Spaces:
Running
Running
Divyansh Agrawal commited on
Update README.md
Browse files
README.md
CHANGED
|
@@ -15,8 +15,6 @@ Our work starts with **SpectralQuant**: a calibration-aware quantization approac
|
|
| 15 |
|
| 16 |
Our goal is simple: **make smaller models feel less compressed.**
|
| 17 |
|
| 18 |
-
---
|
| 19 |
-
|
| 20 |
## What We Believe
|
| 21 |
|
| 22 |
Modern language models are becoming more capable, but access is increasingly shaped by memory, bandwidth, hardware, and deployment cost. Quantization is one of the most important bridges between frontier capability and everyday usability.
|
|
@@ -29,8 +27,6 @@ We believe the next generation of model compression should be:
|
|
| 29 |
* **Research-transparent**, with clear claims, clear limits, and reproducible comparisons.
|
| 30 |
* **User-practical**, focused on models that people can download, run, inspect, and build with.
|
| 31 |
|
| 32 |
-
---
|
| 33 |
-
|
| 34 |
## The SpectralQuant Method
|
| 35 |
|
| 36 |
Quantization is usually framed as a strict tradeoff: smaller files cost quality, and higher-quality quants cost more bytes. SpectralQuant explores whether we can bend that curve by improving what happens *inside* the same footprint.
|
|
@@ -39,8 +35,6 @@ Instead of treating compression as a purely local rounding problem, SpectralQuan
|
|
| 39 |
|
| 40 |
The result is a compact `Q4`-style artifact designed to preserve model behavior where it matters most, without breaking standard `llama.cpp` compatibility.
|
| 41 |
|
| 42 |
-
---
|
| 43 |
-
|
| 44 |
## Evaluation Philosophy
|
| 45 |
|
| 46 |
We do not treat file size and model quality as separate claims. A smaller model only matters if the quality story is visible, testable, and honest.
|
|
@@ -54,8 +48,6 @@ Spectral Labs releases aim to include:
|
|
| 54 |
|
| 55 |
The goal is not to declare universal wins from one benchmark, but to publish compact models with evidence that researchers, builders, and local-inference users can inspect.
|
| 56 |
|
| 57 |
-
---
|
| 58 |
-
|
| 59 |
## Featured Models
|
| 60 |
|
| 61 |
Spectral Labs models are optimized for local inference, edge/laptop deployment, fast prototyping, and research into behavior-preserving compression.
|
|
@@ -64,8 +56,6 @@ Spectral Labs models are optimized for local inference, edge/laptop deployment,
|
|
| 64 |
| :--- | :---: | :---: | :--- |
|
| 65 |
| **Qwen3.5 0.8B SpectralQuant** | GGUF (`Q4_K_M`) | 4.52 BPW | First SpectralQuant release candidate. Built with calibration-aware error shaping. |
|
| 66 |
|
| 67 |
-
---
|
| 68 |
-
|
| 69 |
## Current Focus & Roadmap
|
| 70 |
|
| 71 |
We see Spectral Labs as a home for practical model-compression research: methods that are technically interesting, but also ship as usable artifacts. We are currently executing on:
|
|
|
|
| 15 |
|
| 16 |
Our goal is simple: **make smaller models feel less compressed.**
|
| 17 |
|
|
|
|
|
|
|
| 18 |
## What We Believe
|
| 19 |
|
| 20 |
Modern language models are becoming more capable, but access is increasingly shaped by memory, bandwidth, hardware, and deployment cost. Quantization is one of the most important bridges between frontier capability and everyday usability.
|
|
|
|
| 27 |
* **Research-transparent**, with clear claims, clear limits, and reproducible comparisons.
|
| 28 |
* **User-practical**, focused on models that people can download, run, inspect, and build with.
|
| 29 |
|
|
|
|
|
|
|
| 30 |
## The SpectralQuant Method
|
| 31 |
|
| 32 |
Quantization is usually framed as a strict tradeoff: smaller files cost quality, and higher-quality quants cost more bytes. SpectralQuant explores whether we can bend that curve by improving what happens *inside* the same footprint.
|
|
|
|
| 35 |
|
| 36 |
The result is a compact `Q4`-style artifact designed to preserve model behavior where it matters most, without breaking standard `llama.cpp` compatibility.
|
| 37 |
|
|
|
|
|
|
|
| 38 |
## Evaluation Philosophy
|
| 39 |
|
| 40 |
We do not treat file size and model quality as separate claims. A smaller model only matters if the quality story is visible, testable, and honest.
|
|
|
|
| 48 |
|
| 49 |
The goal is not to declare universal wins from one benchmark, but to publish compact models with evidence that researchers, builders, and local-inference users can inspect.
|
| 50 |
|
|
|
|
|
|
|
| 51 |
## Featured Models
|
| 52 |
|
| 53 |
Spectral Labs models are optimized for local inference, edge/laptop deployment, fast prototyping, and research into behavior-preserving compression.
|
|
|
|
| 56 |
| :--- | :---: | :---: | :--- |
|
| 57 |
| **Qwen3.5 0.8B SpectralQuant** | GGUF (`Q4_K_M`) | 4.52 BPW | First SpectralQuant release candidate. Built with calibration-aware error shaping. |
|
| 58 |
|
|
|
|
|
|
|
| 59 |
## Current Focus & Roadmap
|
| 60 |
|
| 61 |
We see Spectral Labs as a home for practical model-compression research: methods that are technically interesting, but also ship as usable artifacts. We are currently executing on:
|