Tiny Aya Global

Cohere · cohere/tiny-aya-global

ProviderCohere
Familyaya
Typellm-chat
Statusactive
Released2026-02-13
Updated2026-07-07
Parameters3.3B
Open weightsyes

Lineage

RelationshipModel
Is a finetune ofCohereLabs/tiny-aya-base

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sbf166.7 GB~2194535.9~8778143.6 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf166.7 GB~1672.0~6688.1 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf166.7 GB~836.0~3344.1 q4
NVIDIA B200180.0 GB7700.0 GB/sbf166.7 GB~804.7~3218.7 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf166.7 GB~771.2~3084.9 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf166.7 GB~627.0~2508.0 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf166.7 GB~553.9~2215.4 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf166.7 GB~501.6~2006.4 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf166.7 GB~407.6~1630.2 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf166.7 GB~350.1~1400.3 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf166.7 GB~334.4~1337.6 q4
Google TPU v5p95.0 GB2765.0 GB/sbf166.7 GB~288.9~1155.8 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf166.7 GB~213.1~852.3 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf166.7 GB~209.0~836.0 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf166.7 GB~187.3~749.1 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf166.7 GB~171.2~684.7 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf166.7 GB~167.2~668.8 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf166.7 GB~162.5~650.0 q4
Google TPU v432.0 GB1200.0 GB/sbf166.7 GB~125.4~501.6 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf166.7 GB~105.3~421.4 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf166.7 GB~100.3~401.3 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sbf166.7 GB~100.3~401.3 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf166.7 GB~97.8~391.3 q4
NVIDIA L40S48.0 GB864.0 GB/sbf166.7 GB~90.3~361.2 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf166.7 GB~85.6~342.3 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sbf166.7 GB~83.6~334.4 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf166.7 GB~83.6~334.4 q4
Google TPU v5e16.0 GB800.0 GB/sbf166.7 GB~83.6~334.4 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sbf166.7 GB~76.9~307.7 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sbf166.7 GB~70.2~280.9 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sbf166.7 GB~66.9~267.5 q4
Apple M4 Max128.0 GB546.0 GB/sbf166.7 GB~57.1~228.2 q4
Apple M1 Max64.0 GB400.0 GB/sbf166.7 GB~41.8~167.2 q4
Apple M2 Max96.0 GB400.0 GB/sbf166.7 GB~41.8~167.2 q4
Apple M3 Max128.0 GB400.0 GB/sbf166.7 GB~41.8~167.2 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sbf166.7 GB~37.6~150.5 q4
NVIDIA L424.0 GB300.0 GB/sbf166.7 GB~31.4~125.4 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sbf166.7 GB~30.1~120.4 q4
Apple M4 Pro64.0 GB273.0 GB/sbf166.7 GB~28.5~114.1 q4
Apple M432.0 GB120.0 GB/sbf166.7 GB~12.5~50.2 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging FaceCohereLabs/tiny-aya-globalgated

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.
BenchmarkEvaluated asScoreEvidence dateSource kind
aa_lcrTiny Aya Global0.0%2026-09-10 evaluatedindependent evaluatorsource
critptTiny Aya Global0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondTiny Aya Global30.51%2026-09-10 evaluatedindependent evaluatorsource

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub