Falcon3-7B-Instruct

Llama family · 7.0B params

1recorded run

19.9best tok/s

0verified

Evidence: 0 sourced runs · 1 awaiting review · 1 independent source · first recorded Aug 30, 2026. What these labels mean.

What the Falcon3-7B-Instruct numbers mean

The fastest result on our board is 19.9 output tok/s on an Tenstorrent P150 at bfp_bf8 with a 2048-token context. That is about 50.3 ms per generated token — 500 tokens takes roughly 25.1 seconds. Prompt processing on the same record runs at 1821.4 tok/s, so a 2,000-token prompt is read in about 1.1 seconds.

How much VRAM does Falcon3-7B-Instruct need?

Falcon3-7B-Instruct has 7.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	5.3 GB	8 GB
Q8_0 (8-bit)	9.4 GB	12 GB
FP16 (16-bit)	16.9 GB	24 GB

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

Tenstorrent P15019.9

Models with a similar footprint

Model	Params	Q4_K_M weights
AgentTest-7B	7B	3.9 GB
AceReason-Nemotron-1.1-7B	8B	4.5 GB
DeepSeek-R1-0528-Qwen3-8B	8B	4.5 GB
DeepSeek-R1-Distill-Qwen-7B	8B	4.5 GB
gemma-4-E4B-it	8B	4.5 GB
granite-4.2-8b	8B	4.5 GB

Ranked by distance from Falcon3-7B-Instruct's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Tenstorrent P150 · 32 GB	Falcon3-7B-Instruct	bfp_bf8	vllm —	2048	19.9 …	1821.4	—	src