Faysal Aziz

faysalabdulaziz

AI & ML interests

I enjoy experimenting with emerging technologies, building real-world prototypes, and turning complex ideas into reliable working systems.

Recent Activity

new activity 20 days ago

unsloth/GLM-5.3-Flash-GGUF:UD-Q3_K_XL split across two DGX Sparks (llama.cpp RPC over RDMA): memory, speed, 10 users, and what tripped me up

new activity 23 days ago

Qwen/Qwen-Image-2.1-Turbo:Qwen-Image-2.1-Turbo BF16 on one DGX Spark (stable-diffusion.cpp): timings and before/after

new activity 23 days ago

nvidia/Qwen3.8-27B-NVFP4:Ran this on a single DGX Spark - numbers, 10 users at once, and it couldn't find Buckingham Palace

View all activity

Organizations

New activity in unsloth/GLM-5.3-Flash-GGUF 20 days ago

UD-Q3_K_XL split across two DGX Sparks (llama.cpp RPC over RDMA): memory, speed, 10 users, and what tripped me up

#28 opened 25 days ago by

faysalabdulaziz

New activity in Qwen/Qwen-Image-2.1-Turbo 23 days ago

Qwen-Image-2.1-Turbo BF16 on one DGX Spark (stable-diffusion.cpp): timings and before/after

#15 opened 23 days ago by

faysalabdulaziz

New activity in nvidia/Qwen3.8-27B-NVFP4 23 days ago

Ran this on a single DGX Spark - numbers, 10 users at once, and it couldn't find Buckingham Palace

👍 1

#7 opened 23 days ago by

faysalabdulaziz

liked a model 23 days ago

nvidia/Qwen3.8-27B-NVFP4

Text Generation • 18B • Updated Sep 17 • 260k • 235

liked a model 25 days ago

nvidia/Nemotron-3.5-Embed-1B-BF16

Visual Document Retrieval • 2B • Updated 19 days ago • 21k • 61

upvoted 2 articles 25 days ago

Article

Give Your Coding Agents a Memory You Own

dacorvo

•

Sep 3

• 172

Article

Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

iamleonie, burtenshaw, sergiopaniego

•

Sep 3

• 159

liked a Space 25 days ago

FLUX 3 Image Klein

🌲

280

Open weights for image editing, generation and layouts

upvoted an article 25 days ago

Article

Welcome FLUX.3 Image Klein in Diffusers

ariG23498, sayakpaul, linoyts

•

26 days ago

• 51

liked 3 models 25 days ago

IQ2-mix + DFlash2 drafter + vision on one DGX Spark (llama.cpp b11515): tool-calling tests in all four thinking modes, 10 users, DFlash vs plain

👍 1

#7 opened 25 days ago by

faysalabdulaziz

New activity in ggml-org/Clef-Flash-GGUF 25 days ago

BF16 on a DGX Spark (llama.cpp b11576, /v1/systemone): message labelling in English, Malay and Chinese, speed, memory, next to two other models

#2 opened 25 days ago by

faysalabdulaziz

New activity in unsloth/Qwen-Image-Edit-2511-GGUF 25 days ago

Q6_K on a DGX Spark with stable-diffusion.cpp sd-server: create and edit timings, Lightning 4-step vs 8-step vs 40-step, memory, with images

#14 opened 25 days ago by

faysalabdulaziz

New activity in sakamakismile/Huihui-Qwen3.8-27B-abliterated-NVFP4 25 days ago

Running it with vLLM on one DGX Spark: settings, memory, MTP, all four thinking modes, and a photo test

#2 opened 25 days ago by

faysalabdulaziz

New activity in unsloth/GLM-5.3-Flash-GGUF 25 days ago

UD-Q3_K_XL split across two DGX Sparks (llama.cpp RPC over RDMA): memory, speed, 10 users, and what tripped me up

#28 opened 25 days ago by

faysalabdulaziz

Faysal Aziz

AI & ML interests

Recent Activity

Organizations

faysalabdulaziz's activity

UD-Q3_K_XL split across two DGX Sparks (llama.cpp RPC over RDMA): memory, speed, 10 users, and what tripped me up

Qwen-Image-2.1-Turbo BF16 on one DGX Spark (stable-diffusion.cpp): timings and before/after

Ran this on a single DGX Spark - numbers, 10 users at once, and it couldn't find Buckingham Palace

Give Your Coding Agents a Memory You Own

Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

FLUX 3 Image Klein

Welcome FLUX.3 Image Klein in Diffusers

IQ2-mix + DFlash2 drafter + vision on one DGX Spark (llama.cpp b11515): tool-calling tests in all four thinking modes, 10 users, DFlash vs plain

BF16 on a DGX Spark (llama.cpp b11576, /v1/systemone): message labelling in English, Malay and Chinese, speed, memory, next to two other models

Q6_K on a DGX Spark with stable-diffusion.cpp sd-server: create and edit timings, Lightning 4-step vs 8-step vs 40-step, memory, with images

Running it with vLLM on one DGX Spark: settings, memory, MTP, all four thinking modes, and a photo test

UD-Q3_K_XL split across two DGX Sparks (llama.cpp RPC over RDMA): memory, speed, 10 users, and what tripped me up