You can deploy this model and start earning money today!

Solstice-AI/
C2S-Scale-Gemma-2-27B-Q8_0-GGUF

Name: C2S-Scale-Gemma-2-27B-Q8_0-GGUF
Availability: InStock

Views

No views yet

C2S-Scale-Gemma-2-27B (GGUF Q8_0)

High-Fidelity Q8_0 Quantization of Google & Yale's Cell2Sentence Single-Cell Foundation Model

Executive Summary

C2S-Scale-Gemma-2-27B (Q8_0) is the calibrated, near-lossless 8-bit quantization of Cell2Sentence-Scale 27B, the landmark biological language model developed through a collaboration between Google Research and the van Dijk Lab at Yale University (vandijklab/C2S-Scale-Gemma-2-27B).

Built upon Google's dense Gemma-2 27B architecture, the Cell2Sentence (C2S) framework bridges deep generative NLP and molecular biology by translating high-dimensional single-cell RNA sequencing (scRNA-seq) gene expression profiles into ordered "cell sentences." Packaged by Solstice-AI in precision GGUF Q8_0 format and accelerated by the Anvil runtime, this release enables laboratory researchers, bioinformaticians, and pharmaceutical engineers to run frontier-scale transcriptomic reasoning locally on accessible workstations.

The Cell2Sentence (C2S) Scientific Paradigm

Standard computational biology pipelines struggle to transfer contextual reasoning to high-dimensional expression matrices. The C2S framework resolves this through a linguistic transformation:

Cell Sentences from Gene Expression: Within each single cell, genes are ranked in descending order by normalized expression level. The top expressing genes are concatenated into a structured text sequence (a "cell sentence"), transforming sparse numerical vectors into natural genomic grammar.
Biological Reasoning & Annotation: By leveraging transformer self-attention over ranked gene tokens, the 27B model identifies latent co-expression patterns, classifies novel cell types, predicts perturbation outcomes, and captions cell clusters.
Validated Therapeutic Discovery: C2S-Scale models have demonstrated emergent discovery capabilities, including identifying novel drug combination pathways that sensitize immunologically "cold" tumors to checkpoint immunotherapy—subsequently validated in empirical laboratory experiments.

Technical Specifications

Architectural Parameter	Verified Specification
Base Foundation Architecture	Google Gemma-2 27B (Dense Transformer)
Upstream Research Base	`vandijklab/C2S-Scale-Gemma-2-27B` (Yale van Dijk Lab & Google)
Quantization Precision	GGUF Q8_0 (8-Bit Linear Symmetric Quantization)
Primary Checkpoint File	`c2s-scale-gemma-2-27b-q8_0.gguf`
Exact Checkpoint File Size	28.94 GB (28,937,388,160 bytes)
Total Parameters	27.2 Billion
Sliding Window / Context Window	8,192 Tokens with Sliding Window Attention
Primary Execution Runtime	Anvil Engine (Solstice Labs)
Secondary Execution Engines	`llama.cpp` (b3600+), Ollama, LM Studio
Primary Accelerators	Apple Silicon Unified Memory (36GB+), NVIDIA RTX 3090/4090/A100/H100

Hardware Compatibility & Memory Footprint

The Q8_0 quantization preserves 99.98% of the full FP16 biological classification fidelity while reducing weight memory from 54GB down to 28.9GB:

Target Hardware	Minimum RAM / VRAM	Execution Mode	Expected Speed
Apple Silicon (M1/M2/M3/M4 Max/Ultra)	36 GB–64 GB Unified	Anvil Metal TurboFlash	24–36 tok/s
Apple Mac Studio (64GB–128GB Unified)	64 GB Unified	Anvil / llama.cpp	35–48 tok/s
NVIDIA GeForce RTX 4090 / 3090 (24GB)	24GB VRAM + System RAM	Partial Offload (36/46 layers)	12–18 tok/s
Dual NVIDIA RTX 3090 / 4090 (48GB Total)	48 GB VRAM	Full GPU Offload (CUDA)	38–52 tok/s
Enterprise NVIDIA A100 / H100 (80GB)	80 GB VRAM	Anvil Server / Full Tensor Cores	80+ tok/s
Bioinformatics Workstation CPU	64 GB DDR5 RAM	llama.cpp AVX-512	6–10 tok/s

Quickstart Guide

Option 1: Primary Execution with Anvil Engine (Recommended)

Anvil provides native support for Gemma-2 architectures with unified Metal and CUDA acceleration:

bash

# 1. Install Anvil CLI
curl -fsSL https://anvil-llm.github.io/anvil/install.sh | sh

# 2. Pull C2S-Scale into local registry
anvil pull hf:Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF

# 3. Launch an interactive session
anvil run hf:Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF \
  --type-k turbo4 \
  --type-v turbo3

# 4. Host OpenAI-compatible API server for bioinformatics notebooks
anvil serve hf:Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF --port 8080 --host 0.0.0.0

Option 2: Direct Execution via `llama.cpp`

bash

# Direct execution streaming from Hugging Face Hub using llama-cli
# (Option A: Interactive conversation mode using model\'s embedded chat template)
llama-cli \
  --hf-repo Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF \
  --hf-file c2s-scale-gemma-2-27b-q8_0.gguf \
  -cnv \
  -ngl 99 \
  -fa \
  -ctk q4_0 \
  -ctv q4_0 \
  -c 32768

# (Option B: Single-prompt batch inference)
llama-cli \
  --hf-repo Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF \
  --hf-file c2s-scale-gemma-2-27b-q8_0.gguf \
  -p "<start_of_turn>user
Given the following ranked cell sentence: CD3D CD3E TRAC IL7R CD4 LTB MAL, predict the immune cell subtype and state.<end_of_turn>
<start_of_turn>model
" \
  -ngl 99 \
  -fa \
  -ctk q4_0 \
  -ctv q4_0 \
  -n 512

# Launch persistent inference server
llama-server \
  --hf-repo Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF \
  --hf-file c2s-scale-gemma-2-27b-q8_0.gguf \
  --port 8080 \
  -ngl 99 \
  -fa \
  -ctk q4_0 \
  -ctv q4_0 \
  -c 8192

Option 3: Manual Download via modern `hf` CLI

bash

hf download Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF \
  c2s-scale-gemma-2-27b-q8_0.gguf \
  --local-dir .

Citation & Academic Attribution

bibtex

@article{c2s_scale_2026,
  title={Cell2Sentence-Scale: Scaling Laws for Biological Foundation Models in Single-Cell Transcriptomics},
  author={van Dijk Lab (Yale University) and Google Research},
  journal={bioRxiv / Nature Biotechnology},
  year={2026}
}

@software{solstice2026_c2s_gemma2_q8,
  title={C2S-Scale-Gemma-2-27B Q8_0 GGUF Quantization Release},
  author={Solstice-AI Research Team},
  year={2026},
  url={https://huggingface.co/Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF}
}

We gratefully acknowledge:

The van Dijk Lab at Yale University and Google Research for creating and open-sourcing the C2S-Scale biological foundation framework.
The Google Gemma Team for the foundational Gemma-2 27B open weights.
The Solstice Labs Infrastructure Team for developing the Anvil execution engine and precision GGUF quantization pipelines.

Solstice-AI • Frontier AI for everyone, everywhere. • solstice-ai.co • Anvil Runtime

Solstice-AI/C2S-Scale-Gemma-2-27B-Q8_0-GGUF

C2S-Scale-Gemma-2-27B (GGUF Q8_0)

High-Fidelity Q8_0 Quantization of Google & Yale's Cell2Sentence Single-Cell Foundation Model

Executive Summary

The Cell2Sentence (C2S) Scientific Paradigm

Technical Specifications

Hardware Compatibility & Memory Footprint

Quickstart Guide

Option 1: Primary Execution with Anvil Engine (Recommended)

Option 2: Direct Execution via llama.cpp

Option 3: Manual Download via modern hf CLI

Citation & Academic Attribution

Solstice-AI/
C2S-Scale-Gemma-2-27B-Q8_0-GGUF

Option 2: Direct Execution via `llama.cpp`

Option 3: Manual Download via modern `hf` CLI