blogs
Research
-
Findings: Karpathy-style autoresearch on a crypto backtester (local LLM)
Local Qwen 3.5 autoresearch on my crypto DB + Nautilus-style backtester (~2h, 30+ iter, $0 API): tool-calling blocker, run observations, human-in-the-loop steering, GA contrast, diversity, gates.
-
Brief Review on Generative Modeling by Estimating Gradients of the Data Distribution
Review notes on score matching, DSM, and score-based generative modeling.
-
Brief Review on SRDiff: Single Image Super-Resolution with Diffusion Probabilistic Models
A review of SRDiff, a diffusion-based method for single image super-resolution with residual conditioning.
-
Brief Review on High-Resolution Image Synthesis with Latent Diffusion Models (LDM)
Review notes on latent diffusion models, latent-space denoising, and guidance methods.
-
Code for Denoising Diffusion Probabilistic Models (DDPM)
Code walkthrough notes for implementing DDPM with U-Net and diffusion scheduling.
-
Denoising Diffusion Probabilistic Models (DDPM) from Bayes' Theorem
DDPM derivation notes from a Bayes perspective.
-
Brief Review on Denoising Diffusion Implicit Models (DDIM)
Review notes on DDIM and its non-Markovian sampling derivation.
-
Brief Review on Denoising Diffusion Probabilistic Models (DDPM)
Review notes on DDPM forward and reverse diffusion processes.
-
Reparameterization Trick
Notes on the reparameterization trick in VAE and diffusion.
Journey
-
IELTS - After Class Note, Week 8
Final-week listening, writing, speaking, and reading reminders for IELTS.
-
IELTS - After Class Note, Week 7
Listening, writing task structure, and vocabulary from IELTS week 7.
-
IELTS - After Class Note, Week 6
Exam strategy and section-specific tips from IELTS week 6.
-
IELTS - After Class Note, Week 5
Collocation and usage notes from IELTS week 5.
-
IELTS - After Class Note, Week 4
Usage patterns, listening traps, and reading notes from IELTS week 4.
-
IELTS - After Class Note, Week 3
Grammar usage notes from IELTS week 3 after-class session.
-
IELTS - After Class Note, Week 2
Usage, collocation, and common mistakes from IELTS week 2 notes.
-
IELTS - After Class Note, Week 1
Introduction to IELTS format and after-class notes from week 1.
Bug-fixes & Workaround
-
qwen3.6-enhanced.jinja: CoT leakage into tool turns and why preserve_thinking works now
Why Qwen 3.6 with qwen3.5-enhanced.jinja forced preserve_thinking=false, and how qwen3.6-enhanced.jinja restores full Qwen 3.6-series capability—self-healing think/tool boundaries, safe preserve_thinking. Launch recipe tested on vLLM v0.19.0.
-
Qwen 3.6-27B-FP8 on vLLM: enhanced.jinja, qwen3_coder, and fixing NCCL after Studio Driver 595.79
Same qwen3.5-enhanced.jinja and mixed-GPU stack as earlier Qwen 3.5 notes; switching to qwen3_coder for 3.6, mandatory preserve_thinking=false, and NCCL overrides that stopped deadlocks on NVIDIA Studio 595.79—plus a 180k-token agentic run.
-
Qwen 3.6 35B-A3B on vLLM: do the Qwen 3.5 tool-calling fixes carry over?
Follow-up testing: same qwen3_xml parser, qwen3.5-enhanced.jinja template, and mixed-GPU tuning as Qwen 3.5-27B—plus three agentic runs comparing official vs enhanced configs on Qwen3.6-35B-A3B-FP8.
-
Claude Code with local vLLM: client validation, model aliases, and a working settings.json
Run Claude Code against local vLLM without Anthropic API access: why common env-only recipes fail, the alias + settings.json pattern that works, and when this matters if you cannot register or use the Claude API.
-
Stable tool calling for Qwen 3.5 27B/35B on vLLM: template, parser, and mixed-GPU fixes
Debugging notes on Jinja chat templates, qwen3_xml vs qwen3_coder parsers, mixed-GPU FP8 drift, and SFT-distilled checkpoints when running Qwen 3.5 27B/35B-class models for long agentic sessions on vLLM.
-
Workaround for Enabling NCCL P2P Communication for NVIDIA RTX 4090 Workstations
What NCCL P2P means, why it matters on multi-GPU workstations, how Resizable BAR fits in, and a concrete setup path for RTX 4090.
-
Workaround for Debugging Windows 11 Installation
A journal of debugging Windows 11 installation.
Reflection
-
反思：港元穩定幣的宿命——不可能三角視角下的貨幣迷思
從不可能三角視角探討港元穩定幣的結構性宿命，分析香港聯繫匯率制度下穩定幣缺乏獨立貨幣價值的根本原因，以及其作為人民幣國際化探路者的戰略角色。
-
Reflection: The Fate of the HKD Stablecoin — A Monetary Delusion Through the Lens of the Impossible Trinity
Examining the structural fate of HKD stablecoins through the Impossible Trinity framework, analyzing why they lack independent monetary value under Hong Kong's Linked Exchange Rate System and their role as pathfinders for RMB internationalization.
-
Why I built this blog?
Why I built this website and what I will document here.