You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

DeepSeek-V4-Flash KR L10 HF-FP4 Sidecar GGUF Set

Private DS4 bit-lift sidecar artifact set for Layer 10 Korean Think MAX experiments.

This repository contains three sidecar GGUF files. They are not standalone models. Use them with a DS4 runtime that supports --bitlift-sidecar.

Files

file expert coverage local size sha256
DeepSeek-V4-Flash-KR-ThinkTop64-L10-HF-FP4.sidecar.gguf 64 experts 864M 34527cc57a6294590687b16eb741ed374930e13f284310a7eb06dcc8bb3581b8
DeepSeek-V4-Flash-KR-ThinkTop128-L10-HF-FP4.sidecar.gguf 128 experts 1.7G f4e5c12b41c82d257b8cc63753c7bb83115d338e448bb6d96d12bdb6fdb49df9
DeepSeek-V4-Flash-KR-Full256-L10-HF-FP4.sidecar.gguf 256 experts 3.4G 9f4279fe27eb4da78f16ed872e444143a868825d87d0c0fc21e5d39a0d38246a

Runtime Contract

./ds4 \
  -m /path/to/ds4flash.gguf \
  --bitlift-sidecar ./DeepSeek-V4-Flash-KR-ThinkTop128-L10-HF-FP4.sidecar.gguf \
  --ctx 393216 \
  --think-max \
  -n 128 \
  -p 'ํ•œ๊ตญ์–ด๋กœ ์งง๊ฒŒ ๋‹ตํ•˜์„ธ์š”.'

Optional route tracing:

DS4_BITLIFT_TRACE_HITS=1 ./ds4 \
  -m /path/to/ds4flash.gguf \
  --bitlift-sidecar ./DeepSeek-V4-Flash-KR-ThinkTop128-L10-HF-FP4.sidecar.gguf \
  --ctx 393216 --think-max -n 128 -p 'ํ•œ๊ตญ์–ด๋กœ ์งง๊ฒŒ ๋‹ตํ•˜์„ธ์š”.'

Included Metadata

The artifacts/ directory includes:

  • source expert manifests
  • sidecar runtime plans
  • build summaries
  • evaluation summaries
  • local SHA256 checksums

Notes

These sidecars were generated from the locally cached HF FP4 source shard for Layer 10 and requantized into Q4_K sidecar tensors. The current practical recommendation remains ThinkTop128 for the Layer 10 experiment path, with ThinkTop64 as a smaller smoke artifact and Full256 as the coverage ceiling.

Downloads last month
5
GGUF
Model size
6B params
Architecture
deepseek4-bitlift-sidecar
Hardware compatibility
Log In to add your hardware

We're not able to determine the quantization variants.

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support