Yes 2 machines a DGX Spark and a Linux workstation and they connected but a 100Gbps dedicated RoCE network
Csaba Kecskemeti PRO
csabakecskemeti
AI & ML interests
None yet
Recent Activity
liked
a model
2 days ago
Ex0bit/GLM-4.7-PRISM
replied to
their
post
4 days ago
Just sharing a result of a homelab infrastructure experiment:
I've managed to setup a distributed inference infra at home using a DGX Spark (128GB unified gddr6) and a linux workstation with an RTX 6000 Pro (96GB gddr7) connected via 100Gbps RoCEv2. The model I've used (https://lnkd.in/gx6J7YuB) is about 140GB so could not fit either of the GPU. Full setup and tutorial soon on devquasar.com
Screen recording:
https://lnkd.in/gKM9H5GJ
updated
a model
4 days ago
DevQuasar/XiaomiMiMo.MiMo-V2-Flash-GGUF
Organizations
replied to
their
post
4 days ago
posted
an
update
5 days ago
Post
3082
Just sharing a result of a homelab infrastructure experiment:
I've managed to setup a distributed inference infra at home using a DGX Spark (128GB unified gddr6) and a linux workstation with an RTX 6000 Pro (96GB gddr7) connected via 100Gbps RoCEv2. The model I've used (https://lnkd.in/gx6J7YuB) is about 140GB so could not fit either of the GPU. Full setup and tutorial soon on devquasar.com
Screen recording:
https://lnkd.in/gKM9H5GJ
I've managed to setup a distributed inference infra at home using a DGX Spark (128GB unified gddr6) and a linux workstation with an RTX 6000 Pro (96GB gddr7) connected via 100Gbps RoCEv2. The model I've used (https://lnkd.in/gx6J7YuB) is about 140GB so could not fit either of the GPU. Full setup and tutorial soon on devquasar.com
Screen recording:
https://lnkd.in/gKM9H5GJ
-
- 3 replies
posted
an
update
27 days ago
Post
1253
FYI: Mistral.Ministral-3 dequantizer FP8->BF16
https://github.com/csabakecskemeti/ministral-3_dequantizer_fp8-bf16
(The instruct model weights are in FP8)
https://github.com/csabakecskemeti/ministral-3_dequantizer_fp8-bf16
(The instruct model weights are in FP8)
replied to
their
post
28 days ago
I've used this:
https://huggingface.co/meituan/DeepSeek-R1-Channel-INT8/tree/main/inference
Hoped I can make it work on my CPU... :P
replied to
their
post
28 days ago
@ubergarm you might have the resources!? ๐
posted
an
update
28 days ago
Post
2070
Looking for some help to test an INT8 Deepseek 3.2:
SGLang supports Channel wise INT8 quants on CPUs with AMX instructions (Xeon 5 and above AFAIK)
https://lmsys.org/blog/2025-07-14-intel-xeon-optimization/
Currently uploading an INT8 version of Deepseek 3.2 Speciale:
DevQuasar/deepseek-ai.DeepSeek-V3.2-Speciale-Channel-INT8
I cannot test this I'm on AMD
"AssertionError: W8A8Int8LinearMethod on CPU requires that CPU has AMX support"
(I assumed it can fall back to some non optimized kernel but seems not)
If anyone with the required resources (Intel Xeon 5/6 + ~768-1TB ram) can help to test this that would be awesome.
If you have hints how to make this work on AMD Threadripper 7000 Pro series please guide me.
Thanks all!
SGLang supports Channel wise INT8 quants on CPUs with AMX instructions (Xeon 5 and above AFAIK)
https://lmsys.org/blog/2025-07-14-intel-xeon-optimization/
Currently uploading an INT8 version of Deepseek 3.2 Speciale:
DevQuasar/deepseek-ai.DeepSeek-V3.2-Speciale-Channel-INT8
I cannot test this I'm on AMD
"AssertionError: W8A8Int8LinearMethod on CPU requires that CPU has AMX support"
(I assumed it can fall back to some non optimized kernel but seems not)
If anyone with the required resources (Intel Xeon 5/6 + ~768-1TB ram) can help to test this that would be awesome.
If you have hints how to make this work on AMD Threadripper 7000 Pro series please guide me.
Thanks all!
-
- 8 replies
posted
an
update
about 2 months ago
Post
303
Recently there are so much activity on token efficient formats, I've also build a package (inspired by toon).
Deep-TOON
My goal was to token efficiently handle json structures with complex embeddings.
So this is what I've built on the weekend. Feel free try:
https://pypi.org/project/deep-toon/0.1.0/
Deep-TOON
My goal was to token efficiently handle json structures with complex embeddings.
So this is what I've built on the weekend. Feel free try:
https://pypi.org/project/deep-toon/0.1.0/
posted
an
update
7 months ago
Post
3070
Has anyone ever backed up a model to a sequential tape drive, or I'm the world first? :D
Just played around with my retro PC that has got a tape driveโdid it just because I can.
Just played around with my retro PC that has got a tape driveโdid it just because I can.
- 5 replies
posted
an
update
7 months ago
Post
475
Deepseek R1 0528 Q2 locally.
(I believe it has overthinking it a bit :) )
https://youtu.be/Iqu5s9aFaXA?si=QWZe293iTKf_3ELU
DevQuasar/deepseek-ai.DeepSeek-R1-0528-GGUF
(I believe it has overthinking it a bit :) )
https://youtu.be/Iqu5s9aFaXA?si=QWZe293iTKf_3ELU
DevQuasar/deepseek-ai.DeepSeek-R1-0528-GGUF
posted
an
update
9 months ago
Post
2119
Local Llama4 Maverick Q2
https://youtu.be/4F8g_LThli0?si=MGba2SUTHt6xYw3T
Quants uploading now
Big thanks to @ngxson !
https://youtu.be/4F8g_LThli0?si=MGba2SUTHt6xYw3T
Quants uploading now
Big thanks to @ngxson !
posted
an
update
9 months ago
Post
1750
Why the 'how many r's in strawberry' prompt "breaks" llama4? :D
Quants DevQuasar/meta-llama.Llama-4-Scout-17B-16E-Instruct-GGUF
Quants DevQuasar/meta-llama.Llama-4-Scout-17B-16E-Instruct-GGUF
- 3 replies
posted
an
update
9 months ago
Post
3424
I'm collecting llama-bench results for inference with a llama 3.1 8B q4 and q8 reference models on varoius GPUs. The results are average of 5 executions.
The system varies (different motherboard and CPU ... but that probably that has little effect on the inference performance).
https://devquasar.com/gpu-gguf-inference-comparison/
the exact models user are in the page
I'd welcome results from other GPUs is you have access do anything else you've need in the post. Hopefully this is useful information everyone.
The system varies (different motherboard and CPU ... but that probably that has little effect on the inference performance).
https://devquasar.com/gpu-gguf-inference-comparison/
the exact models user are in the page
I'd welcome results from other GPUs is you have access do anything else you've need in the post. Hopefully this is useful information everyone.
posted
an
update
9 months ago
Post
2412
Managed to get my hands on a 5090FE, it's beefy
| llama 8B Q8_0 | 7.95 GiB | 8.03 B | CUDA | 99 | pp512 | 12207.44 ยฑ 481.67 |
| llama 8B Q8_0 | 7.95 GiB | 8.03 B | CUDA | 99 | tg128 | 143.18 ยฑ 0.18 |
Comparison with others GPUs
http://devquasar.com/gpu-gguf-inference-comparison/
| llama 8B Q8_0 | 7.95 GiB | 8.03 B | CUDA | 99 | pp512 | 12207.44 ยฑ 481.67 |
| llama 8B Q8_0 | 7.95 GiB | 8.03 B | CUDA | 99 | tg128 | 143.18 ยฑ 0.18 |
Comparison with others GPUs
http://devquasar.com/gpu-gguf-inference-comparison/
replied to
their
post
10 months ago
Follow-up
With the smaller context length dataset the training has succeeded.
posted
an
update
10 months ago
Post
1851
GTC new model announcement now from Nvidia
nvidia/Llama-3_3-Nemotron-Super-49B-v1
GGUFs:
DevQuasar/nvidia.Llama-3_3-Nemotron-Super-49B-v1-GGUF
Enjoy!
nvidia/Llama-3_3-Nemotron-Super-49B-v1
GGUFs:
DevQuasar/nvidia.Llama-3_3-Nemotron-Super-49B-v1-GGUF
Enjoy!
reacted to
clem's
post with ๐
10 months ago
Post
4785
We just crossed 1,500,000 public models on Hugging Face (and 500k spaces, 330k datasets, 50k papers). One new repository is created every 15 seconds. Congratulations all!
- 3 replies
posted
an
update
10 months ago
Post
597
Cohere Command-a Q2 quant
DevQuasar/CohereForAI.c4ai-command-a-03-2025-GGUF
6.7t/s on a 3gpu setup (4080 + 2x3090)
(q3, q4 currently uploading)
DevQuasar/CohereForAI.c4ai-command-a-03-2025-GGUF
6.7t/s on a 3gpu setup (4080 + 2x3090)
(q3, q4 currently uploading)
replied to
their
post
10 months ago
No success so far, the training data contains some larger contexts and it fails just before complete the first epoch.
(dataset: DevQuasar/brainstorm-v3.1_vicnua_1k)
If anyone has further suggestion to the bnb config (with ROCm on MI100)?
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_use_double_quant=True,
bnb_4bit_compute_dtype=torch.bfloat16
)
Now testing with my other dataset that is smaller seems I have a lower memory need
DevQuasar/brainstorm_vicuna_1k
replied to
their
post
10 months ago
It's failed by the morning, need to find more room to decrease the memory