Card: badges and links on one line; padded, rounded, 2x charts; vertical diagram

Browse files

Files changed (9) hide show

.gitattributes +2 -0
README.md +4 -20
assets/accuracy-by-subset.png +0 -0
assets/accuracy-per-parameter.png +0 -0
assets/compute.png +0 -0
assets/hero.png +0 -0
assets/how-it-works.png +0 -0
assets/latency.png +0 -0
assets/leaderboard.png +2 -2

.gitattributes CHANGED Viewed

@@ -35,3 +35,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 assets/leaderboard.png filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text

 *tfevents* filter=lfs diff=lfs merge=lfs -text
 assets/leaderboard.png filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+assets/accuracy-per-parameter.png filter=lfs diff=lfs merge=lfs -text
+assets/hero.png filter=lfs diff=lfs merge=lfs -text

README.md CHANGED Viewed

@@ -24,29 +24,13 @@ tags:
 lev answers typed questions about a piece of context in a single forward pass. You give it a **state** (text, a ticket, an email, or JSON) and a set of yes/no, choice, and score questions. It reads each answer from the logits it already computed and returns calibrated probabilities over exactly the options you supplied. It is a LoRA adapter on Qwen3.5-4B, and it speaks TypeSafe's `/v1/systemone` protocol, so code written for the TypeSafe SDK works against it once you change the base URL.
-<p align="center">
-  <img src="https://img.shields.io/badge/license-Apache--2.0-2a78d6?style=flat-square" alt="Apache-2.0">
-  <img src="https://img.shields.io/badge/base-Qwen3.5--4B-2a78d6?style=flat-square" alt="Qwen3.5-4B">
-  <img src="https://img.shields.io/badge/output%20tokens-0-2a78d6?style=flat-square" alt="Zero output tokens">
-  <img src="https://img.shields.io/badge/API-%2Fv1%2Fsystemone-2a78d6?style=flat-square" alt="/v1/systemone compatible">
-  <a href="https://github.com/Abhinavexists/lev"><img src="https://img.shields.io/badge/code-GitHub-14181f?style=flat-square&logo=github" alt="GitHub"></a>
-</p>
 <h2 align="center">72.5% on S1Bench. 69 ms of compute. 4B parameters.</h2>
-<p align="center"><strong>Qwen3.5-4B + LoRA · one H100 · six S1Bench subsets, 1,999 items.</strong><br>
-Sixth of 22 on the S1Bench board over these subsets, ahead of every model its size or smaller.<br>
-69 ms is compute inside the container. End to end from a laptop it is 414–463 ms.<br>
-<a href="#benchmarks">Benchmarks</a> · <a href="#speed">Speed</a> · <a href="#boundaries-worth-understanding">Boundaries</a></p>
-<p align="center">
-  <a href="#quickstart"><strong>Quickstart</strong></a> ·
-  <a href="#self-hosting-a-jev-compatible-http-server">Self-hosting</a> ·
-  <a href="#benchmarks">Benchmarks</a> ·
-  <a href="#why-it-works">Why it works</a> ·
-  <a href="#training">Training</a> ·
-  <a href="https://github.com/Abhinavexists/lev">GitHub</a>
-</p>
 **No generated tokens. No JSON to parse. No retries until it validates.**
@@ -205,7 +189,7 @@ Most of lev's round trip from a laptop is network and Modal's ingress; its compu
 ## Why it works
-<p align="center"><img src="assets/how-it-works.png" alt="State and questions go through one forward pass, then label-token readout or the candidate-path head, then per-bucket temperatures, then typed answers" width="100%"></p>
 - **Label-token readout.** Each option gets a short code, and the answer is read from the next-token logits over those codes. Codes that would split into two tokens are skipped, so every option stays one token. This handles yes/no, scores, and choice lists up to several hundred options. Choices are read in two option orders and averaged, which cancels position bias.
 - **Candidate-path head.** Past that point, a small learned head matches the state against each option's text, so the number of options has no fixed ceiling.

 lev answers typed questions about a piece of context in a single forward pass. You give it a **state** (text, a ticket, an email, or JSON) and a set of yes/no, choice, and score questions. It reads each answer from the logits it already computed and returns calibrated probabilities over exactly the options you supplied. It is a LoRA adapter on Qwen3.5-4B, and it speaks TypeSafe's `/v1/systemone` protocol, so code written for the TypeSafe SDK works against it once you change the base URL.
+<p align="center"><img src="https://img.shields.io/badge/license-Apache--2.0-2a78d6?style=flat-square" alt="Apache-2.0"> <img src="https://img.shields.io/badge/base-Qwen3.5--4B-2a78d6?style=flat-square" alt="Qwen3.5-4B"> <img src="https://img.shields.io/badge/output%20tokens-0-2a78d6?style=flat-square" alt="Zero output tokens"> <img src="https://img.shields.io/badge/API-%2Fv1%2Fsystemone-2a78d6?style=flat-square" alt="/v1/systemone compatible"> <a href="https://github.com/Abhinavexists/lev"><img src="https://img.shields.io/badge/code-GitHub-14181f?style=flat-square&logo=github" alt="GitHub"></a></p>
 <h2 align="center">72.5% on S1Bench. 69 ms of compute. 4B parameters.</h2>
+<p align="center"><strong>Qwen3.5-4B + LoRA · one H100 · six S1Bench subsets, 1,999 items.</strong><br>Sixth of 22 on the S1Bench board over these subsets, ahead of every model its size or smaller.<br>69 ms is compute inside the container. End to end from a laptop it is 414–463 ms.<br><a href="#benchmarks">Benchmarks</a> · <a href="#speed">Speed</a> · <a href="#boundaries-worth-understanding">Boundaries</a></p>
+<p align="center"><a href="#quickstart"><strong>Quickstart</strong></a> · <a href="#self-hosting-a-jev-compatible-http-server">Self-hosting</a> · <a href="#benchmarks">Benchmarks</a> · <a href="#why-it-works">Why it works</a> · <a href="#training">Training</a> · <a href="https://github.com/Abhinavexists/lev">GitHub</a></p>
 **No generated tokens. No JSON to parse. No retries until it validates.**
 ## Why it works
+<p align="center"><img src="assets/how-it-works.png" alt="State and questions go through one forward pass, then label-token readout or the candidate-path head, then per-bucket temperatures, then typed answers" width="460"></p>
 - **Label-token readout.** Each option gets a short code, and the answer is read from the next-token logits over those codes. Codes that would split into two tokens are skipped, so every option stays one token. This handles yes/no, scores, and choice lists up to several hundred options. Choices are read in two option orders and averaged, which cancels position bias.
 - **Candidate-path head.** Past that point, a small learned head matches the state against each option's text, so the number of options has no fixed ceiling.