A home for my notes, small builds, and writeups as I study LLMs and deep learning from scratch.

LittleLearner Experiments: Teaching Math Beyond K–5

I trained 16 LoRA adapters on K–5-constrained LittleLearner models and found that useful mathematical skill transfer changed with…

September 2, 2026

Can a Warmth-Trained Model Learn When Not to Agree?

I compared reasons, a matched-length control, and character training after warmth SFT; none increased judged warmth, while M4…

August 20, 2026

When Emergent Misalignment Crosses Languages

English risky-financial fine-tuning transferred coherent emergent misalignment into Hindi, Marathi, and Urdu, while matched…

August 17, 2026

Making an Image Editor with LongCat and a Small LoRA

My first image-generation research experiment: teaching LongCat-Image-Dev source-conditioned editing, finding global…

August 12, 2026

Narrow Fine-Tuning, Broad Misalignment in a 3B Model

I fine-tuned a 3.35B model on narrowly risky financial advice and found a reproducible shift toward coherent misalignment on…

August 10, 2026

Before It Codes: Catching Qwen3.5-4B Planning With J-Lens

I used Anthropic's new Jacobian lens to test whether Qwen3.5-4B represents an algorithm before it writes code, then built a live…

July 30, 2026

Hybrid Token-Efficient Routing Agent

How I built TOKENMAN for AMD Developer Hackathon ACT II: a deterministic, local-model, and Fireworks routing agent that climbed…

July 22, 2026

World Fan Arena

A public football fan simulation where AI fan agents from different countries react to matches, rivalries, upsets, predictions…

July 16, 2026

[ICML ’26 Effort] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G

My ICML 2026 AdaptFM effort to optimize Qwen3.5-4B for low-latency inference on a single NVIDIA A10G through 100+ runtime…

June 30, 2026

Looking Inside Qwen3-4B With Sparse Autoencoders

I trained a sparse autoencoder on Qwen3-4B-Base, labeled its learned features, and tested one by steering the model toward…

June 12, 2026

Pre-Training My First Base Language Model From Scratch

Took a model from random initialization through 2B tokens of FineWeb-Edu on a single H100 and watched it learn next-token…

May 13, 2026

Parameter Golf: My OpenAI Model Craft Experiment

I didn't come close to the Parameter Golf leaderboard, but I still had a lot of fun running scattered H100 experiments on Modal…

May 4, 2026

CrisisOps: My (Final Round) OpenEnv Hackathon Project

How I built CrisisOps for the OpenEnv hackathon finale, then trained a small model with GRPO using TRL, Unsloth, and Modal.

April 29, 2026

My First Successful RL Training Run With OpenEnv And Wordle

A full fine-tune of Qwen3-1.7B on Wordle using OpenEnv, with a reward curve that actually went up and shaped rewards that taught…

April 2, 2026

My Second RL Experiment On Prime Intellect Lab

My second RL experiment while studying GRPO: moving to Prime Intellect Lab, testing smaller and larger setups, and finally…

March 25, 2026

First RL Experimental Project: RLVR Using GRPO With TRL On Modal

My first ever RL experiment: RLVR on GSM8K using Hugging Face TRL, Qwen2.5 1.5B Instruct, and an NVIDIA H100 on Modal, with notes…

March 23, 2026

My Current Approach in Learning and Experimenting with LLMs and Deep Learning

How I study LLMs by going deep on specific topics instead of starting from math.

March 18, 2026

Teaching a 3B Base Model to Emit Reasoning Traces

A first small-scale AI research experiment: cold-start SFT on Nanbeige4-3B-Base using 2,160 distilled reasoning triplets, trained…

March 11, 2026

September 2, 2026reads

LittleLearner Experiments: Teaching Math Beyond K–5

I trained 16 LoRA adapters on K–5-constrained LittleLearner models and found that useful mathematical skill transfer changed with model scale.

August 20, 2026reads

Can a Warmth-Trained Model Learn When Not to Agree?

I compared reasons, a matched-length control, and character training after warmth SFT; none increased judged warmth, while M4 most strongly resisted explicit persona attacks.

August 17, 2026reads

When Emergent Misalignment Crosses Languages

English risky-financial fine-tuning transferred coherent emergent misalignment into Hindi, Marathi, and Urdu, while matched prudent controls stayed near zero.

August 12, 2026reads

Making an Image Editor with LongCat and a Small LoRA

My first image-generation research experiment: teaching LongCat-Image-Dev source-conditioned editing, finding global regeneration, and improving preservation by filtering the data.

August 10, 2026reads

Narrow Fine-Tuning, Broad Misalignment in a 3B Model

I fine-tuned a 3.35B model on narrowly risky financial advice and found a reproducible shift toward coherent misalignment on unrelated prompts.

July 30, 2026reads

Before It Codes: Catching Qwen3.5-4B Planning With J-Lens

I used Anthropic's new Jacobian lens to test whether Qwen3.5-4B represents an algorithm before it writes code, then built a live visualizer around the result.

July 22, 2026reads

Hybrid Token-Efficient Routing Agent

How I built TOKENMAN for AMD Developer Hackathon ACT II: a deterministic, local-model, and Fireworks routing agent that climbed from 6,101 counted tokens to zero.

July 16, 2026reads

World Fan Arena

A public football fan simulation where AI fan agents from different countries react to matches, rivalries, upsets, predictions and tournament drama inside chat rooms.

June 30, 2026reads

[ICML ’26 Effort] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G

My ICML 2026 AdaptFM effort to optimize Qwen3.5-4B for low-latency inference on a single NVIDIA A10G through 100+ runtime, compiler, and quantization experiments.

June 12, 2026reads

Looking Inside Qwen3-4B With Sparse Autoencoders

I trained a sparse autoencoder on Qwen3-4B-Base, labeled its learned features, and tested one by steering the model toward cooking instructions.

May 13, 2026reads

Pre-Training My First Base Language Model From Scratch

Took a model from random initialization through 2B tokens of FineWeb-Edu on a single H100 and watched it learn next-token prediction from nothing.

May 4, 2026reads

Parameter Golf: My OpenAI Model Craft Experiment

I didn't come close to the Parameter Golf leaderboard, but I still had a lot of fun running scattered H100 experiments on Modal, hunting tiny BPB improvements while watching ideas collapse against artifact size limits, and figuring out the hard way why squeezing a capable model into 16 MB is trickier than it sounds.

April 29, 2026reads

CrisisOps: My (Final Round) OpenEnv Hackathon Project

How I built CrisisOps for the OpenEnv hackathon finale, then trained a small model with GRPO using TRL, Unsloth, and Modal.

April 2, 2026reads

My First Successful RL Training Run With OpenEnv And Wordle

A full fine-tune of Qwen3-1.7B on Wordle using OpenEnv, with a reward curve that actually went up and shaped rewards that taught real gameplay behavior.

March 25, 2026reads

My Second RL Experiment On Prime Intellect Lab

My second RL experiment while studying GRPO: moving to Prime Intellect Lab, testing smaller and larger setups, and finally getting a reward curve that went up.

March 23, 2026reads

First RL Experimental Project: RLVR Using GRPO With TRL On Modal

My first ever RL experiment: RLVR on GSM8K using Hugging Face TRL, Qwen2.5 1.5B Instruct, and an NVIDIA H100 on Modal, with notes on where the reward function broke and what I want to fix in v2.

March 18, 2026reads

My Current Approach in Learning and Experimenting with LLMs and Deep Learning

How I study LLMs by going deep on specific topics instead of starting from math.

March 11, 2026reads

Teaching a 3B Base Model to Emit Reasoning Traces

A first small-scale AI research experiment: cold-start SFT on Nanbeige4-3B-Base using 2,160 distilled reasoning triplets, trained with LoRA on an H100, with notes on what worked and what I want to try next.