Downloads · 30 days
31
6% of all-time downloads
LFM2.5-8B-A1B-DSpark-3L
tugot17/LFM2.5-8B-A1B-DSpark-3L
LFM2.5-8B-A1B-DSpark-3L is a machine learning model from tugot17. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for sglang. The card lists the license as other.
DSpark speculative-decoding draft model for the MoE target LiquidAI/LFM2.5-8B-A1B: a Qwen3-style GQA block drafter (3 layers) with a low-rank Markov transition head (rank 256) and a confidence head. Block size 7 (anch…
All-time downloads
531
Public
Parameters
231M
462 MB on disk
Likes
2
Public
Repo makeup
Click a slice to open those files.
.safetensors462 MB · 100%
At a glance
- Library
- sglang
- License
- other
- Model type
- qwen3
- Access
- Public
- Created
- Jul 10, 2026
- Updated
- Jul 10, 2026
- SHA
- 7621e184
Base models
Model card
From the Hugging Face model README
LFM2.5-8B-A1B DSpark draft (3-layer)
DSpark speculative-decoding draft model for the MoE target LiquidAI/LFM2.5-8B-A1B: a Qwen3-style GQA block drafter (3 layers) with a low-rank Markov transition head (rank 256) and a confidence head. Block size 7 (anchor + 6 drafts), interleaved (GPT-J) RoPE.
The draft ships no vocab weights; the embedding and LM head are tied from the target at load time.
Provided to exercise LFM2-MoE DSpark support in SGLang.
Usage (SGLang)
python -m sglang.launch_server \
--model-path LiquidAI/LFM2.5-8B-A1B \
--speculative-algorithm DSPARK \
--speculative-draft-model-path tugot17/LFM2.5-8B-A1B-DSpark-3L \
--speculative-draft-attention-backend flashinfer
Requires an SGLang build with DSpark and LFM2 support.
- Library
- sglang
- Type
- qwen3
- License
- other
- Created
- Jul 10, 2026
- Updated
- Jul 10, 2026