Hacker News: codelion

New comment by codelion in "Velonus – Open-source AppSec scanner that deduplicates SAST noise"

codelion — Fri, 15 May 2026 04:35:40 +0000

You can consider using Frame for the SAST part - https://github.com/lambdasec/frame

New comment by codelion in "Ollama is now powered by MLX on Apple Silicon in preview"

codelion — Tue, 31 Mar 2026 04:50:16 +0000

How does it compare to some of the newer mlx inference engines like optiq that support turboquantization - https://mlx-optiq.pages.dev/

Zero-Dependency Programming

codelion — Sun, 29 Mar 2026 22:53:42 +0000

Article URL: https://conjure.pages.dev/

Comments URL: https://news.ycombinator.com/item?id=47568297

Points: 2

# Comments: 0

New comment by codelion in "US Court of Appeals: TOS may be updated by email, use can imply consent [pdf]"

codelion — Mon, 09 Mar 2026 07:53:39 +0000

the key issue is the interpretation of "consent" when continued use is the only option. aree users truly consenting, or are they simply left with no alternative?

Scaling Pedagogical Pre-Training: From Optimal Mixing to 10B Tokens

codelion — Mon, 09 Mar 2026 07:52:37 +0000

Article URL: https://huggingface.co/blog/codelion/scaling-pedagogical-pretraining-10-billion-tokens

Comments URL: https://news.ycombinator.com/item?id=47305981

Points: 2

# Comments: 0

From HashHop to Memory-Augmented Language Models

codelion — Sat, 31 Jan 2026 05:40:42 +0000

Article URL: https://huggingface.co/blog/codelion/reverse-engineering-magic-hashhop

Comments URL: https://news.ycombinator.com/item?id=46833832

Points: 2

# Comments: 0

The Optimal Architecture for Small Language Models

codelion — Fri, 26 Dec 2025 07:46:59 +0000

Article URL: https://huggingface.co/blog/codelion/optimal-model-architecture

Comments URL: https://news.ycombinator.com/item?id=46390114

Points: 1

# Comments: 0

Enhancing LLMs with LoRA – Standardized Recipes for Capability Enhancement

codelion — Tue, 16 Dec 2025 05:56:21 +0000

Article URL: https://huggingface.co/blog/codelion/ellora-lora-recipes

Comments URL: https://news.ycombinator.com/item?id=46285260

Points: 3

# Comments: 0

OpenEvolve: Teaching LLMs to Discover Algorithms Through Evolution

codelion — Tue, 09 Dec 2025 22:54:33 +0000

Article URL: https://algorithmicsuperintelligence.ai/blog/openevolve-overview/index.html

Comments URL: https://news.ycombinator.com/item?id=46211861

Points: 53

# Comments: 9

Ellora: Enhancing LLMs with LoRA Standardized Recipes for Capability Enhancement

codelion — Wed, 03 Dec 2025 04:09:57 +0000

Article URL: https://huggingface.co/blog/codelion/ellora-lora-recipes

Comments URL: https://news.ycombinator.com/item?id=46130216

Points: 1

# Comments: 0

The 1B Token Challenge: Finding the Perfect Pre-Training Mix

codelion — Sat, 15 Nov 2025 15:07:35 +0000

Article URL: https://huggingface.co/blog/codelion/optimal-dataset-mixing

Comments URL: https://news.ycombinator.com/item?id=45937921

Points: 6

# Comments: 0

The 1B Token Challenge: Finding the Perfect Pre-Training Mix

codelion — Mon, 03 Nov 2025 07:42:34 +0000

Article URL: https://huggingface.co/blog/codelion/optimal-dataset-mixing

Comments URL: https://news.ycombinator.com/item?id=45796734

Points: 3

# Comments: 0

New comment by codelion in "Rats filmed snatching bats from air"

codelion — Sun, 02 Nov 2025 23:19:44 +0000

When mammals hunt other mammals strange things can happen.

New comment by codelion in "Don't Build Multi-Agents"

codelion — Mon, 01 Sep 2025 22:54:51 +0000

Whom to believe? Devin or Claude? - https://www.anthropic.com/engineering/multi-agent-research-s...

Pivotal Token Search (PTS): Targeting Critical Decision Points in LLM Training

codelion — Mon, 18 Aug 2025 04:33:43 +0000

Article URL: https://huggingface.co/blog/codelion/pts

Comments URL: https://news.ycombinator.com/item?id=44937442

Points: 2

# Comments: 1

New comment by codelion in "GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2"

codelion — Mon, 11 Aug 2025 06:43:07 +0000

It is by design. OpenAI is not going to reveal any architectural innovation they have made in their own commercial models.

Internal Coherence Maximization(ICM): Label-Free Unsupervised Training Framework

codelion — Sat, 09 Aug 2025 13:29:10 +0000

Article URL: https://github.com/codelion/icm

Comments URL: https://news.ycombinator.com/item?id=44846304

Points: 1

# Comments: 0

Unsupervised Model Improvement via Internal Coherence Maximization

codelion — Sun, 03 Aug 2025 13:53:15 +0000

Article URL: https://huggingface.co/blog/codelion/internal-coherence-maximization

Comments URL: https://news.ycombinator.com/item?id=44776598

Points: 1

# Comments: 0

Show HN: PTS Library – Analyze LLM reasoning through "thought anchors"

codelion — Wed, 23 Jul 2025 04:09:58 +0000

I built PTS (Pivotal Token Search), an open-source library for mechanistic interpretability analysis of language models. The core feature is generating "thought anchors" - identifying which specific sentences in a model's reasoning chain significantly impact task success.

What it does:

- Generates chain-of-thought reasoning traces from any LLM

- Uses counterfactual analysis to measure impact of each reasoning step

- Identifies critical sentences that make-or-break task completion

- Exports semantic embeddings for clustering analysis

- Provides systematic failure mode categorization

Example use case:

I used PTS to compare Qwen3-0.6B vs DeepSeek-R1-Distill-1.5B on math problems and discovered they have fundamentally different reasoning architectures:

- DeepSeek: concentrated reasoning (fewer, high-impact steps)

- Qwen3: distributed reasoning (impact spread across multiple steps)

Quick start:

# Generate thought anchors

pts run --model="your-model" --dataset="gsm8k" --generate-thought-anchors

# Export for analysis

pts export --format="thought_anchors" --output-path="analysis.jsonl"

The library implements the thought anchors methodology from Bogdan et al. (2025) with extensions for:

- Comprehensive metadata collection

- 384-dimensional semantic embeddings

- Causal dependency tracking

- Systematic failure analysis

Why this matters: Most interpretability tools focus on individual tokens or attention patterns. Thought anchors operate at the sentence level, revealing which complete reasoning steps actually matter for getting correct answers.

Limitations: Currently focused on mathematical reasoning tasks. Planning to extend to other domains and larger models.

Links:

- GitHub: https://github.com/codelion/pts

- Research example: https://huggingface.co/blog/codelion/understanding-model-rea...

- Generated datasets: Available on HuggingFace

Would appreciate feedback on extending this to other reasoning domains or interpretability approaches.

Comments URL: https://news.ycombinator.com/item?id=44655663

Points: 2

# Comments: 0

Automated Discovery of High-Performance GPU Kernels with OpenEvolve

codelion — Sat, 28 Jun 2025 01:03:46 +0000

Article URL: https://huggingface.co/blog/codelion/openevolve-gpu-kernel-discovery

Comments URL: https://news.ycombinator.com/item?id=44401609

Points: 4

# Comments: 0