Benchmark
Submitted bylukas
Public report
This is the public version of the benchmark. It includes model, runtime, protocol, and throughput information, without telemetry, raw samples, or signing metadata.
Compare throughput at the same token counts. A signed report does not independently verify benchmark execution.
Model
- Canonical ID
- google/gemma-4-12B-it-qat-q4_0-unquantized
- Identity resolution
- artifact_verified
- Identity verification
- verified
- Verified at
- Name
- Gemma-4 12B IT (smart Q4_0, QAT-lossless)
- Architecture
- gemma4
- Quantization
- Q4_0
- Artifact
- Provider
- huggingface
- Repo ID
- unsloth/gemma-4-12B-it-qat-GGUF
- Revision
- 980b060c40a8539ac159e0501a3e0f66a6365af3
- Path
- gemma-4-12B-it-qat-UD-Q4_K_XL.gguf
- SHA256
- 90fd44e29e0d7cffeb0fd00dc73cfdab9ed0b0e95306ecf7821ea634c940c370
Runtime
- Name
- llama-cpp
- Version
- b10901 (28ff09582)
Benchmark
- Mode
- text
- Chip
- Apple M5 Max
- Backend
- MTL,BLAS
- Protocol
- Report schema
- computearena-benchmark/1
- Benchmark schema
- computearena-measurements/1
- Throughput schema
- llama-bench-independent-pp-tg/1
- Decode initial context tokens
- 0
- Conditioning schema
- computearena-conditioning/1
- Conditioning mode
- runtime_native_warmup
Report metadata
- Schema
- computearena-public-benchmark/1
- Submission ID
- 0d00cd50-feec-4314-9bf5-c7e54f50d030
- Run ID
- 29a1c71ef64a38a717ccb8bc0987fcb6
- Benchmark date
- (Unix ms: 1789083516694)
- Submitted by
- lukas