01. identity

artificial intelligence enthusiastic

i like building small language models that can code, reason like claude (via distillation) and making it to play chess very well!

view operations huggingface github

What is the capital of France?

▶ Thinking Process

Analyzing request about the capital of France.

Straightforward factual question.

Retrieving data: Paris.

Formulating concise response.

The capital of France is Paris.

Known globally as its cultural, political, and economic hub, Paris has been the city's heart since the 17th century. 🎨🗼

02. current_state

training

03. development_index

projects

project_01

LFM2.5-1.2B-Distilled-Claude

a claude sonnet 4.6 adaptive reasoning distillation into a LFM2.5-1.2b-Thinking via LoRA.

> open collection > open best model

project_02

open-i3

the open-source release of the i3 architecture. promoting transparency in ai development.

> open repo

project_03

ChessSLM

small language models trained exclusively to play chess. available variants: base, rl, neo, and nano.

> play against it

project_04

koda-waf

a small, SOTA open-weights WAF model development by me for NetGoat.

> open repo

project_05

literallynothing.js

the name is self-explanatory.

> open npm

04. model_zoo

hugging face models

model_id	params	ctx_window	hardware	access
NanoSR-6x	N/A	NaN	nvidia t4	view_card
LFM2.5-1.2B-Distilled-Claude-4.6	1.2B	12k	nvidia t4	view_card view_gguf
ChessSLM	30.3m	1024	nvidia p100	view_card

model_id	params	ctx_window	hardware	access
MathGPT2	81.9m	1024	nvidia t4	view_card
ChessSLM	30m	1024	nvidia p100	view_card
FlameF0X/Qwen2-0.2B-it	~0.2B	2048	nvidia p100	view_card

05. capabilities

technical capability stack

01. language python 3.10+

02. framework pytorch / cuda

03. distribution huggingface

04. processing hf datasets

06. contributions

fuel the research

computing power is the primary constraint. if my models aided your development, consider supporting the next training sequence.

support_via_kofi vast.ai_template