Update README.md
Browse files
README.md
CHANGED
|
@@ -22,6 +22,9 @@ This repository provides **FS-DFM checkpoints** from the paper:
|
|
| 22 |
Amin Karimi Monsefi, Nikhil Bhendawade, Manuel R. Ciosici, Dominic Culver, Yizhe Zhang, Irina Belousova (Jan 9, 2026)
|
| 23 |
ArXiv: 2509.20624
|
| 24 |
|
|
|
|
|
|
|
|
|
|
| 25 |
FS-DFM is a **token-space diffusion / flow-matching language model** designed for **fast long-text generation** by explicitly training for a **user-specified step budget** (e.g., 1–8 steps), while preserving a CTMC-based discrete flow formulation.
|
| 26 |
|
| 27 |
## What’s in this repo
|
|
@@ -42,6 +45,15 @@ FS-DFM is a **token-space diffusion / flow-matching language model** designed fo
|
|
| 42 |
|
| 43 |
**Formulation:** discrete flow-matching over a **CTMC** on token sequences; sampling uses custom solvers (e.g., `mixture_euler_with_cumulative_scalar`).
|
| 44 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
---
|
| 46 |
|
| 47 |
## Architecture
|
|
|
|
| 22 |
Amin Karimi Monsefi, Nikhil Bhendawade, Manuel R. Ciosici, Dominic Culver, Yizhe Zhang, Irina Belousova (Jan 9, 2026)
|
| 23 |
ArXiv: 2509.20624
|
| 24 |
|
| 25 |
+
[Github Link](https://github.com/apple/ml-fs-dfm/tree/main)
|
| 26 |
+
[Paper Link](https://arxiv.org/abs/2509.20624)
|
| 27 |
+
|
| 28 |
FS-DFM is a **token-space diffusion / flow-matching language model** designed for **fast long-text generation** by explicitly training for a **user-specified step budget** (e.g., 1–8 steps), while preserving a CTMC-based discrete flow formulation.
|
| 29 |
|
| 30 |
## What’s in this repo
|
|
|
|
| 45 |
|
| 46 |
**Formulation:** discrete flow-matching over a **CTMC** on token sequences; sampling uses custom solvers (e.g., `mixture_euler_with_cumulative_scalar`).
|
| 47 |
|
| 48 |
+
|
| 49 |
+
## Comparison of Methods
|
| 50 |
+
|
| 51 |
+
| ARM | DFM | FS-DFM (Ours) |
|
| 52 |
+
|-----|-----|---------------|
|
| 53 |
+
|  |  |  |
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
|
| 57 |
---
|
| 58 |
|
| 59 |
## Architecture
|