Downloads · 30 days
0
Siddharth85/kvbridge-qwen3-1.7b-to-qwen3-4b
kvbridge-qwen3-1.7b-to-qwen3-4b is a machine learning model from Siddharth85. 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 kvbridge. The card lists the license as mit.
A fitted KV cache mapper: it converts a Qwen3-1.7B KV cache into one Qwen3-4B can generate from, so the prompt is prefilled once on the small model instead of on the big one.
Downloads · 30 days
0
Access
Public
Updated Aug 11, 2026
Repo size
1.8 GB
Likes
1
Public
Click a slice to open those files.
.safetensors1.8 GB · 100%
From the Hugging Face model README
A fitted KV cache mapper: it converts a Qwen3-1.7B KV cache into one
Qwen3-4B can generate from, so the prompt is prefilled once on the
small model instead of on the big one.
from kvbridge import Mapper, transfer
mapper = Mapper.from_pretrained("Siddharth85/kvbridge-qwen3-1.7b-to-qwen3-4b")
small_cache = small(prompt_ids, use_cache=True).past_key_values
big_cache = transfer(small_cache, mapper)
big.generate(input_ids=prompt_ids, past_key_values=big_cache, max_new_tokens=128)
Code, results and fitting instructions: https://github.com/Sidd03192/kvbridge
Fit on 500 FineWeb-Edu sequences x 1024 tokens, stride-4 subsampled, ridge lambda=0.01. Mean in-sample R^2: K 0.813, V 0.718. Weights are bfloat16.
Method: Cross-Model KV Cache Transfer in LLM Families (Heo et al., NVIDIA, arXiv:2608.03893).