Model reference · open weights

Inkling-Small-DSpark

LLMs RadixArk Text gen 1 build Licence not stated 35k dl/mo

Inkling-Small-DSpark is an open-weight language model from RadixArk. Inkling-Small-DSpark-Preview (BF16) weighs 1.8 GB; the smallest configuration that runs it is RTX 3060 12 GB.

For self-hosting, compare this build's requirements below with the GPU you plan to rent.

  • Inkling-Small-DSpark is a speculative decoding draft model developed by RadixArk to accelerate inference for the Inkling-Small NVFP4 target.
  • It features 1.4B parameters and supports a context length of 1,048,576 tokens.
  • The model is released as a research preview and requires the specific target checkpoint and tokenizer for operation.

Summary of the RadixArk/Inkling-Small-DSpark-Preview model card, 2026-10-01

What it is

Released byRadixArk
Released2026-07-30
Parameters1.4B
VRAM1.8 GB for the weights

What it runs on

Memory and cards for Inkling-Small-DSpark-Preview (BF16)

1.8 GBweights, file size
20 MBcache per 1K tokens
770 MBruntime overhead, at least
1,048,576 tokenscontext max
CardRequests at onceContext maxMemory
8K each32K each
RTX 3060 12 GB5313431K11.6 GB
RTX 4060 Ti 16 GB7619612K15.4 GB
RTX 3090 24 GB12431993K23.4 GB
RTX 4090 24 GB12330991K23.4 GB
RTX 5090 32 GB16942all 1024K31.0 GB
L40S 48 GB24661all 1024K44.0 GB
A100 80 GB450112all 1024K78.2 GB
H100 80 GB421105all 1024K78.1 GB
RTX PRO 6000 Blackwell 96 GB515128all 1024K93.8 GB
DGX Spark (GB10) 128 GB unified595148all 1024K107 GB
H200 141 GB778194all 1024K138 GB
B200 180 GB1000+250all 1024K176 GB
Memory needed at each load
Requests at once8K tokens each32K tokens each
12.7 GB3.2 GB
53.4 GB5.9 GB
83.9 GB7.9 GB
165.3 GB13.3 GB
327.9 GB24.0 GB
6413.3 GB45.5 GB

One card, with vLLM's small-card settings.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms