Model reference · open weights

Kimi-K3-DSpark

Kimi-K3-DSpark is an open-weight language model from RadixArk, listed in the AxForge catalogue. AxForge can bring it up on EU-owned hardware for you on request — with the licence handled where one is required.

Licence fee required LLMs RadixArk 1 variants 3.3M downloads/mo
Request a licence + hosting quote All served models Not on the shared API today — deployed on request.

About

What Kimi-K3-DSpark is

Kimi K3 DSpark speculator Overview A long-context DSpark speculator for Kimi K3. It supports context lengths of up to 1 million tokens. A DSpark speculator for the Kimi K3 target, enabling faster inference through speculative decoding. DSpark extends the DFlash parallel-draft backbone with a Markov logit-bias head and a per-position confidence head. This checkpoint was trained with SpecForge using hidden states from a live SGLang target engine. Model Specifications - Base model: moonshotai/Kimi-K3 - Format: Safetensors (single-file BF16, 2,249,289,601 parameters) - Draft: 5 full-attention Qwen3-style GQA layers, hidden size 7168, 64 query heads / 16 KV heads, and blocksize=7 - Verification width: 1 current token + 7 draft tokens - Auxiliary target layers: [7, 23, 51, 67, 83] - Trained context: 65,536 tokens - Target weights: embedding and unembedding weights are not included Evaluation Results acclen is SGLang's histogram-native request acceptance length, averaged within each question and then equally across questions. RULER V2 uses the 1M input configuration. Actual prompts span 1,000,432–1,047,925 tokens; partition acclen is 4.4658 for MK, 4.3081 for MV, and 3.9919 for QA. AIME26 acclen by output length Serving with SGLang SGLang Cookbook provides Kimi K3 deployment recipes. YaRN-16 is enabled in the published draft config by default with originalmaxpositionembeddings=65536 and maxpositionembeddings=1048576; no separate draft config override is required. Training Details - Framework: SpecForge online distillation, with hidden states captured from a frozen Kimi K3 target served by a live SGLang engine. Draft trained from random initialization. - Loss: 0.1 CE + 0.9 L1 distillation + 1.0 confidence BCE, decay gamma 4.0, with 512 sampled anchors per sequence and blocksize=7. - Topology: 4 nodes × 4 GB300 (16 ranks) — 2 × TP8 target replicas, DP2 sampler, FSDP16 SHARDGRADOP on the draft, TP-batch scatter. Batch 8 per replica × 32 accumulation steps × 2 replicas = global batch 512.

Summarised from the published model card. Read the full card on the HuggingFace links below.

Specifications

What it is

MakerRadixArk
TypeLanguage models
Parameters (lead)2.2B
Context1024k tokens
Variants1
Runs withtransformers
Released2026-07-27
Popularity3.3M downloads / month
Likes54
LicenceCommercial licence needed

How it works

How language models work

Your prompttext / messagesTransformerattention over tokensNext-token loopgenerate + streamResponsetext · tool callsA language model reads your tokens and predicts the next one, again and again, streaming the reply back.

Variants

Sizes & precisions

Open weights ship in several sizes and precisions. One page, all the variants — pick the one that fits your GPU. VRAM figures are estimates from model size.

VariantParamsPrecisionVRAMFits 16 GBWeights
Kimi-K3-DSpark2.2BBF16~5.2 GBWeights ↗

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys kimi-k3-dspark for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (kimi-k3-dspark below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3-dspark","messages":[{"role":"user","content":"Hello"}]}'

Details

Languages, data & research

Tags

transformers safetensors qwen3 feature-extraction speculative-decoding dspark dflash specforge sglang long-context text-generation custom_code text-generation-inference

Licence

Commercial licence needed

The weights are open but its licence needs a commercial agreement for business use. AxForge can arrange that licence and host the model for you — you pay AxForge, we settle with the model’s maker. Ask us for a quote. Read the licence ↗

Sources

Weights & code

Want Kimi-K3-DSpark on EU-owned hardware?

Request a licence + hosting quote See what’s served now

Explore

More language models

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