Qwen A/B Evaluation¶
scripts/qwen_ab_eval.py compares a HuggingFace baseline against a TurboQuant
diagnostic path for the same model and scenarios.
Run¶
python scripts/qwen_ab_eval.py \
--model Qwen/Qwen2.5-3B \
--output-root reports/qwen_ab \
--dtype bfloat16
Use local model files when needed:
Recorded Fields¶
The script records:
- scenario name,
- prompt and input length,
- generation settings,
- raw generated text,
- prompt/decode/total timing,
- tokens per second,
- CUDA peak allocation and reservation when CUDA is available,
- CPU and RAM samples when
psutilis available, - dense KV byte estimate,
- compressed byte estimate for the TurboQuant adapter,
- output comparison notes,
- artifact manifests.
Artifact Isolation¶
Runs use a run id and separate directories for:
- baseline outputs,
- TurboQuant outputs,
- shared scenario manifest,
- metrics JSON,
- raw JSONL outputs.
The script is intended for controlled A/B checks, not for production serving claims.