doing something
Something new in Hugging Face #ai #hf
Hmm, okay it's their DuckDB WASM
> -- The SQL console is powered by DuckDB WASM and runs entirely in the browser.
Hence, it's slow
> -- The SQL console is powered by DuckDB WASM and runs entirely in the browser.
Hence, it's slow
đź‘€3
https://github.com/mit-han-lab/deepcompressor
https://github.com/mit-han-lab/nunchaku
#ai #optimization
https://github.com/mit-han-lab/nunchaku
#ai #optimization
GitHub
GitHub - mit-han-lab/deepcompressor: Model Compression Toolbox for Large Language Models and Diffusion Models
Model Compression Toolbox for Large Language Models and Diffusion Models - mit-han-lab/deepcompressor
A small investigation on WER difference beween two w2v2 models:
https://github.com/egorsmkv/speech-recognition-uk/issues/49
... and why it's important to evaluate models with own dataset.
#ai #speech
https://github.com/egorsmkv/speech-recognition-uk/issues/49
... and why it's important to evaluate models with own dataset.
#ai #speech
GitHub
`Yehor/w2v-bert-uk` vs. `Yehor/w2v-bert-uk-v2.1` · Issue #49 · egorsmkv/speech-recognition-uk
Benchmark table shows that Yehor/w2v-bert-uk is better than Yehor/w2v-bert-uk-v2.1
Published Open Source Crimean Tatar Text-to-Speech dataset to Hugging Face:
https://huggingface.co/datasets/Yehor/qirimtatar-tts
#ai #speech
https://huggingface.co/datasets/Yehor/qirimtatar-tts
#ai #speech
❤2
doing something
https://neuralmagic.com/blog/multimodal-model-quantization-support-through-llm-compressor/ #ai #optimization
At least, it works somehow
Quantized it using https://pypi.org/project/llmcompressor/
Data used for calibration: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean-punctuated
Quantized model: https://huggingface.co/Yehor/whisper-large-v2-quantized-uk
#ai #speech
Quantized it using https://pypi.org/project/llmcompressor/
Data used for calibration: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean-punctuated
Quantized model: https://huggingface.co/Yehor/whisper-large-v2-quantized-uk
#ai #speech
doing something
At least, it works somehow Quantized it using https://pypi.org/project/llmcompressor/ Data used for calibration: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean-punctuated Quantized model: https://huggingface.co/Yehor/whisper-large-v2-quantized…
But I couldn't run it on L4, it needs about 25 GB of RAM to init
https://javascript-conference.com/progressive-web-apps/optimizing-prime-video-with-webassembly-and-rust/
#rust
#rust
International JavaScript Conference
Optimizing Prime Video with WebAssembly and Rust - International JavaScript Conference
Prime Video delivers content to millions of customers, all over the world, on a variety of devices such as: game consoles, set-top boxes, streaming sticks, and Smart TVs. These devices have a vast range of hardware capabilities and performance characteristics.…
doing something
At least, it works somehow Quantized it using https://pypi.org/project/llmcompressor/ Data used for calibration: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean-punctuated Quantized model: https://huggingface.co/Yehor/whisper-large-v2-quantized…
Some tweaks helped, now it uses 3.4 GB of GPU, but spikes when vllm starts here - about 26.5 GB
Decrease
Decrease
gpu_memory_utilization parameter to get lower GPU usage
doing something
At least, it works somehow Quantized it using https://pypi.org/project/llmcompressor/ Data used for calibration: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean-punctuated Quantized model: https://huggingface.co/Yehor/whisper-large-v2-quantized…
How to quantize:
https://colab.research.google.com/drive/1TsCMxwq9kqsWV8jabihFN7J78RKgyvnD?usp=sharing
How to inference:
https://colab.research.google.com/drive/1fd6orSZqisnvjMm3Via6CdU247I5JbKo?usp=sharing
#ai #speech #whisper
https://colab.research.google.com/drive/1TsCMxwq9kqsWV8jabihFN7J78RKgyvnD?usp=sharing
How to inference:
https://colab.research.google.com/drive/1fd6orSZqisnvjMm3Via6CdU247I5JbKo?usp=sharing
#ai #speech #whisper
Google
LLM Compressor for Whisper Large V2: Calibrate the model on Yehor/cv10-uk-testset-clean-punctuated dataset.ipynb
Colab notebook
Forwarded from Hacker News
Terence Eden’s Blog
Why are QR Codes with capital letters smaller than QR codes with lower-case letters?
Take a look at these two QR codes. Scan them if you like, I promise there's nothing dodgy in them. Left is upper-case HTTPS://EDENT.TEL/ and right is lower-case https://edent.tel/ You can clearly see that the one on the left is a "smaller" QR as it has…