mlabonne/llm-course
373 views
doing something
doing something
https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models #llm
https://colab.research.google.com/drive/1iCVMUmguLmZ_rfltcp83mStjJUJdIbWi?usp=sharing

#llm
381 views
doing something
A post to bookmark projects about scientific discovery

#science
134 views
doing something
Love to use polars internal parallelizarion for intensive tasks

#rust #python
393 viewsedited  
doing something
https://bentoml.com/llm/

#llm
Modular
LLM Inference Handbook
A practical handbook for engineers building, optimizing, scaling and operating LLM inference systems in production.
124 views
doing something
https://developers.googleblog.com/en/genai-processors/

#genai
Googleblog
Google for Developers Blog - News about Web, Mobile, AI and Cloud
Learn about GenAI Processors, a new open-source Python library for building powerful and flexible Gemini applications with streamlined, real-time multimodal processing.
119 views
doing something
https://github.com/EricLBuehler/mistral.rs/blob/master/docs/GEMMA3N.md

Inference engine written in Rust

#rust #gemma
GitHub
mistral.rs/docs/GEMMA3N.md at master · EricLBuehler/mistral.rs
Blazingly fast LLM inference. Contribute to EricLBuehler/mistral.rs development by creating an account on GitHub.
95 views
doing something
https://github.com/google/yamlfmt

#go
GitHub
GitHub - google/yamlfmt: An extensible command line tool or library to format yaml files.
An extensible command line tool or library to format yaml files. - google/yamlfmt
95 views
doing something
https://github.com/cocoindex-io/cocoindex

#rust
GitHub
GitHub - cocoindex-io/cocoindex: Incremental engine for long horizon agents 🌟 Star if you like it!
Incremental engine for long horizon agents 🌟 Star if you like it! - cocoindex-io/cocoindex
97 views
doing something
https://docs.unsloth.ai/basics/unsloth-dynamic-2.0-ggufs

#optimization
unsloth.ai
Unsloth Dynamic 2.0 GGUFs | Unsloth Documentation
A big new upgrade to our Dynamic Quants!
99 views
doing something
doing something
https://github.com/beowolx/rensa #rust #nlp
speed example
95 views
doing something
Discovered https://pypi.org/project/polars-lts-cpu/

#rust
PyPI
polars-lts-cpu
Blazingly fast DataFrame library
99 views
doing something
doing something
speed example
On larger machines speed is better (it uses 2 cores), ofc

But for big datasets Rensa requires a lot of memory

16 million of text samples requires ~100 gb of memory
105 viewsedited  
doing something
https://pypi.org/project/parquet-tools/

#parquet #data_engineering
106 views
doing something
doing something
On larger machines speed is better (it uses 2 cores), ofc But for big datasets Rensa requires a lot of memory 16 million of text samples requires ~100 gb of memory
Do it on your own:

https://gist.github.com/egorsmkv/ea16104f4e167702f3a7f65a442eed6f
108 viewsedited  
doing something
HF dataset viewer support chat conversations

Example: https://huggingface.co/datasets/HuggingFaceTB/smoltalk

#hf
113 viewsedited  
doing something
My time to use dask, because 16 GB jsonl with non-trivial structure is not fittable into 50 gb of memory...

#data_engineering
120 views
doing something
doing something
My time to use dask, because 16 GB jsonl with non-trivial structure is not fittable into 50 gb of memory... #data_engineering
Ohhh, it does not save original structure... List became "Large String"...

Splitted the original file into smaller lines using split and then processed using polars
128 views
doing something
datasets 4.0.0 rocks now

https://huggingface.co/docs/datasets/en/use_with_polars
136 viewsedited