Carnice-V3-27b โ€” Pollard

Pollard shrank this model: 54.72 GB (f16) โ†’ 12.08 GB โ€” 78% smaller, 4.5ร— down.

The smallest rung here; larger, higher-fidelity rungs are listed below.

format this model's size
f16 54.72 GB
Q8_0 ~29.00 GB
Q6_K ~22.44 GB
Q4_K_M ~15.87 GB
PollardMix (this repo's IQ3_S) 12.08 GB

Pollard builds of kai-os/Carnice-V3 made with Pollard Weights โ€” a ladder of measured-allocation quants (bits placed by per-layer sensitivity, not a uniform crush).

Standard GGUF โ€” runs in stock llama.cpp / ik_llama.cpp, Ollama, LM Studio. Trellis (IQ*_KT) files need ik_llama.cpp; the K-quants run anywhere.

Model details

Parameter count ~27.4B
Architecture qwen3_5
Input support text
imatrix no
Perplexity measured yes โ€” table below

Which file should I choose?

Every rung is the same weights, sized to a different RAM budget by the measured allocation. Pick the largest one that fits your machine with room for context:

  • ~22 GB RAM / VRAM โ†’ Q6_K (19.63 GB).
  • ~18 GB RAM / VRAM โ†’ IQ4_XS (15.72 GB).
  • ~14 GB RAM / VRAM โ†’ IQ3_S (12.08 GB).

Available files

file PPL size tok/s Mean KLD notes
Carnice-V3-27b-Pollard-IQ3_S.gguf 7.88 12.08 GB ~6.9 โ€” IQ3_S
Carnice-V3-27b-Pollard-IQ4_XS.gguf โ€” 15.72 GB 24 GB tier โ€” IQ4_XS
Carnice-V3-27b-Pollard-Q6_K.gguf โ€” 19.63 GB 32 GB tier โ€” Q6_K

tok/s is hardware-specific; the machine it was measured on is stated in the errata.

Multimodal

Vision needs the projector shipped alongside: mmproj-Carnice-V3-BF16.gguf โ€” download it too and pass it with --mmproj. It is not quantized; it is small and the text ladder is where the size lives.

llama-server -m Carnice-V3-27b-Pollard-IQ3_S.gguf --mmproj mmproj-Carnice-V3-BF16.gguf -ngl 99

Download a specific file

pip install -U "huggingface_hub[cli]"
hf download PollardWeights/Carnice-V3-27b-Pollard \
  --include "Carnice-V3-27b-Pollard-IQ3_S.gguf" --local-dir ./

How to run

These are standard GGUF and run with llama.cpp:

llama-server -hf PollardWeights/Carnice-V3-27b-Pollard:IQ3_S

or from a local file:

llama-cli    -m Carnice-V3-27b-Pollard-IQ3_S.gguf -ngl 99 -p "Explain why the sky is blue."
llama-server -m Carnice-V3-27b-Pollard-IQ3_S.gguf -ngl 99      # OpenAI-compatible API + web UI at :8080

They also work in anything built on llama.cpp โ€” LM Studio, koboldcpp, Jan, ramalama, Ollama (ollama run hf.co/PollardWeights/Carnice-V3-27b-Pollard).

ARM / AVX

llama.cpp repacks weights into an interleaved layout at load time for faster inference on ARM and AVX machines โ€” no special file needed, online repacking covers these quants. The old Q4_0_4_4/4_8/8_8 variants are not required.

Errata

  • Trellis (IQ*_KT) quants need ik_llama.cpp to build/run; K-quants run in any recent llama.cpp.
  • Measured allocation places bits by per-layer sensitivity under a size budget.
  • Single machine; replication invited.

Credits & license

  • Base model: kai-os/Carnice-V3
  • Quantization tooling: llama.cpp (ggml-org)
  • Method + tooling: Pollard Weights โ€” measure first, no claim before a number.
  • License: apache-2.0, inherited from the base model.

Built with Pollard Weights โ€” frontier models, small hardware, no compromise.

Downloads last month
560
GGUF
Model size
27B params
Architecture
qwen35
Hardware compatibility
Log In to add your hardware

3-bit

4-bit

6-bit

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

Model tree for PollardWeights/Carnice-V3-27b-Pollard

Base model

Qwen/Qwen3.8-27B
Quantized
(5)
this model