I'm planning more systematic CPU latency and evaluations to better characterize the accuracy efficiency trade off. Appreciate the thoughtful feedback
Mahwiz Khalil
mahwizzzz
AI & ML interests
I build ASR, TTS, NLP and fine-tuned LLM systems from dataset creation and evaluation to optimization, ONNX deployment and production inference.
Recent Activity
repliedto their post 5 days ago
Released: Urdu Streaming Zipformer 20M
Introducing Urdu Streaming ASR, a compact causal Zipformer transducer developed for Urdu speech recognition.
FLEURS Urdu (299 utterances):
WER: 28.03%
CER: 11.96%
Model: https://huggingface.co/mahwizzzz/Urdu-Streaming-Zipformer-20M
Source: https://github.com/mwzkhalil/Urdu-Streaming-ASR liked a model 6 days ago
muose/Muose-50M-Decision posted an update 6 days ago
Released: Urdu Streaming Zipformer 20M
Introducing Urdu Streaming ASR, a compact causal Zipformer transducer developed for Urdu speech recognition.
FLEURS Urdu (299 utterances):
WER: 28.03%
CER: 11.96%
Model: https://huggingface.co/mahwizzzz/Urdu-Streaming-Zipformer-20M
Source: https://github.com/mwzkhalil/Urdu-Streaming-ASROrganizations
replied to their post 5 days ago
posted an update 6 days ago
Post
1,029
Released: Urdu Streaming Zipformer 20M
Introducing Urdu Streaming ASR, a compact causal Zipformer transducer developed for Urdu speech recognition.
FLEURS Urdu (299 utterances):
WER: 28.03%
CER: 11.96%
Model: mahwizzzz/Urdu-Streaming-Zipformer-20M
Source: https://github.com/mwzkhalil/Urdu-Streaming-ASR
Introducing Urdu Streaming ASR, a compact causal Zipformer transducer developed for Urdu speech recognition.
FLEURS Urdu (299 utterances):
WER: 28.03%
CER: 11.96%
Model: mahwizzzz/Urdu-Streaming-Zipformer-20M
Source: https://github.com/mwzkhalil/Urdu-Streaming-ASR
- 2 replies
posted an update about 2 months ago
Post
284
I implemented the attention-free bidirectional encoder architecture Avey-B for Urdu a compact 24.87M-parameter language encoder built for efficient Urdu NLP research.
Original Avey-B paper: Avey-B (2602.15814)
Urdu model: https://huggingface.co/mahwizzzz/avey-b-ur
Original Avey-B paper: Avey-B (2602.15814)
Urdu model: https://huggingface.co/mahwizzzz/avey-b-ur
reacted to ProCreations's post with ๐ 3 months ago
Post
3,443
I have hit 300 followers, and I think this calls for a bit of a giveaway ๐ a unique one, too.
I have had countless AI projects I have wanted to make but have been (brutally) blocked by compute. Now that I finally have just enough compute to sort of get around (i still don't have enough ๐ญ) and for hitting 300 followers (tysm!) I will be funding three of the communities projects via HuggingFace jobs, giving them 150 dollars max worth of compute each. I will personally be picking the winners, I am looking for projects that genuinely hit the compute wall: great ideas, blocked by compute, just like the countless ideas I've had.
To join, head over to https://giveaway.ssh.codes
RULES:
- Final result must be open weight or open source
- Only one submission per person
- Have fun!
I have had countless AI projects I have wanted to make but have been (brutally) blocked by compute. Now that I finally have just enough compute to sort of get around (i still don't have enough ๐ญ) and for hitting 300 followers (tysm!) I will be funding three of the communities projects via HuggingFace jobs, giving them 150 dollars max worth of compute each. I will personally be picking the winners, I am looking for projects that genuinely hit the compute wall: great ideas, blocked by compute, just like the countless ideas I've had.
To join, head over to https://giveaway.ssh.codes
RULES:
- Final result must be open weight or open source
- Only one submission per person
- Have fun!
- 36 replies
posted an update 3 months ago
Post
220
If you're into running models locally, this might be worth checking out.
https://local.ai/mahwiz/invite
https://local.ai/mahwiz/invite
posted an update 5 months ago
Post
407
Released lafzyn , built over Qwen, an Urdu language model that converts Urdu text into IPA phonetic transcription, with GGUF builds for local inference.
Release contents:
- mahwizzzz/lafzyn: full weights
- mahwizzzz/lafzyn-gguf: quantized builds
Try it out ๐ค
Demo: https://huggingface.co/spaces/mahwizzzz/Lafzyn
Release contents:
- mahwizzzz/lafzyn: full weights
- mahwizzzz/lafzyn-gguf: quantized builds
Try it out ๐ค
Demo: https://huggingface.co/spaces/mahwizzzz/Lafzyn
reacted to Kseniase's post with ๐ over 1 year ago
Post
6,589
12 Foundational AI Model Types
Letโs refresh some fundamentals today to stay fluent in the what we all work with. Here are some of the most popular model types that shape the vast world of AI (with examples in the brackets):
1. LLM - Large Language Model (GPT, LLaMA) -> Large Language Models: A Survey (2402.06196)
+ history of LLMs: https://www.turingpost.com/t/The%20History%20of%20LLMs
It's trained on massive text datasets to understand and generate human language. They are mostly build on Transformer architecture, predicting the next token. LLMs scale by increasing overall parameter count across all components (layers, attention heads, MLPs, etc.)
2. SLM - Small Language Model (TinyLLaMA, Phi models, SmolLM) A Survey of Small Language Models (2410.20011)
Lightweight LM optimized for efficiency, low memory use, fast inference, and edge use. SLMs work using the same principles as LLMs
3. VLM - Vision-Language Model (CLIP, Flamingo) -> An Introduction to Vision-Language Modeling (2405.17247)
Processes and understands both images and text. VLMs map images and text into a shared embedding space or generate captions/descriptions from both
4. MLLM - Multimodal Large Language Model (Gemini) -> A Survey on Multimodal Large Language Models (2306.13549)
A large-scale model that can understand and process multiple types of data (modalities) โ usually text + other formats, like images, videos, audio, structured data, 3D or spatial inputs. MLLMs can be LLMs extended with modality adapters or trained jointly across vision, text, audio, etc.
5. LAM - Large Action Model (InstructDiffusion, RT-2) -> Large Action Models: From Inception to Implementation (2412.10047)
Understands and generates action sequences by predicting action tokens (discrete/continuous instructions) that guide agents. Trained on behavior datasets, LAMs generalize across tasks, environments, and modalities - video, sensor data, etc.
Read about LRM, MoE, SSM, RNN, CNN, SAM and LNN below๐
Also, subscribe to the Turing Post: https://www.turingpost.com/subscribe
Letโs refresh some fundamentals today to stay fluent in the what we all work with. Here are some of the most popular model types that shape the vast world of AI (with examples in the brackets):
1. LLM - Large Language Model (GPT, LLaMA) -> Large Language Models: A Survey (2402.06196)
+ history of LLMs: https://www.turingpost.com/t/The%20History%20of%20LLMs
It's trained on massive text datasets to understand and generate human language. They are mostly build on Transformer architecture, predicting the next token. LLMs scale by increasing overall parameter count across all components (layers, attention heads, MLPs, etc.)
2. SLM - Small Language Model (TinyLLaMA, Phi models, SmolLM) A Survey of Small Language Models (2410.20011)
Lightweight LM optimized for efficiency, low memory use, fast inference, and edge use. SLMs work using the same principles as LLMs
3. VLM - Vision-Language Model (CLIP, Flamingo) -> An Introduction to Vision-Language Modeling (2405.17247)
Processes and understands both images and text. VLMs map images and text into a shared embedding space or generate captions/descriptions from both
4. MLLM - Multimodal Large Language Model (Gemini) -> A Survey on Multimodal Large Language Models (2306.13549)
A large-scale model that can understand and process multiple types of data (modalities) โ usually text + other formats, like images, videos, audio, structured data, 3D or spatial inputs. MLLMs can be LLMs extended with modality adapters or trained jointly across vision, text, audio, etc.
5. LAM - Large Action Model (InstructDiffusion, RT-2) -> Large Action Models: From Inception to Implementation (2412.10047)
Understands and generates action sequences by predicting action tokens (discrete/continuous instructions) that guide agents. Trained on behavior datasets, LAMs generalize across tasks, environments, and modalities - video, sensor data, etc.
Read about LRM, MoE, SSM, RNN, CNN, SAM and LNN below๐
Also, subscribe to the Turing Post: https://www.turingpost.com/subscribe
- 2 replies
reacted to hesamation's post with โค๏ธ over 1 year ago
Post
10,723
Google published a 69-page whitepaper on Prompt Engineering and its best practices, a must-read if you are using LLMs in production:
> zero-shot, one-shot, few-shot
> system prompting
> chain-of-thought (CoT)
> ReAct
LINK: https://www.kaggle.com/whitepaper-prompt-engineering
> code prompting
> best practices
> zero-shot, one-shot, few-shot
> system prompting
> chain-of-thought (CoT)
> ReAct
LINK: https://www.kaggle.com/whitepaper-prompt-engineering
> code prompting
> best practices
reacted to AlexBodner's post with ๐ about 2 years ago
Post
3,858
๐พ๐ง How much VRAM will you need for training your AI model? ๐พ๐ง
Check out this app where you convert:
Pytorch/tensorflow summary -> needed VRAM
or
Parameter count -> needed VRAM
Use it in: http://howmuchvram.com
And everything is open source! Ask for new functionalities or contribute in:
https://github.com/AlexBodner/How_Much_VRAM
If it's useful to you leave a star ๐and share it to someone that will find the tool useful!
Check out this app where you convert:
Pytorch/tensorflow summary -> needed VRAM
or
Parameter count -> needed VRAM
Use it in: http://howmuchvram.com
And everything is open source! Ask for new functionalities or contribute in:
https://github.com/AlexBodner/How_Much_VRAM
If it's useful to you leave a star ๐and share it to someone that will find the tool useful!
- 3 replies
posted an update over 2 years ago
Post
2,793
[ { "from": "human", "value": "First post ๐ค" }]
- 4 replies
reacted to victor's post with ๐ almost 3 years ago
Post
Let's play a little game, how would you build the Rabbit R1 with open source tech? Here is my stack:
- openai/whisper-small for awesome Speech-to-Text with low latency
- mistralai/Mixtral-8x7B-Instruct-v0.1 for an awesome super powerful LLM Brain
- coqui/XTTS-v2 for a nice and clean voice
Which stack will you personally choose?
- openai/whisper-small for awesome Speech-to-Text with low latency
- mistralai/Mixtral-8x7B-Instruct-v0.1 for an awesome super powerful LLM Brain
- coqui/XTTS-v2 for a nice and clean voice
Which stack will you personally choose?
- 19 replies
reacted to akhaliq's post with ๐ almost 3 years ago
Post
Here is my selection of papers for today (9 Jan)
https://huggingface.co/papers
AGG: Amortized Generative 3D Gaussians for Single Image to 3D
MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts
DiarizationLM: Speaker Diarization Post-Processing with Large Language Models
TeleChat Technical Report
Soaring from 4K to 400K: Extending LLM's Context with Activation Beacon
AST-T5: Structure-Aware Pretraining for Code Generation and Understanding
Has Your Pretrained Model Improved? A Multi-head Posterior Based Approach
Blending Is All You Need: Cheaper, Better Alternative to Trillion-Parameters LLM
GPT-4V(ision) is a Human-Aligned Evaluator for Text-to-3D Generation
CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution
Mixtral of Experts
https://huggingface.co/papers
AGG: Amortized Generative 3D Gaussians for Single Image to 3D
MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts
DiarizationLM: Speaker Diarization Post-Processing with Large Language Models
TeleChat Technical Report
Soaring from 4K to 400K: Extending LLM's Context with Activation Beacon
AST-T5: Structure-Aware Pretraining for Code Generation and Understanding
Has Your Pretrained Model Improved? A Multi-head Posterior Based Approach
Blending Is All You Need: Cheaper, Better Alternative to Trillion-Parameters LLM
GPT-4V(ision) is a Human-Aligned Evaluator for Text-to-3D Generation
CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution
Mixtral of Experts
reacted to akhaliq's post with ๐ almost 3 years ago
Post
Here is my selection of papers for today (4 Jan)
https://huggingface.co/papers
Efficient Hybrid Zoom using Camera Fusion on Mobile Phones
Incremental FastPitch: Chunk-based High Quality Text to Speech
CoMoSVC: Consistency Model-based Singing Voice Conversion
SIGNeRF: Scene Integrated Generation for Neural Radiance Fields
From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations
aMUSEd: An Open MUSE Reproduction
Image Sculpting: Precise Object Editing with 3D Geometry Control
A Vision Check-up for Language Models
Multilingual Instruction Tuning With Just a Pinch of Multilinguality
WordArt Designer API: User-Driven Artistic Typography Synthesis with Large Language Models on ModelScope
Moonshot: Towards Controllable Video Generation and Editing with Multimodal Conditions
GPT-4V(ision) is a Generalist Web Agent, if Grounded
https://huggingface.co/papers
Efficient Hybrid Zoom using Camera Fusion on Mobile Phones
Incremental FastPitch: Chunk-based High Quality Text to Speech
CoMoSVC: Consistency Model-based Singing Voice Conversion
SIGNeRF: Scene Integrated Generation for Neural Radiance Fields
From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations
aMUSEd: An Open MUSE Reproduction
Image Sculpting: Precise Object Editing with 3D Geometry Control
A Vision Check-up for Language Models
Multilingual Instruction Tuning With Just a Pinch of Multilinguality
WordArt Designer API: User-Driven Artistic Typography Synthesis with Large Language Models on ModelScope
Moonshot: Towards Controllable Video Generation and Editing with Multimodal Conditions
GPT-4V(ision) is a Generalist Web Agent, if Grounded
Interesting paper! 'LLaMA Beyond English'
reacted to akhaliq's post with ๐ almost 3 years ago
Post
Here is my selection of papers for today (3 Jan) on Hugging Face
paper pages: https://huggingface.co/papers
En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D Synthetic Data
Boundary Attention: Learning to Find Faint Boundaries at Any Resolution
Taming Mode Collapse in Score Distillation for Text-to-3D Generation
LLaMA Beyond English: An Empirical Study on Language Capability Transfer
Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models
A Comprehensive Study of Knowledge Editing for Large Language Models
Q-Refine: A Perceptual Quality Refiner for AI-Generated Image
LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning
DocLLM: A layout-aware generative language model for multimodal document understanding
VideoDrafter: Content-Consistent Multi-Scene Video Generation with LLM
TrailBlazer: Trajectory Control for Diffusion-Based Video Generation
paper pages: https://huggingface.co/papers
En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D Synthetic Data
Boundary Attention: Learning to Find Faint Boundaries at Any Resolution
Taming Mode Collapse in Score Distillation for Text-to-3D Generation
LLaMA Beyond English: An Empirical Study on Language Capability Transfer
Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models
A Comprehensive Study of Knowledge Editing for Large Language Models
Q-Refine: A Perceptual Quality Refiner for AI-Generated Image
LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning
DocLLM: A layout-aware generative language model for multimodal document understanding
VideoDrafter: Content-Consistent Multi-Scene Video Generation with LLM
TrailBlazer: Trajectory Control for Diffusion-Based Video Generation
- 1 reply
