The Open-Source AI Stack
RSS
All models

Models · gemma

Gemma 2 27B IT

Source-available Google DeepMind · 2024-06-27 · Gemma Terms of Use

Distilled from a larger Gemini teacher, demonstrating knowledge-distillation as a path to mid-scale open-weights quality. Source-available license restricts some uses.

Cost

— / Mtok input
— / Mtok output

Together AI · as of 2026-05-19

via Artificial Analysis ↗

Speed

— tok/sec output

Together AI · as of 2026-05-19

via Artificial Analysis ↗

Architecture

tokens in Embedding vocab 256,000 · gemma tokenizer × N layers GQA + Sliding-window (interleaved) RoPE context 8,192 tokens Dense MLP SwiGLU activation (standard) 27.2B active params Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
dense
Total params
27.2B
Active params
27.2B
Context window
8K tokens
Attention
hybrid-gqa-sliding
Position encoding
rope
Pretraining tokens
13.0T
Training hardware
TPU v5p
Post-training
sft, rlhf, knowledge-distillation
OSI-approved
no
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

Recommended use cases

  • mid-tier deployment
  • distillation reference
  • Google-stack integration

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
AWQ Activation-aware 4-bit weight quantization for GPU serving. runs on vLLM, SGLang
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX
FP8 8-bit float, frequently a native release on Hopper / Blackwell GPUs. runs on vLLM, SGLang, TensorRT-LLM
bitsandbytes On-the-fly NF4 / INT8 weight quantization inside Transformers. runs on Transformers

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · Knowledge distillation from frontier teacher
  • · Sliding-window + full-attention interleave

Known limitations

  • · Gemma Terms is source-available, not OSI-approved. source ↗

Lineage

Distilled from a larger Gemini teacher; sibling line to the Gemini API.

Sources