Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

RadixArk · qwen3

Kimi-K3-DSpark

Kimi-K3-DSpark by RadixArk: a 2.2B-parameter dense open-weight model under the unknown license.

unknownDense1049K contextAuto-imported · not yet reviewed

3,550,124 downloads · 56 likes · Model card · synced 2026-09-19

Specifications

Parameters
2.2B
Architecture
dense · qwen3
Layers
5
Hidden size
7,168
KV heads · head dim
16 · 64
Vocabulary
163,840
Native dtype
Context window
1,048,576 tokens
Released
2026-07-27
Last modified on Hub
2026-08-16

Features & licensing

License
unknown
Commercial use
No
Modalities
text
Tool / function calling
No
Reasoning mode
No
Languages
EN
Quantised variants on Hub
FP8, GGUF
Pipeline
text-generation
transformerssafetensorsqwen3feature-extractionspeculative-decodingdsparkdflashspecforgesglanglong-contexttext-generationcustom_codetext-generation-inference

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run Kimi-K3-DSpark

Weights need about 5 GB at FP16, 3 GB at INT8 and 1 GB at INT4. The KV cache adds roughly 20 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 1.8 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
1 × NVIDIA L4 (24 GB)24 GB8 %~0.5 s$584
1 × NVIDIA L40S (48 GB)48 GB4 %~0.2 s$1,387
1 × NVIDIA A100 (80 GB)recommended80 GB2 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB2 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.