ANLP Assignment 3 Part 2: graph classifiers for UPFD GossipCop
VANSH NAWANDER
vanshnawander
AI & ML interests
machine learning, Large Language Models, NLP, neural networks, TensorFlow
Recent Activity
updated a model 1 day ago
vanshnawander/anlp-a3-hybrid published a model 1 day ago
vanshnawander/anlp-a3-hybrid updated a model 1 day ago
vanshnawander/anlp-a3-deltanetOrganizations
SAPC2 Track 2: compact CPU-streamable ASR for dysarthric speech (Moonshine Tiny + RL, distilled students, CTC packages); full-Dev WER/CER, latency.
-
vanshnawander/sapc2-moonshine-tiny-rl
Automatic Speech Recognition • 27.1M • Updated • 84 -
vanshnawander/sapc2-moonshine-m24
Automatic Speech Recognition • 23.8M • Updated • 93 -
vanshnawander/sapc2-moonshine-m15s
Automatic Speech Recognition • 14.6M • Updated • 117 -
vanshnawander/sapc2-moonshine-m10s
Automatic Speech Recognition • 10.7M • Updated • 92
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%. Saved before evaluation.
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%.
Assignment 2 MoE model checkpoints
SAPC2 Track 1: large ASR models adapted to dysarthric speech (Cohere Transcribe LoRA / SOAP SFT / RL, other selected checkpoints); full-Dev WER/CER.
Fine-tuned SliceGPT checkpoints for LLaDA and Dream.
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%.
Assignment 2 Optimizer model checkpoints
ANLP Assignment 3 Part 2: graph classifiers for UPFD GossipCop
SAPC2 Track 1: large ASR models adapted to dysarthric speech (Cohere Transcribe LoRA / SOAP SFT / RL, other selected checkpoints); full-Dev WER/CER.
SAPC2 Track 2: compact CPU-streamable ASR for dysarthric speech (Moonshine Tiny + RL, distilled students, CTC packages); full-Dev WER/CER, latency.
-
vanshnawander/sapc2-moonshine-tiny-rl
Automatic Speech Recognition • 27.1M • Updated • 84 -
vanshnawander/sapc2-moonshine-m24
Automatic Speech Recognition • 23.8M • Updated • 93 -
vanshnawander/sapc2-moonshine-m15s
Automatic Speech Recognition • 14.6M • Updated • 117 -
vanshnawander/sapc2-moonshine-m10s
Automatic Speech Recognition • 10.7M • Updated • 92
Fine-tuned SliceGPT checkpoints for LLaDA and Dream.
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%. Saved before evaluation.
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%.
SparseGPT, sink-aware, and SliceGPT at 20%, 30%, and 40%.
Assignment 2 Optimizer model checkpoints
Assignment 2 MoE model checkpoints