afrideva's picture

afrideva

afrideva

·

afri_deva

AI & ML interests

None yet

Recent Activity

reacted to hexgrad's post with 👍 17 days ago

I wrote an article about G2P: https://hf.co/blog/hexgrad/g2p G2P is an underrated piece of small TTS models, like offensive linemen who do a bunch of work and get no credit. Instead of relying on explicit G2P, larger speech models implicitly learn this task by eating many thousands of hours of audio data. They often use a 500M+ parameter LLM at the front to predict latent audio tokens over a learned codebook, then decode these tokens into audio. Kokoro instead relies on G2P preprocessing, is 82M parameters, and thus needs less audio to learn. Because of this, we can cherrypick high fidelity audio for training data, and deliver solid speech for those voices. In turn, this excellent audio quality & lack of background noise helps explain why Kokoro is very competitive in single-voice TTS Arenas.

reacted to davanstrien's post with ❤️ 24 days ago

Why choose between strong LLM reasoning and efficient models? Use DeepSeek to generate high-quality training data, then distil that knowledge into ModernBERT https://huggingface.co/answerdotai/ModernBERT-base for fast, efficient classification. Blog post: https://danielvanstrien.xyz/posts/2025/deepseek/distil-deepseek-modernbert.html

liked a model 27 days ago

deepseek-ai/Janus-Pro-1B

View all activity

Organizations

afrideva's activity

upvoted an article 4 months ago

Article

Model2Vec: Distill a Small Fast Model from any Sentence Transformer

By

and 1 other •

Oct 14, 2024

• 72

upvoted a paper 8 months ago

Conciseness: An Overlooked Language Task

Paper • 2211.04126 • Published Nov 8, 2022 • 2

upvoted a collection 9 months ago

IrokoBench

a human-translated benchmark dataset for 16 African languages covering three tasks: NLI, MMLU and MGSM • 6 items • Updated May 31, 2024 • 18

upvoted a paper 9 months ago

Arcee's MergeKit: A Toolkit for Merging Large Language Models

Paper • 2403.13257 • Published Mar 20, 2024 • 20

upvoted 2 collections about 1 year ago

Medical Evaluation Datasets

44 items • Updated 11 days ago • 8

Foundation Text-Generation Models Below 360M Parameters

Great candidates for fine-tuning targeting Wllama and Transformers.js for mobile devices, ordered by number of parameters. • 30 items • Updated 2 days ago • 22

upvoted 2 papers about 1 year ago

Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translation

Paper • 2401.08417 • Published Jan 16, 2024 • 35

Self-Rewarding Language Models

Paper • 2401.10020 • Published Jan 18, 2024 • 146

upvoted 2 collections about 1 year ago

Open LLM Leaderboard best models ❤️‍🔥

A daily uploaded list of models with best evaluations on the LLM leaderboard: • 64 items • Updated 2 days ago • 543

Trained Models 🏋️

They may be small, but they're training like giants! • 8 items • Updated Dec 3, 2024 • 17

upvoted 2 papers about 1 year ago

EIPE-text: Evaluation-Guided Iterative Plan Extraction for Long-Form Narrative Text Generation

Paper • 2310.08185 • Published Oct 12, 2023 • 8

TinyGSM: achieving >80% on GSM8k with small language models

Paper • 2312.09241 • Published Dec 14, 2023 • 39

upvoted 3 collections about 1 year ago

ChatGPT-Mini

A collection of fine-tuned GPT-2 models each designed to deploy a ChatGPT-like model at home. These models can also be deployed on an old computer. • 8 items • Updated Nov 16, 2023 • 5

Merged Models

Using mergekit • 10 items • Updated Mar 1, 2024 • 3

smol llama

🚧"raw" pretrained smol_llama checkpoints - WIP 🚧 • 4 items • Updated Apr 29, 2024 • 6

upvoted 5 collections over 1 year ago

Coding datasets

3 items • Updated Nov 23, 2023 • 4

Indic language fine-tunes

Halted State: Attempting to create acceptable quality fine-tunes of different models • 1 item • Updated Nov 23, 2023 • 1

PIC (Partner-in-Crime) project

Empathetic, small, really useful personalised models. • 3 items • Updated Dec 10, 2023 • 2

Cramp(ed) Models

Smaller models trained locally on my 2xA6000 Lambda Vector • 3 items • Updated Oct 10, 2023 • 1

Shrink Llama - V1

Parts of Meta's LlamaV2 models, chopped up and trained. CoreX means the first X layers were kept. • 2 items • Updated Sep 12, 2023 • 2