4× DGX Spark
35.7tok/sEveryday
1 request, prose prompt, with DFlash2, thinking on (per recipe) Source
Intelligence45with thinking
- Custom kernel required
- Non-commercial
Local AI · Recipe · 3× DGX Spark
GLM-5.3 753B EXL3 2.75 bpw with TensorFold on 3× DGX Spark
GLM-5.3 753B EXL3 2.75 bpw with TensorFold on 3× DGX Spark: 31.3 tok/s according to github.com (dataset as of Sep 29, 2026).
by MiaAI-Lab
Engine
- Engine
- TensorFold v0.6.0 + 110 Patches, Image ghcr.io/miaai-lab/glm-5.3-exl3-3x-dgx-sparks-tensorfold:v0.6.0-4f3223245926, DSpark + Copy-Drafts, Context Parallelism, RoCE
- Quantization
- EXL3 2,75 bpw Ø (Experten 2–4 bit), 4-bit dense, BF16 sonst, FP4 KV (Default CP=1)
- Model family
- GLM-5.3
- Context
- 499,712
- Parameters
- ~753B (HF safetensors der Basis)
- Creator
- MiaAI-Lab
- GitHub stars
- 21
- Repo updated
- Oct 4, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Decode · parallel requests
Sums over all streams are marked; per request it is correspondingly less.
Prefill by context
What you need
- Hardware
- 3 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsMia-AiLab/GLM-5.3-EXL3-2.75bpw-TensorFold other
- Draft modelRedHatAI/GLM-5.3-speculator.dspark other
- Basezai-org/GLM-5.3 other
- Engine
- TensorFold v0.6.0 + 110 Patches, Image ghcr.io/miaai-lab/glm-5.3-exl3-3x-dgx-sparks-tensorfold:v0.6.0-4f3223245926, DSpark + Copy-Drafts, Context Parallelism, RoCE
- Context
- 499,712 tokens
Notes
What matters before you rebuild it.
- Custom kernel required
- tool-calling
- License unclear
Sources
- github.com/MiaAI-Lab/GLM-5.3-EXL3-3x-DGX-Sparks-TensorFold#performance
- github.com/MiaAI-Lab/GLM-5.3-EXL3-3x-DGX-Sparks-TensorFold#concurrent-requests-0134-0138-0139
- github.com/MiaAI-Lab/GLM-5.3-EXL3-3x-DGX-Sparks-TensorFold/blob/main/README.md
- github.com/MiaAI-Lab/GLM-5.3-EXL3-3x-DGX-Sparks-TensorFold/blob/main/scripts/config.sh
Related recipes
1× Mac
6.14tok/sMixed
1 request, mixed prompt set, context 1k, no speculative decoding Source
Intelligence45with thinkingOriginal model's value. Heavily compressed (~2.1 bit), intelligence is likely much lower.
- ≤ 3 bit: quantization may cost quality
- Streams from SSD
- License unclear
4× DGX Spark
53.3tok/sPeak
1 request, structured output end-to-end (count-to-300); DCP4 lane 2: NVFP4 KV Source
Intelligence45with thinking
- Custom kernel required
- License unclear
3× DGX Spark
51tok/sEveryday
1 request, prose prompt, with MTP/EAGLE (spekulativ, ~59 ms/step) (per recipe) Source
Intelligence40with thinkingno thinking 25
- Custom kernel required
3× DGX Spark
41.9tok/sEveryday
1 request, prose prompt, with DSpark (per recipe) Source
Intelligenceno independent value
- Custom kernel required
3× DGX Spark
39.6tok/sEveryday
1 request, prose prompt, with DFlash2 (per recipe) Source
Intelligence42with thinking
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
- Non-commercial
- License unclear
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.