Skip to content
View PineappleBlowsnow's full-sized avatar
2 followers · 3 following

Achievements

Achievement: Pull Shark

Achievements

Achievement: Pull Shark

Highlights

  • Pro

Block or report PineappleBlowsnow

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.

Content in all repositories owned by your account will be closed.
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
PineappleBlowsnow/README.md

Ying Jin

I studied applied mathematics in the MVA master's program at ENS Paris-Saclay and worked on neural-network growth with the Inria TAU team. My projects cover language-model training, multimodal learning, model analysis and tool-using agents.

MVA projects

Selected work

Project What I worked on MVA course
Chess move prediction Compact Transformer, custom tokenization, training and move-legality evaluation. LLMs for Code and Proof
Multimodal regression Conditional density estimation, mixture models and gated experts. Probabilistic Graphical Models
Memory-efficient LLM training Optimizer implementation, training/resume and memory analysis. Training and Deploying Large-Scale Models
Molecular graph-to-text Graph-text alignment, GINE encoders and language-model conditioning. Advanced Learning for Text and Graph Data (ALTeGraD)
LLM attention and activation analysis Input ablations and cross-model interpretability diagnostics. Explainable AI
Tool-using text-adventure agent ReAct, failure memory, state handling and recovery. LLMs for Code and Proof
More MVA projects and practical implementations
Project Implementation focus MVA course
KV caching and LoRA KV-cached autoregressive decoding and LoRA adaptation in a small PyTorch Transformer. LLMs for Code and Proof
Tiled attention in Triton Online-softmax attention forward, a PyTorch backward, and correctness/performance experiments. LLMs for Code and Proof
Constrained text decoding Logit masks and beam search for token and sequence constraints. LLMs for Code and Proof
Decentralized training with Gossip SGD PyTorch/Gloo training across graph topologies and communication budgets. Training and Deploying Large-Scale Models
Sequence and image modeling with Mamba S6 Collaborative PyTorch implementation for sequences and bidirectional image classification. Deep Learning
Image restoration with consistency models Consistency training/distillation and comparisons with diffusion posterior sampling. Generative Models for Images
Static mean-field games Annealed TRPA and exploratory principal intervention. Interactions

Each repository links to its implementation, available reports and experiment evidence, and explains individual contributions, collaborators and course-provided components.

Other work

  • ADVI and DADVI — Joint M1 research project at Université Paris Dauphine–PSL on variational Bayesian inference.
  • GroMo — My fork of the collaborative neural-network-growth library associated with my Inria TAU internship.
  • Neural-network components from scratch — Independent implementation exercises.

LinkedIn

Pinned Loading

  1. apollo-llm-memory-optimization apollo-llm-memory-optimization Public

    APOLLO optimizer experiments for LLM training: optimizer-state memory and convergence on TinyStories.

    Python

  2. chess-move-transformer chess-move-transformer Public

    Training a compact Transformer to predict chess moves, with a custom tokenizer, saved weights and legality evaluation.

    Python

  3. llm-attention-sinks-and-outliers llm-attention-sinks-and-outliers Public

    Forked from Tristan22400/xai-outlier-repro

    Investigating attention sinks and activation outliers in language models through input ablations and diagnostics.

    Python

  4. mixture-density-networks mixture-density-networks Public

    Conditional density estimation with mixture density networks and gated experts on multimodal regression tasks.

    Python

  5. molecular-graph-to-text molecular-graph-to-text Public

    Molecular graph captioning with GINE graph-text retrieval and graph-conditioned language models.

    Jupyter Notebook

  6. zork-llm-agent zork-llm-agent Public

    A ReAct language-model agent for text adventures with failed-action memory and recovery from repeated actions.

    Python