Hailo-10H Pi AI HAT+ 2 Llama-3.2-1B compiled-model TPS
9.9 tok/s (Hailo compiled model, vendor TPS table)
Fields
id
benchmarks/comm-hailo-10h-llama32-1b-compiled-tpsMetric
decode_tok_s
Value
9.9
Unit
tok/s (Hailo compiled model, vendor TPS table)
Model
Llama-3.2-1B
Methodology
Hailo own compiled-model TPS table in the Hailo Model Zoo GenAI models doc (genai/models.rst), measured on the Hailo-10H as shipped in the Raspberry Pi AI HAT+ 2. Vendor-compiled model number, not a llama.cpp/GGUF run: no open GGUF path, and the compiled set tops out at <=2B (same table: Qwen2.5-1.5B 7.4 tok/s, Qwen3-1.7B ~4.8 tok/s). Prompt/output lengths and batch size are not stated, so this is a vendor headline TPS rather than a reproducible llama-bench cell. vendor_claim: true.
Measured by
vendor
Reproducible
no
Thermal state
unknown
notes
Previously recorded only as prose inside gotcha-hailo-pi-ai-hat-2-capped-at-2b-models. Filed as its own benchmark so the 9.9 tok/s figure is citable. Vendor claim: the table does not disclose ISL/OSL or concurrency.
Notes
Previously recorded only as prose inside gotcha-hailo-pi-ai-hat-2-capped-at-2b-models. Filed as its own benchmark so the 9.9 tok/s figure is citable. Vendor claim: the table does not disclose ISL/OSL or concurrency.