Downloads · 30 days
0
dsv4flash-mtp-aligned
avlp12/dsv4flash-mtp-aligned
dsv4flash-mtp-aligned is a machine learning model from avlp12. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
On-policy chain-alignment of mtp.0's non-expert parameters (74.3M, bf16) for DeepSeek-V4-Flash-0731, trained teacher-forced on the model's own greedy generations to improve chained (depth-k) draft acceptance.
Access
Public
Updated Aug 24, 2026
Repo size
301 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors301 MB · 100%
At a glance
- License
- mit
- Access
- Public
- Created
- Aug 23, 2026
- Updated
- Aug 24, 2026
- SHA
- f2025622
Base models
Model card
From the Hugging Face model README
DSv4-Flash aligned MTP (non-expert sidecar)
On-policy chain-alignment of mtp.0's non-expert parameters (74.3M, bf16) for
DeepSeek-V4-Flash-0731, trained teacher-forced on the model's own greedy
generations to improve chained (depth-k) draft acceptance.
Measured (single M3 Ultra, 4-bit pack, fixed depth-3 chain):
- conditional acceptance d2 43.2% -> 51.1%, d3 9.8% -> 22.2%
- tokens/verify-cycle 2.25 -> 2.31
Load after the 4-bit pack by dequantizing mtp.0's non-expert linears to bf16
and applying these tensors (see train_align.py / serving sidecar in
github.com/avlp12/dsv4flash_tp2_stack). Speculative decoding is lossless, so
this changes speed only, never outputs.
- License
- mit
- Created
- Aug 23, 2026
- Updated
- Aug 24, 2026