Skip to content
View yashb98's full-sized avatar
10 followers · 7 following

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx2Achievement: YOLO

Block or report yashb98

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
yashb98/README.md

Yash Bishnoi — ML Engineer, Edinburgh

Most of my work is reproduction under a strict evidence gate: a claim doesn't count until the number traces back to a file and survives an adversarial check. Currently working through inference-time kernels, quantisation, and torch.compile internals.

Now

BuildFromScratch — single-file LM reproductions at bit-exact parity

SmolLM2-135M and Qwen3-0.6B-Base rebuilt as single-file PyTorch models, verified against the reference weights, then used as a controlled base for architecture ablations at matched compute.

Check Result
SmolLM2-135M parity vs HuggingFace max|Δlogits| = 0.0
Qwen3-0.6B-Base parity vs HuggingFace max|Δlogits| = 0.0
Qwen3 faithful baseline ~2.14× reference, ~275,000× less training data
Modernised IMU-1 bundle @ matched compute +18% over the faithful baseline (23.52 vs 28.65)

Three builds — faithful, modernised, exploratory partial-RoPE — each with parity scripts, training pipelines and an lm-evaluation-harness wrapper.

→ github.com/yashb98/BuildFromScratch

Deep-Learning for Facial 3D Reconstruction — MSc dissertation, University of Dundee

End-to-end 3D head reconstruction pipeline: PCA normalisation, a 1,200-image rotation sweep, SixDRepNet pose estimation, and synthetic dataset rendering — built as the dissertation project for my MSc in Computer Science.

→ github.com/yashb98/Deep-Learning-for-Facial-3D-Reconstruction

multi-agent-patterns — orchestration for long-running agent systems

Patterns for agent systems that run for hours rather than turns: JobPulse daily automation and MindGraph knowledge visualisation as the working examples.

→ github.com/yashb98/multi-agent-patterns

Open source

Patches to the libraries I use daily — kernels, quantisation dispatch, and compiler internals.

Project Contribution Status
pytorch/pytorch [dynamo] Split FrozensetVariable off SetVariable to match CPython semantics (+421 / −279) In review
linkedin/Liger-Kernel Fix FLCE torch.compile failure from the aten.addmm.dtype_out overload In review
bitsandbytes-foundation/bitsandbytes Add an sm121 (GB10) tier to the 4-bit GEMM dispatch heuristic In review

Stack

PyTorch, Triton, CUDA, torch.compile, Dynamo, Transformers, TRL, LoRA, Unsloth, lm-eval-harness, TypeScript

Also built
Project What it is
LetsFineTune Fine-tuned LMs with LoRA, Unsloth and Transformers — configs, scripts, evals
Velox_AI Voice-agent platform for real phone calls: Twilio + Gemini + Deepgram, hybrid RAG
Prism Research agent decomposing million-token corpora into verifiable reasoning chains
90-days-of-ml 30+ projects across core ML, MLOps on AWS, and production capstones

Elsewhere

yashbishnoi.io · LinkedIn

Pinned Loading

  1. BuildFromScratch BuildFromScratch Public

    Single-file PyTorch reproductions of SmolLM2-135M and Qwen3-0.6B at bit-exact parity with the reference weights, plus matched-compute architecture ablations.

    Python 1

  2. Deep-Learning-for-Facial-3D-Reconstruction---Simulator Deep-Learning-for-Facial-3D-Reconstruction---Simulator Public

    MSc Computer Science project, University of Dundee. End-to-end 3D head pipeline: PCA normalisation, 1,200-image rotation sweep, SixDRepNet pose estimation, and synthetic dataset rendering.

    Jupyter Notebook

  3. multi-agent-patterns multi-agent-patterns Public

    Orchestration patterns for long-running multi-agent LLM systems, with JobPulse daily automation and MindGraph knowledge visualisation.

    Python 2

  4. Velox_AI Velox_AI Public

    Enterprise voice-agent platform for real phone calls: Twilio + Gemini + Deepgram, visual flow design, multi-agent LLM routing, hybrid RAG and production observability.

    TypeScript

  5. LetsFineTune LetsFineTune Public

    Fine-tuned language models built with LoRA, Unsloth and Hugging Face Transformers - configs, training scripts and evaluation results included.

    Jupyter Notebook

  6. 90-days-of-ml 90-days-of-ml Public

    90 days of building ML end to end: 30+ projects across core ML, MLOps on AWS, and production capstones - code, notebooks and infrastructure for each.

    Python 3