Skip to content
Updated minutes ago
NVIDIA

Florence 2 Large

Microsoft · dense · 0.77B parameters · 2,048 context

Quality
50.0

Parameters

0.77B

Context Window

2K tokens

Architecture

Dense

Best GPU

RTX 4060

Intelligence Brief

Florence 2 Large is a 0.77B parameter DENSE model from Microsoft, featuring Multi-Head Attention (MHA) with 24 layers and 1,024 hidden dimensions. With a 2,048 token context window, it supports vision, structured output. For self-hosted inference, RTX 4060 delivers optimal throughput at $209/month.

Architecture Details

TypeDENSE
Total Parameters0.77B
Active Parameters0.77B
Layers24
Hidden Dimension1,024
Attention Heads16
KV Heads16
Head Dimension64
Vocab Size51,289

Memory Requirements

BF16 Weights

1.5 GB

FP8 Weights

0.8 GB

INT4 Weights

0.4 GB

KV-Cache per Token98304 bytes
Activation Estimate0.20 GB

GPU Compatibility Matrix

Florence 2 Large is compatible with 100% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
B300288GB
B100 SXM192GB
GB200 NVL72 (per GPU)192GB
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
H200 SXM141GB
H2096GB
GH20096GB
Ada Lovelace(11 GPUs)
L40S48GB
L4048GB
RTX 6000 Ada48GB
L2048GB
Ampere(16 GPUs)
A100 80GB SXM80GB
A100 80GB PCIe80GB
A1664GB
RTX A600048GB
Legend:No fitVery tightTightModerateGoodExcellent

GPU Recommendations

RTX 4060optimal

BF16 · 1 GPU · vllm

90/100

score

Throughput

906.0 tok/s

Latency (ITL)

1.1ms

Est. TTFT

0ms

Cost/Month

$209

Cost/M Tokens

$0.09

Use this config →
RTX 3070optimal

BF16 · 1 GPU · vllm

90/100

score

Throughput

1.5K tok/s

Latency (ITL)

0.7ms

Est. TTFT

0ms

Cost/Month

$85

Cost/M Tokens

$0.02

Use this config →
B200 SXMoptimal

FP8 · 1 GPU · tensorrt-llm

83/100

score

Throughput

3.5K tok/s

Latency (ITL)

0.3ms

Est. TTFT

0ms

Cost/Month

$4261

Cost/M Tokens

$0.46

Use this config →

Deployment Options

API

API Deployment

No API pricing available

Self-Hosted

Single GPU

RTX 4060

$209/mo

Min VRAM: 1 GB

Scale

Multi-GPU

RTX 4060

906.0 tok/s

Best available config

API Pricing Comparison

No API pricing data available for this model.

Performance Estimates

Throughput by GPU

RTX 4060
906.0 tok/s
RTX 3070
1.5K tok/s
B200 SXM
3.5K tok/s

VRAM Breakdown (RTX 4060, BF16)

Weights
KV
Act
Weights 1.5 GBKV-Cache 1.6 GBActivations 1.6 GBOverhead 0.1 GB

Precision Impact

bf16

1.5 GB

weights/GPU

~906.0 tok/s

fp8

0.8 GB

weights/GPU

int4

0.4 GB

weights/GPU

Capabilities

Features

Tool Use Vision Code Math Reasoning Multilingual Structured Output

Supported Frameworks

vllm

Supported Precisions

BF16 (default)FP8INT4

Where to Deploy Florence 2 Large

Similar Models

Larger contextCompare →
Larger contextCompare →

BGE M3

0.568B params · dense

Quality: 50

from $0.01/M

Larger contextCompare →

Frequently Asked Questions

How much VRAM does Florence 2 Large need for inference?

Florence 2 Large requires approximately 1.5 GB of VRAM at BF16 precision, 0.8 GB at FP8, or 0.4 GB at INT4 quantization. Additional VRAM is needed for KV-cache (98304 bytes per token) and activations (~0.20 GB).

What is the best GPU for Florence 2 Large?

The top recommended GPU for Florence 2 Large is the RTX 4060 using BF16 precision. It achieves approximately 906.0 tokens/sec at an estimated cost of $209/month ($0.09/M tokens). Score: 90/100.

How much does Florence 2 Large inference cost?

Florence 2 Large inference costs vary by provider and GPU setup. Use our calculator for detailed cost estimates across all providers.