SShortSingh.
Back to feed

TokenPrint Lets Developers Visualize LLM Inference Layer by Layer in 3D

0
·50 views

A developer has released TokenPrint, a free open-source tool that lets users visually explore how transformer-based large language models process tokens during inference. The interactive 3D environment allows users to navigate through each layer of a model and inspect individual operations such as attention scores, embeddings, normalization, and MLP projections. Rather than presenting the model as a static diagram or raw code, TokenPrint frames inference as a step-by-step computational journey through the network. Users can select any component to view plain-language explanations, mathematical equations, tensor dimensions, parameter counts, and actual model data. The project is available to try online at tokenprint.in and its source code is hosted publicly on GitHub.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

OpenAI Introduces GPT-6.2 Astra and Sol Models

OpenAI has announced two new models, GPT-6.2 Astra and GPT-6.2 Sol, through a post on its website. The announcement was shared on Hacker News, where it drew around 100 points and 23 comments. The available source material contains only the headline, the link to OpenAI's announcement, and the discussion metrics, so details about capabilities, pricing, and availability cannot be confirmed from this item. Readers seeking specifics are directed to OpenAI's original post for the full information.

0
ProgrammingDEV Community ·

CyberMoranda Advances Reverse Proxy Mode Ahead of Open-Source Release

Development of the Reverse Proxy mode for CyberMoranda's CIDS is progressing, with recent work focused on request handling, upstream forwarding, trusted proxy checks, and resource management. The developer is keeping the TARPIT and DECEPTION enforcement modes isolated so that heavy load on one does not impact the other. New tests have been added covering request limits, HTTP forwarding, concurrency, and resource recovery. Further testing, particularly with real traffic and under load, remains before the mode is released as open source. The team says it wants to confirm the foundation is reliable first.

0
ProgrammingDEV Community ·

Anima ships seven ComfyUI workflows for Wan-based pose-driven video generation

A release bundles seven workflow files built around the Anima pipeline, totaling 308 nodes across graphs that range from a 9-node text-to-skeletal-motion graph to a 116-node Wan 2.2 video-extension graph. The write-up analyzes the graphs directly, listing the models, VAEs, text encoders, and LoRAs each one loads. Several LoRAs run at non-default weights, such as 0.7, which the author treats as deliberate tuning choices. The workflows also use different latent geometries, including 928x1664 at one frame, and frame count is highlighted as more significant than pixel size for timing. The author cautions that node count does not indicate quality and that the files contain no benchmarks, so any quality claim requires the reader's own comparison. Model weights are not included and must be supplied separately.

0
ProgrammingDEV Community ·

Vetis Matcher API Offers Declarative HTTP Routing Rules in Rust

The Vetis Matcher API provides a fluent, declarative way to match HTTP traffic before a hook or handler runs, covering factors such as method, query values, headers, and protocol version. The approach is drawn from the project's integration tests and public API, which exercise the matchers against live TLS-enabled HTTP requests. Vetis separates server configuration from matching logic, so requests and responses can be evaluated independently of how the server itself is set up. Its public surface includes request matchers, response matchers, grouped matchers, and CEL-based expression matchers, which can be composed into more complex conditions. Examples include path matchers with wildcard support, header checks for values such as API keys, method filters, and query parameter matchers for tenant-specific routing.